Clean Electricity from Photovoltaics

kT

exp

kT

-1

(3.2)

where A** is the Richardson constant and V is the forward bias over the metalsemiconductor junction. In addition, the barrier height is reduced by image effects that are a function of the doping level. Even when assuming 100% light transmission through the metal, the predicted efficiency of solar cells based on metalsemiconductor junctions will be below 10% (see e.g. Hovel, 1975). A relatively new approach to overcome this limitation consists in using the absorption in the thin metal layer to excite 'hot carriers' over the barrier. Schmidt (1994) predicted that this could enhance the efficiency by 10%, but no experimental confirmation of this approach has yet been reported. From the technological point of view, it is not straightforward to realise metal-semiconductor junctions with reproducible barrier heights, because these are strongly influenced by interfacial contamination and crystal orientation. However, the interpolation of an interfacial oxide between the metal and semiconductor can be used advantageously to increase the barrier height and thereby Thin insulator = tunneling layer Metal /

Semiconductor

nitride induced n -type inversion layer (by nitride layer with large fixed positive charge) p -Si substrate

backside contact (a)

(b)

Figure 3.5 (a) Band-gap diagram for a cell with an MIS contact under equilibrium conditions; (b) Schematic cross-section of an MIS-IL Si cell.

Principles of Cell Design

99

decrease the dark current. When this approach is used, the thickness (2-4 nm) of the oxide is chosen so that the minority carriers can tunnel through the oxide (see the band-gap scheme of Fig. 3.5a). Such a device is called a metal-insulatorsemiconductor (MIS) solar cell. Much effort has gone into the development of Si MIS-technology. The most popular design is the MIS-inversion layer (IL) cell (see Fig. 3.5b). This cell configuration is realised by depositing a nitride layer, containing a large fixed positive charge, on a p-type Si-substrate. This induces an rc-type inversion-type emitter, which is contacted through an MIS contact. An alternative is the MIS-contacted n+-p Si solar cell. The ongoing research to optimise the latter MIS-approach has resulted in cell efficiencies above 20% (see Metz, 1997) and efficiencies as high as 23% are predicted (Kuhlmann, 1997) for a MIS-inversion layer cell.

3.3 Optical design of cells This section on optical cell design is divided into two parts. Section 3.3.1 deals with the design of the front and back surfaces to ensure that the light is efficiently coupled into the cell body. Different light-trapping schemes and their practical implementation are reviewed. Section 3.3.2 deals with the properties of anti-reflective coatings. The optical optimisation of cell surfaces divides into two questions: how can light be coupled efficiently into the cell and how can it be kept in the cell once it has entered the cell to ensure maximal absorptance? The first question basically addresses the reduction of the primary reflectance of the front surface whereas the second is related to the development of light-trapping schemes. The second item will be dealt with first because the light-trapping schemes discussed also affect the front-surface reflectance and are of primary importance for the cell's open-circuit voltage because they allow reduction of the necessary thickness of the cell, and thereby the volume for bulk recombination (see eq. 3.1a). Module design also influences optical design. Low-concentration approaches (e.g. prismatic covers) and bifacial cells are additional means to improve optical absorptance, but these will be omitted from the subsequent discussion.

3.3.1 Light trapping Light trapping has been extensively studied in the context of Si solar cells. The indirect band gap of Si results in an absorption coefficient that is lower than 10 cm"1

100

J. Poortmans, J. Nijs and R. Mertens

in the wavelength region where the terrestrial solar spectrum contains most energy (450-750 nm). The reader should therefore not be surprised that the bulk of theoretical and experimental work on light trapping has been done for Si, although improving the absorption of weakly absorbed light is in principle interesting for all types of cells. Two basically different types of light-trapping schemes can be distinguished: schemes based on lambertian surfaces, and those based on geometrical light-trapping schemes. A lambertian surface is one that scatters light uniformly in all [forward] directions, and lambertian light-trapping schemes are based on full internal randomisation of the light ray direction by a lambertian surface. Two possible implementations are shown in Figs. 3.6a and b. The first is based on a perfect lambertian rear reflector and a front surface with zero reflectance for the incoming light (a reflectance as low as 5% is readily feasible). The second is equivalent as far as the final result is concerned, but relies on a randomising front surface and a perfect rear reflector.

Lambertian frontside

Critical angle

KV&V^^V^^^^^ Lambertian backside

(a)

(b)

Figure 3.6 (a) Schematic of a cell with a perfect lambertian rear reflector and front surface with zero reflectance; (b) schematic of a cell with a perfect specular rear reflector and a lambertian front surface.

A detailed analysis of randomising schemes has been performed by Jablonovitch and Cody (1982). The basic result coming out of their analysis is the intensity enhancement of 2n inside a sheet with refractive index n and a perfect white reflector at the backside. This can easily be understood from basic statistical mechanics arguments. Under equilibrium conditions, the in- and outgoing radiation fluxes have to be equal. Since the fully randomised light inside the cell can only escape over a limited solid angle, determined by the critical angle for transmission of light from an optically dense material to a material with lower refractive index, the intensity inside the cell has to increase. Only under these circumstances can the in- and outgoing

101

Principles of Cell Design

radiation fluxes remain equal. Jablonovitch and Cody showed that this results in an optical path length enhancement of An2 for weakly absorbed light. The spectral absorptance a(k) for a material with refractive index n, thickness W and absorption coefficient a(X) is given by

a(X) =

—

(3.3a)

a(A) + —r— An W The same result is found from a calculation of the average pathlength for weakly absorbed light inside the structure of Fig. 3.6a, by averaging over all possible directions and assuming a l/n2 probability for escape every time the ray strikes the front surface (see e.g. Green, 1995 and references therein). The assumption of total loss of all the light within the escape cone for cells with a lambertian backside has been critically reviewed in a recent publication (Abouelsaood, 1997), in which Fresnel coefficients were used to calculate the amount of light transmitted within the escape cone. The difference between the exact calculation and the approximation of total loss within the escape cone was <2% for Si. The absorptance of weakly absorbed light is further increased by limiting the entrance angle 9. It was proved by Campbell (1986) that the spectral absorptance under such circumstances is given by

"&)=

" ' I

(3-3b)

g(A) + s m , • " "

where 0max is the maximal entrance angle. The effect of such lambertian-based optical enhancement schemes on limiting efficiency and obtainable currents has been studied in numerous papers (see e.g. Tiedje et al., 1982 for calculation of the limiting efficiency in crystalline Si cells, and Sheng, 1984 for current enhancement in a-Si:H cells). Up to now, we have assumed that the backside of the cell in Fig. 3.6a reflects the light perfectly. If this is not the case, the expression for the path-length enhancement A must be modified to 2

(1+Aback)

A=

^ - '

n

(3.3c)

102

J. Poortmans, J. Nijs and R. Mertens

where Rback 1S the backside reflectance. Although theoretically very attractive, the incorporation of perfectly randomising layers is not straightforward from the technical point of view. In the context of crystalline Si solar cells, recent work has identified porous Si as a promising approach to randomise the incoming light (Stalmans, 1998). However, the high surface recombination velocity at the porous Si-Si interface and the residual absorption of high-energy photons in the porous Si layer tend to reduce the benefits of this approach. Examples of cell structures with randomising rear reflectors are shown in Fig. 3.7, demonstrating the exploitation of the naturally occurring surface texture of microcrystalline material. A particularly successful implementation of this structure was the 9.5% efficient ultrathin 1.5 jxm polycrystalline Si-cell, developed by Yamamoto (1997). Another example is afforded by the use of textured contact layers, laid down by various means such as sputtering or etching in dilute HC1 (Kluth, 1997). Light ray

air(n=l)

Figure.?. 7 Practical realisation of a cell with lambertian rear reflector by use of the natural surface texture of microcrystalline Si layers. (After Winz, 1997, reproduced with permission from IEEE.)

Light-trapping structures with lambertian surfaces represent the limiting case for cells with an isotropic response, i.e. cells whose response is independent of the angle of incidence (Tiedje, 1984). Because of the directionality of sunlight, the use of surface structures with dimensions larger than the wavelength of light (geometrical light-trapping schemes) can increase the cell absorptance above the limits of the lambertian schemes. In contrast to the lambertian schemes, it is not possible to describe these schemes by a fully analytical approach; ray-tracing programs are required to perform a detailed analysis of these structures. Two basic types of geometrical light-trapping schemes can be distinguished: 2dimensional and 3-dimensional geometries. In their simplest form, two-dimensional structures can consist simply of a number of parallel grooves at the surface. The

103

Principles of Cell Design

number of passes can be increased by tilting the grooves (Campbell, 1988), as illustrated in Fig. 3.8. Recently a large amount of experimental work has been performed to realise solar cells in grooved multicrystalline Si-substrates. The interested reader is referred to the work of Zechner (1997) and references therein for details.

Figure 3.8

Cross-section of cell with parallel tilted grooves.

A further reduction of the probability of light escape is provided by threedimensional geometries. Instead of 2-dimensional grooves, three-dimensional pyramids are formed. The optical performance of such structures is studied by analysis of the 'image regions', which show where the rays strike the front surface after being reflected by the backside (see Fig. 3.9a).

side view

top view

incident pyramid

image pyramid

(a)

(b)

Figure 3.9 (a) Study of ray behaviour by means of image regions for regular pyramids (see e.g. Green, 1995, p.101); (b) schematic view of the tiler's pattern. This pattern is realised on (100) Si by slightly misaligning the pattern for the inverted pyramids with the <110> direction. (After Green, 1995, p.103, reproduced with permission from The Centre for Photovoltaic Devices and Systems.)

104

J. Poortmans, J. Nijs and R. Mertens

For regular pyramids, the light-trapping efficiency strongly depends on the details of the layout, especially the ratio of the pyramid size to the wafer thickness. A further improvement can be realised by going to the so-called tiler's pattern (see Fig. 3.9b).

w\

^

J ^ ^

(a)

\ \

J /

surface

i / i / **JH

Back surface

(b)

Figure 3.10 (a) Artist's view of a cell with light-trapping structures at both front- and backside. The grooves at the frontside are perpendicular to the grooves at the backside (after Green, 1995, p.104, reproduced with permission from The Centre for Photovoltaic Devices and Systems); (b) Structure, proposed by Jorgensen (1997), consisting of inverted pyramids at the frontside and octagonal structures at the backside. (After Jorgensen, 1997, reproduced with permission from IEEE.)

Additional improvement in light-trapping behaviour is obtained by structuring the backside of the cell. A very efficient method is the use of backside grooves perpendicular to the frontside grooves, as shown in Fig. 3.10a. In Fig. 3.10b, a recently proposed concept with structuring of both sides is shown (Jorgensen et al., 1997). This structure gives rise to a further improvement of light-trapping properties. Figure 3.11 shows a comparison of the optical performance of the different schemes. The 3-dimensional structures discussed above have found widespread use in (100) monocrystalline Si solar cells. Although the double-sided structure of Fig. 3.10a with perpendicular grooves should perform better than the tiler's pattern of Fig. 3.9, the latter has been used in the highest efficiency Si solar cells to date. In the context of industrial production of monocrystalline Si cells, (100) Si-substrates are textured with (lll)-facetted pyramids by anisotropic etching in NaOH- or KOH-containing solutions. This produces a random pyramidal texture as shown in the SEM picture in Fig. 3.12. For multicrystalline Si solar cells, the variation in grain orientation impedes efficient anisotropic texturing of the cell surface: grains with near (lll)-orientation are etched parallel to the surface, whereas the grains near to the (lOO)-orientation exhibit the desired pyramids with (lll)-facets. Therefore grooving (Zechner, 1997), and the use of randomising (Vazsonyi, 1995) or isotropic texturing schemes (Einhaus, 1997) are under investigation for multicrystalline Si cells.

Principles of Cell Design

105

i—'

'—r

randomizing surface

,

E p

80

periodic brickwork

random j pyra^nidst—*.,

5

\

10

T ~ \ j

L,

20

SO

Passes through cell

100 2 0 0

I 60 regular pyramids

6

10

20

60

100 200

Passes through cell

(a)

(b) Figure 3.11 (a) Percentage of light remaining in the cell as a function of the number of passes for cells with structure at the frontside (after Green, 1995, p. 105); (b) Percentage of light remaining in the cell as a function of the number of passes for cells with structure at both the front- and backside (after Green, 1995, p.105, reproduced with permissionfromThe Centre for Photovoltaic Devices and Systems).

j?i+*fr

Figure 3.12 etching.

SEM picture of a randomly textured (100) Si surface, created by NaOH-based anisotropic

106

J. Poortmans, J. Nijs and R. Mertens

Recent studies aimed at calculating the performance of thin-film cells made in conformal layers on glass by Brendel et al. (1997), Brendel (1997) and Thorp (1996) in the so-called pyramidal-film structure shown in Fig. 3.13 confirm the good performance of double-sided structures. One may conclude that high short-circuit currents are achievable, even with a sub-micron layer of Si, despite its indirect band gap and relatively low absorption coefficient in the energy-richest part of the spectrum.

Figure 3.13 Conformal thin-film crystalline Si solar cells on textured glass (pyramidal-film structure). (After Brendel et al., 1997, reproduced with permission from IEEE.)

3.3.2 Anti-reflective coatings: reduction of the first reflection Up to now, the reflection of a ray impinging on the cell front surface has been omitted from the discussion. The light-trapping schemes discussed in Section 3.3.1 inherently give rise to a substantial reduction of this reflectance, because any deviation from a perfectly smooth surface increases the probability that the reflected light ray will strike the surface again. A further reduction of the front-surface reflectance is achieved by the use of socalled anti-reflective coatings (ARC). The ARC layer, which is interposed between the photovoltaic cell material and the surrounding environment (air or the lamination material), acts as a quarter-wavelength impedance matching element between the characteristic impedance of the environment and the photovoltaic material. Under these conditions, the ray, reflected at the interface between the semiconductor (refractive index n2) and the ARC layer, interferes destructively with the ray reflected

107

Principles of Cell Design

at the interface between the ARC and surrounding medium (refractive index n0). The conditions for zero reflectance at wavelength A. are given by eqs. 3.4 and 3.5, where d\ and «i are the thickness and refractive index of the intermediate anti-reflective layer respectively. X dx=

(3.4) 4«, (3-5)

"l = yfnJh

Equations 3.4 and 3.5 define the conditions that reduce the reflectance to zero at a certain wavelength for a planar surface. The solar spectrum, however, extends over a broad wavelength range. The ARC-thickness is therefore chosen to give minimal integrated reflectance weighted by the energy content at each wavelength. On a structured surface, as used in geometrical light-trapping schemes, the minimum reflectance is no longer zero, because the angle at which the ray travels in the ARC layer is no longer unique. However, the better light-trapping properties for the light coupled in, and the broader minimum in the spectral reflectance curve in case of a geometrical light-trapping structure, largely compensate for this non-zero minimum reflectance. An example, given in Fig. 3.14, shows the reflectance of an InP cell coated with an ITO or ITO + MgF2 layer (see e.g. Coutts and Yamaguchi, 1988). n

c

n

i

•

'

•

i

•

•

• '

i

•

•

•

•

i

•

'

•

•

i

•

•

•

•

i

•

• ' •

InP thickness = 0.025 cm ...; ITO thickness - 52.5 nm "-. MgF2 thickness = 100.0 nm

0.40

r g 020

'

InP

"•• InP/ITO...

•

0.10 300 400 nnn

__.

•;:<'.

InP/ITO/MgF;.

500 600 700 800 900 1000 Wavelength (nm)

Figure 3.14 Spectral reflectance curve for an InP cell with an Indium-Tin Oxide (single-layer ARC) and Indium-Tin Oxide + MgF2 layer (double-layer ARC). (After Coutts and Yamaguchi, 1988, reproduced with permission from Academic Press.)

Practical examples of single-layer ARCs for Si are SiN^H^,, Ti0 2 and Ta 2 0 5 . The wavelength range over which the reflectance is low can be increased by going to double-layer ARCs, consisting for instance of ZnS/MgF2 double layers for Si cells.

108

J. Poortmans, J. Nijs and R. Mertens

3.4 Surface recombination losses and their reduction The calculations in Section 3.2 of the theoretically achievable efficiencies of solar cells, fabricated in a certain material were generally based on bulk recombination via radiative or Auger processes. Provided that light-trapping structures to maintain the current at a high level are used, these calculations show that the cell volume should be kept low to minimise bulk recombination and achieve the highest possible open-circuit voltage. In real life, recombination at surfaces and in the bulk is mostly governed by Shockley-Read-Hall recombination. In this section, we will focus on the latter recombination mechanism and how to minimise its influence.

3.4.1 Surface recombination Surface recombination occurs at the extremities of the cell. Its effect is described in terms of the surface recombination velocity S. Conceptually, in a p-type semiconductor, S is the relation between the recombination current density
(3.6)

The relation between S and the properties of the uncharged surface in the presence of a continuous distribution of states Dit(U) and cross-sections cr,(L0 is given by S = vth f^ ° * < " ° + A » + 2A"> iU Juv no+An+n^iU) p0+An+px(U) + op(U) on(U)

(3.7)

where v^ is the thermal velocity of the charge carriers. When the capture crosssections and their energy dependences are known, it is, in principle, possible to calculate S (see e.g. Eades and Swanson, 1985, for Si passivated by its thermal oxide) from capacitance-voltage and deep-level transient spectroscopic measurements. Often the value of S is empirically determined from electrical (e.g. current-voltage measurements) or electro-optical methods (e.g. effective lifetime measurements). The thus-measured value is in most cases interpreted in terms of an effective surface

Principles of Cell Design

109

recombination velocity, taken at a reference plane positioned close to the physical surface. This is done to avoid the complication of the electronic band bending near the surface as a result of surface charges. This suppresses or enhances the actual excess minority carrier concentration, as compared with its values in the quasi-neutral region near the surface. Surface passivation often puts a strong constraint on cell efficiency. This can be illustrated by the following quantitative comparison of the contribution to the dark current of the (volume) Auger recombination term with that associated with the surfaces. For a cell in low-level injection, the dark current contribution from Auger recombination in a bulk region of doping Nb and width Wb (the Auger lifetime is then 1/C Nl) is given by

q-L-CNlWbeKP

qV_

kT

(3.8)

where C is the Auger coefficient (based on the assumption of constant quasi-Fermi potential separation qV over the bulk region). The surface recombination current is created by a surface with surface recombination velocity Stff is 'v=<7-r-S eff exp

(3.9)

kT

It follows that the condition to ensure that surface recombination is not the determining factor for the open-circuit voltage is (3.10)

S*<WbCNl

If one calculates how low the effective surface recombination velocity has to be for a typical silicon cell (200 ftm thick, 1017 cm-3 p-type doping), one arrives at a value of 20 cm s~'. This very low value is not easily realisable, even for a material like Si. However, for high-injection conditions and under concentration, the requirements on the surface passivation are relaxed. In the high-injection regime, the Auger-limited bulk contribution to the recombination is given by (Green, 1995, p. 185) ib=qWbni(Cn+C.,)exp

3qV_ -1 2kT

(3.11)

110

J. Poortmans, J. Nijs and R. Mertens

whereas the surface recombination term is now given by '(qv)

-1

[[ikTJ Equating these two expressions and using the fact that under open-circuit conditions the photogenerated current is exactly balanced by the total recombination current, one arrives at the following expression for Seff, which ensures that the bulk recombination component dominates the dark current: 2

5eff <

~3[wb(Cn+Cp)]~3

I

(3.13)

Using eq. 3.13, the reader can easily verify that under concentration the requirements are relaxed as compared to the low-injection case.

3.4.2 Practical realisation of surface passivation In the following section a distinction is made between two situations. First, surface recombination in the non-contacted areas and the possibility of reducing the surface recombination velocity in these regions by suitable passivation layers will be dealt with. Afterwards, methods to passivate the surfaces of contacted areas are reviewed. Surface passivation of non-contacted areas The ability to passivate a semiconductor surface is strongly dependent on the material. The surface of a material is a major disturbance of the crystal and is characterised by a high surface-state density (of the order 1012-1013 cm-2). A reduction of the surfacestate density is achieved by growing or depositing thin dielectric layers on the surface. The surfaces of most II-VI and III-V semiconductors and Ge are, however, difficult to passivate. Only the Si surface can be readily passivated by a variety of techniques. The oxidation of a Si surface results in a low and stable surface-state density of the order of 1010 cm"2. These values follow from capacitance-voltage measurements (see e.g. Aberle, 1992 for oxide-passivated Si) or from measured surface-state densities and capture cross-sections, as shown in Figs. 3.15a and 3.15b. On n-type material the

Principles of Cell Design

111

surface recombination velocity is two orders of magnitude lower than on p-type material.

10 1 4

10 1 6

Substrate doping, cm" 3 (a)

10 1 8

10 1 4

10 1 6

10 1 8

Substrate doping, c m 3 (b)

Figure 3.15 Calculated surface recombination velocity for (a) p-type Si and (b) for n-type Si. The calculation was based on measured capture cross sections and the measured energy dependence of the interface state density. (After Eades and Swanson, 1985, reprinted with permission from A1P.)

In the context of industrial Si solar cells, the use of nitride layers, grown by the plasma-enhanced chemical vapour deposition technique, has attracted much interest. These combine the desired optical properties (n = 2) with surface and bulk passivation features (see e.g. Chen, 1994). High efficiencies of >17% have been reported for large-area screen-printed Si solar cells (see e.g. Nijs, 1996; Shirasawa, 1994). A relatively new approach to passivation of the cell non-contacted backside is the use of so-called floating junctions. Figure 3.16 shows the basic structure. It consists of an additional non-contacted diffused region at the backside. In principle, this approach allows very low values of the surface recombination velocity. By the light-induced forward bias over the floating junction electrons are injected towards the n+-contact This can be deduced from the equation for the effective surface recombination velocity of a surface passivated with a floating junction (Ghannam, 1991):

112

J. Poortmans, J. Nijs and R. Mertens

•jrff

—

"

NJ qn)

(3.14)

where /op is the saturation current density of hole injection into the «+-region and Nb is the doping in the base Actual realisations of this concept at the backside of Si solar cells resulted in an open-circuit voltage of 717 mV (Zhao, 1994), the highest ever achieved for a Si cell under 1-Sun illumination Emitter fingers

fT\^

JSX

p-type base /?+-contact floating n+-region

Figure 3.16 backside.

Base contact

floating

Schematic cross section of a cell with /j-type floating junctions for passivation of the cell

Whether the concept is more generally applicable to materials with lower minority carrier lifetimes is still a matter of debate. The derivation of 5eff in eq. 3.14 for the floating junction only takes into account the recombination in the quasi-neutral part of the diffused region. Lolgen (1994) proved that for less optimal circumstances (low lifetimes, low injection level), the depletion layer recombination (described by i02) will co-determine the actual Seff, which also becomes injection-dependent. The effective surface recombination velocity is then given by (Lolgen, 1994)

exp 5, ff =

qV_ -1 + k kT qn{

expl^hl

qV_ exp kT

(3.15)

113

Principles of Cell Design

This expression also takes the shunt resistance /?shunt into account, although it neglects generation at the cell backside. To avoid the dominance of depletion-layer recombination, the minority carrier concentration at the backside of the cell should be high (this increases the self-induced forward bias over the floating junction). This suggests the use of the floating junction passivation for bifacial cells, where light can also enter from the backside. Also the shunt resistance between the p+-and n+-regions has to be kept high. Careful optimisation by Honsberg (1996) showed that the effect of a shunt resistance is minimised by going to a shallow diffusion with a high sheet resistance for the floating n+-region. The high sheet resistance will isolate the shunted region from the rest of the floating junction. Surface passivation of metal-contacted areas The region under an ohmic contact is characterised by a large surface recombination velocity which is often assumed to be infinitely large (for example, in Si the surface recombination velocity at an ohmic contact is the thermal velocity of the carriers, which is about 107 cm s"1). This strongly enhances the dark current injected into the base. This is clearly seen in the expressions of the recombination current in the case of a base with a well passivated surface (eq. 3.17) or when the surface recombination velocity is infinite (eq. 3.16) for the case of a short base (diffusion length larger than the base width).

'A (Shack = ° ° ) =

h(SM=0)-

—

NbWb

exp

qV_ -1 kT

( v_ an, •W„ exp q kT Nbtb

(3.16)

(3.17)

where Dk Nb, Wb and «, have their usual meaning and Tb is the base lifetime. As described above, the use of dielectric passivation layers may result in well-passivated surfaces in the non-contacted areas, but if no measures are taken to suppress the recombination velocity near the ohmic contacts, no significant reduction of the recombination current is to be expected. A simple calculation illustrates this. Assuming a metal contact coverage of 1% and a ratio of 105 between the surface recombination velocity between the metallised and non-metallised areas, the contribution of the recombination at the contacts to the total recombination is still a factor of 103 larger than the contribution of the non-contacted area.

114

J. Poortmans, J. Nijs and R. Mertens

A careful analysis of eq. 3.16 suggests how the contribution of the metallised areas to the recombination current can be reduced. The solution is to increase the doping level near the contact. The contact area is thereby screened from the minority carriers by a high-low doping transition, creating a back surface field (or front surface field when the same technique is applied at the frontside of the cell). In n+-p Si solar cells this technique is widely used for the passivation of the backside. From eq. 3.16 the reader might expect the doping level at the backside should be as high as possible to optimise the surface-field effect. There are, however, a number of additional considerations that should be taken into account when optimising the contact passivation.

'eft-

recombination in accumulation layer

Doping ratio small =weak BSF p-region

• / accumulation region

(a)

. V P '-region depletion region

(b)

Figure 3.17 (a) Schematic representation of the different regions to be accounted for when calculating the effective surface recombination velocity at a p-p+-doping transition; (b) qualitative dependence of the effective surface recombination velocity on the doping ratio between the heavily and lightly doped parts of the high-low transition.

At high doping levels, the beneficial effect of the surface field is reduced by heavy-doping-induced band-gap narrowing (see e.g. Jain and Roulston, 1991 and references therein). This effect will increase the effective intrinsic carrier concentration in the heavily doped part and thereby increase the recombination current (see e.g. Brendel etai, 1995 as how to take this effect into account). One also has to account for recombination in the heavily doped layer For the case of Si the reader is referred to the work of Del Alamo (1981) and references therein. A third complication is caused by recombination in the accumulation and depletion region associated with the high-low transition (see Fig. 3.17a). This gives rise to additional recombination terms and, more importantly, it leads to an optimum value for the

115

Principles of Cell Design

doping ratio between the heavily and lightly doped part (see e.g. Singh, 1991 for calculations of the p-p+ transition in Si n+-p-p+ solar cells). When the doping ratio is too high, the low electric field in the accumulation layer results in enhanced recombination. An additional reduction of the recombination at the contacts can be obtained by including a thin layer of a material with larger band gap near the contact. Equation 3.16 is in fact a specific case of the following more general expression for the recombination current in a short base with high backside surface recombination velocity:

exp kT

(3.18)

o Dhn] The integral in eq. 3.18 is called the base Gummel number. The value of the Gummel number can be increased by decreasing the value of n] in part of the short base by including a higher band-gap region in the heavily doped region near the contact. There are numerous examples of this approach where the band gap can be adjusted without running into problems of interface lattice mismatch. An example is the use of an AlGaAs layer at the backside of a cell with a GaAs base (see Andreev, 1997 and references therein). In the case of Si, the heterojunction transition from Si to heavily doped /ic-Si:H to passivate the backside contact regions resulted in an efficiency as high as 23.5% (Okamoto, 1997).

3.4.3 Front surface passivation: homojunction vs. heterojunction design Most p-n homoj unctions and heteroj unctions used for photovoltaic purposes are onesided, strongly asymmetric junctions in which the base layer of the cell has a lower doping level. Such a design is suggested by the strongly peaked character of the carrier generation when the cell is illuminated by sunlight (uniformly absorbed light would not require an asymmetric structure). To keep the photocurrent loss in the top region within reasonable limits with such a peaked generation profile, the top semiconductor layer has to remain thin. Since for most cell designs there is the additional requirement of reducing resistive losses caused by horizontal carrier flow in the top layer (see also Section 3.5.1), one will normally encounter high doping levels in the top layer and Auger recombination as the dominant recombination mechanism.

116

J. Poortmans, J. Nijs and R. Mertens

The lower doping level in the base, on the other hand, is a requirement to obtain reasonable diffusion lengths in the base of the cell and to ensure efficient collection of carriers generated in the bulk of the cell. The practicality of a homojunction approach is dependent on the degree of achievable surface passivation of the front surface. In addition, the ratio of junction depth to absorption depth must be sufficiently small. The latter condition is satisfied for Si, but in CdTe the optical absorption coefficient is so high that this condition is difficult to realise and the homojunction approach is ruled out, practically speaking. The success of the Si homojunction approach is largely based on the ability to passivate its surfaces efficiently, even when the doping level is high. On the other hand, because of the poor surface passivation of most III-V compounds (except InP) and II—VI compounds {e.g. CdTe), a homojunction approach is in most instances not suited for these materials, despite the fact that the band gap of a material like GaAs (1.42 eV) is close to optimal. A high surface recombination velocity will result in significant recombination losses in the emitter. One way to suppress the dark current contribution of the emitter is to use a heterostructure with the emitter as the wideband-gap material or to cap the emitter with a wide-band-gap material. Because of the high band gap in the window layer, the cell becomes less sensitive to the actual surface. Assuming that interfacial recombination is low, the contribution of the wideband-gap emitter to the total recombination current is normally negligible, as can be seen from eq. 3.19 (see e.g. Hovel, 1975), which shows the ratio of the dark currents between base and emitter in such a heterojunction structure. This is caused by the dependence of the intrinsic carrier concentration on the semiconductor band gap, as given in eq. 3.1b. C iab

^(emitter)

(3

19)

n, (base)

The top layer with the larger band gap is called the 'window' layer, whereas the underlying base layer is often called the 'absorber' (see e.g. de Vos, 1976). Table 3.3 shows the band gap and electron affinities for the most common wide-band-gap window layers. In general, the lattice mismatch between the two materials of the heterostructure should be low to avoid large recombination losses at the defective interface. Figure 8.10 gives more information about the lattice constants of semiconducting materials used in photovoltaic devices. The requirement of lattice matching can be relaxed in specific circumstances (see e.g. Tsai, 1980 for a discussion of the ITO/InP heterojunction). The effect of the misfit dislocations can also be minimised when

Principles of Cell Design

Table 3.3

117

Energy gap and electron affinity of commonly used window layers Energy gap/eV

Semiconductor

Electron affinity/eV

CdS

2.42

4.5

ZnS

3.58

3.9

ZnO

3.3

4.35

In203:Sn

3.7-4.4

4.5

Sn0 2 :F

3.9-4.6

4.8

appropriate passivation mechanisms are available. For example, the hydrogenation of Si/SiGe-cells (Said, 1998) leads to low interface recombination velocities of the order of 1000 cm s_l even though the lattice mismatch in that specific experiment was as large as 0.4%. A large mismatch might even be advantageous in circumstances where the cells are subjected to strong radiation, as are space solar cells. Although the beginning-of-life efficiency of InP/Si solar cells is lower than that of high-efficiency space solar cells, based on InGaP/GaAs combinations, InP/Si cells are relatively insensitive to radiation (they are 'predamaged'). As a result their end-of-life efficiency is higher than for their high-efficiency counterparts, which makes this type of cell attractive for use in low-Earth high-radiation-level orbits.

3.4.4 Window layers Contact

2.0

3.0

Absorber

Window

TCO

Contact

4.0

Photon Energy, eV

(a)

(b)

Figure 3.18 (a) Blue response improvement by the use of a wide-band-gap window layer (AlGaAs) on a GaAs cell (after Hovel, 1975, reprinted with permission from Academic Press); (b) band-gap diagram of a CdS / CdTe heterojunction cell with a Transparent Conductive Oxide (TCO) contact layer.

118

J. Poortmans, J. Nijs and R. Mertens

A variety of heteroj unctions with wide-band-gap window layers exists. Eminent examples are AlGaAs window layers on GaAs and CdS layers on CdTe, CuInSe2 layers and InP. Figure 3.18a shows the improvement of the blue response produced by the window layer. In Fig. 3.18b the energy band diagram for a CdS/CdTe cell is shown. 3.4.5 Homojunction emitter design and its relation to surface passivation Although thermal diffusion is certainly not the only way of realising a homojunction emitter, this technique has been extensively studied in Si solar cells and it has also been used for other types of homojunctions (see e.g. InP-homojunctions studied in the work of Yamamoto, 1984). This has allowed a clear model to be constructed for all the phenomena occurring in diffused homojunction emitters. The minority carriers in a thermally diffused emitter are not only moving under the influence of minority carrier concentration gradients. Because of the high and rapidly changing doping level two other forces contribute to the transport of the charge carriers. The first is the electrical field resulting from the doping gradient. As long as the excess carrier concentration is much lower than the doping level (which is normally the case in the emitter), the strength of this field is given by *

)

^ q

^ ax

|

(3.20, /Vj

where Nd (x) is the dopant concentration at depth x in the emitter. For diffused profiles with the maximum dopant concentration at the surface, this electrical field will aid the minority carriers to move in the direction of the junction, and therefore has a beneficial influence on the collection of photogenerated carriers. The term frontsurface field is also used to describe this effect. This positive effect can be maximised by going to extremely shallow junctions, achievable by Rapid Thermal Processing techniques (see e.g. Sivoththaman, 1997). A second contribution to the field in the emitter stems from heavy-doping-induced effects. The high doping level in the emitter reduces the semiconductor energy gap. This is the result of the breaking down of the one-electron approximation at high doping levels. The charge carriers can no longer be considered as non-interacting, hence correlation and exchange contributions have to be included in the energy of the carrier. In addition, the short-range variation of the electrostatic potential caused by doping density fluctuations leads to the formation of energy band tails near the

Principles of Cell Design

119

conduction and valence band edges. Together with the shift of the Fermi level into the majority carrier band, these changes result in the so-called apparent heavy-doping induced band-gap narrowing (BGN). The extent of narrowing depends on the doping level and the dielectric constant of the material. Figure 3.19 shows the band-gap narrowing for heavily doped n-type Si. _ 60

_

o A D

••

•

X

Theory Dumke Mertens Wieder Neugroschel et al Possin Del Alamo Tang

^ y ^

2*° /, cP ^/

0°

/ "V / / , ,*

t

10,s

•

•

,

,

10 1 9

.

1

10 2 '

Dopant concentration (cm"3)

Figure 3.19 Heavy-doping induced BGN as a function of doping for heavily doped n-type Si. Jain and Roulston, 1991, reprinted with permission from Pergamon Press.)

(From

As a result, the band gap and the position of the band edges become positiondependent, leading to a quasi-electrical field that points in the same direction for minority and majority carriers. The magnitude £(BGN) of this quasi-field is given by £(BGN) =

kTdnj 1 q dx n)e

(3.21)

where ( At/ 4 (BGN)^ nie = n, exp kT

(3.22)

Here AUg is the apparent band-gap narrowing taken from Fig. 3.19, n, is the intrinsic carrier concentration at low doping levels, and nie is referred to as the effective intrinsic carrier concentration. The quasi-field points in the direction of the higher doping level and hinders the transport of minority carriers towards the junction. The total emitter field is the sum of these two contributions. As a result the flux j for the minority carriers in an emitter has to be written as j = normal diffusion + drift by dopant gradient + drift by band-gap variation

(3.23a)

120

J. Poortmans, J. Nijs and R. Mertens

For minority hole current moving through an n-type emitter this becomes kT dp i„=-Wp——-
kT d/Vrf (x) 1 dnl 1 " +qpnp-^— q dx Nd(x) dx nie

(3.23b)

These effects make it difficult to find analytical solutions for the photogenerated and dark current in the emitter. Numerical methods are required instead (see Tsaur, 1972 for GaAs-cells and Fossum, 1976 for Si-cells).

0) & 101 " . "i : 1 m

• ^ •

noTCA, 1000C, W / F G A no TCA, 950C, w / F G A w / TCA, A1 annealed fit for no TCA. 1000C, w/FGA

10 17 10 1 8 10 19 10 2 0 Surface Phosphorus Concentration {cm"3)

(a)

(b)

Figure 3.20 (a) Surface recombination velocity as a function of surface concentration for phosphorusdoped n*-p Si solar cells (after King, 1990, reprinted with permission from IEEE); (b) schematic crosssection of the selective emitter structure.

When the minority carriers, injected in the emitter under forward bias, are able to reach the emitter surface, the emitter is said to be 'transparent' and the forward dark current, injected into the emitter, is sensitive to the surface recombination velocity at the emitter surface. One can then normally observe a strong increase of the surface recombination velocity with doping level. This is seen in Fig. 3.20a, where the dependence of the surface recombination velocity on the surface concentration is shown for n-type Si passivated by a thermally grown oxide. At first sight this would lead the designer of a photovoltaic cell to reduce the surface doping concentration of a transparent emitter. This will however lead to significant recombination losses at the contact regions, where the interface to the ohmic contact is anyway characterised by a large surface recombination velocity.

Principles of Cell Design

121

The contact resistance between metal contact and emitter layer should also be taken into account. In certain cases {e.g. for screen-printed contacts), a high surface concentration is essential for a low contact resistance. As a result, the design of a homogeneous emitter is a trade-off between the wish to avoid dead layers at the surface and the reduction of contact resistance and the screening of the high surface recombination velocity at the contacts at the other. This trade-off can be avoided by employing a selective emitter, as shown schematically in Fig. 3.20b. The emitter in the region between the fingers is different from the emitter beneath the metal fingers. The inter-finger emitter is shallow and the emitter surface should be well passivated. Beneath the finger, the emitter profile is deep with a high dopant surface concentration to screen the contact. This ensures a low contact resistance, acts as a buffer layer hindering the in-diffusion of metal impurities from the contact and also as a gettering site for impurities in substrates with high metallic impurity content. The highest efficiencies for Si solar cells have been realised in cells with a selective emitter.

3.5 Bulk recombination losses and their reduction In the context of this section, the term 'bulk of the cell' covers the lightly doped part of the cell, either in the homo- or heterojunction case. In contrast to the homojunction emitter or the window layer in the heteroj unction case, where the sheet resistance is a primary concern and high doping levels are the rule, the bulk is characterised by a lower doping level to maintain high minority carrier diffusion lengths. Most of the photogenerated current in a Si solar cell is produced by absorption of photons in the base and collection of the photogenerated carriers by the emitter-base junction. The collection efficiency is strongly influenced by the minority-carrier lifetime. This is especially true in indirect semiconductors, where the base must be sufficiently wide to allow the useful photons to be absorbed. Efficient carrier collection by the junction requires sufficiently high values of the minority-carrier lifetime, as determined by the various recombination processes. In an indirect semiconductor such as crystalline Si, radiative recombination is unimportant and Auger recombination is the only fundamental recombination process. This is generally important only in the heavily doped emitter or back-surface-field regions where the doping levels exceed 1018 cm"3. Near-to-Auger-limited diffusion lengths can be achieved by the use of high purity material with low defect densities. In the case of Si, the best quality monocrystalline material is grown and purified by the float-zone technique.

122

J. Poortmans, J. Nijs and R. Mertens

However, highly purified material with low defect density is in most instances not compatible with low cost, a condition for further market penetration of photovoltaic devices. The following paragraphs deal with the situation where grain boundaries and other crystallographic flaws are present in the polycrystalline materials used nowadays in photovoltaic thin-film technologies (CdTe, CuInSe2 and related alloys, poly- and microcrystalline Si). A short description and classification of the effect of grain boundaries is presented, followed by an overview of techniques and designs to overcome the detrimental effects of grain boundaries and impurities or, more generally, what possibilities exist to cope with materials with relatively low diffusion lengths.

3.5.1 Electronic description of grain boundaries Grain boundaries are the transition regions between adjacent grains of crystalline material with different orientation. In principle, it should be possible to calculate the electronic structure of the material near the grain boundaries when the grain boundary properties are known. However, except for low-angle grain boundaries, the required geometrical information is usually completely missing. The problem is further complicated by the chemical decoration of the boundary regions by impurities that diffuse preferentially towards these regions. All these factors exclude, for the overwhelming majority of situations, an ab initio calculation of grain boundary effects. Nevertheless, two different classes of grain boundary effects can be distinguished. In the first model, which is the more popular, the basic band structure of the semiconductor is conserved in the grain boundary region. The deviations from the perfect crystal are incorporated as electronic states existing in the forbidden energy gap (see Fig. 3.21a). These states can have donor or acceptor character and are distributed throughout the band gap. Depending on the occupancy, a charge sheet will be formed at the grain boundary surface, and this electrostatically induces a depletion or accumulation charge around the grain boundary region. Table 3.4 shows the experimentally observed character of grain boundaries for different semiconductors. The reader can see that for CuInSe2 and CdTe the grain boundaries are relatively inactive. This useful property, together with their high absorption coefficients, makes these materials and their related alloys prime candidates for polycrystalline thin-film solar cell technologies.

123

Principles of Cell Design

Table 3.4

Character of grain boundaries for a number of common semiconductors Semiconductor

Grain boundary behaviour

Ge

Strongly p —» accumulation in p -type

Si GaAs

midgap —» depletion in n- and p-type

CuInSe2

midgap —> depletion in n- and p-type upper midgap —» depletion in p-type —> accumulation in n-type no pinning

CdTe

low grain boundary activity

InP

The second model of grain boundaries, shown in Fig. 3.21b, describes the grain boundary region as a thin layer of material with different properties (Sutton, 1989 and references therein). Whatever the model used, the crucial element is the formation of interface states near the grain boundary region. In the case of Si, the interface state density, measured at the grain boundaries, is always in the region of 1012-10l3cirf2 (Sutton, 1989; Evrard, 1995).

J^ uf.

J^ (a)

(b)

Figure 3.21 (a) Model to describe the effect of a grain boundary by means of additional localised states at the interface; (b) Model to describe the effect of a grain boundary by considering the material near the grain boundary as a different semiconductor material.

3.5.2 Effect of grain boundaries on majority carrier transport From these models, the effect of grain boundaries on majority-carrier transport is readily understood. From Table 3.4, we see that, the grain boundary usually gives rise to a depletion layer around itself. Depending on the grain size, different situations might occur. As long as the grain size is such that the grain is only partially depleted, the depletion layer represents an energy barrier for the majority carriers (Fig. 3.22a).

124

J. Poortmans, J. Nijs and R. Mertens

The dominant charge carrier transport mechanism taking majority carriers across the barrier is thermionic emission. At high doping levels, tunnelling provides an additional transport mechanism. Thermally activated conduction is indeed found for multi- and polycrystalline semiconductors (for Si see e.g. MSller, 1993 and references therein, for GaAs see e.g. Cohen, 1980; for CdTe see e.g. Thorpe, 1986). The height of this barrier is a function of illumination level. Under illumination the excess minority carrier charge will lead to reduction of the charge associated with the barrier and as a result the barrier height decreases, behaviour also observed for surface states.

(a)

(b)

Figure 3.22 (a) Schematic band-gap representation of the grain boundary effect when the grainboundary-associated depletion layers do not overlap. This situation is typical for large-grained material (e.g. multicrystalline Si); (b) schematic band-gap representation of the grain boundary effect when the grain-boundary-associated depletion layers overlap. This situation is typical for small-grained lightly doped material. One can see that the barrier height becomes smaller as compared with Fig. 3.22a.

When the grain size becomes smaller than the depletion layer width associated with the grain boundary charge sheet, the grain will be fully depleted and the height of the barrier is reduced (see Fig. 3.22b). Under such circumstances the mobility can be high again, although scattering at the numerous grain boundaries will reduce the mobility in most instances below its value in the single-crystal material.

3.5.3 Effect of grain boundaries on carrier recombination The effect of grain boundaries on minority carrier transport is easily visualised. In the case of a depletion layer around the grain boundary, the minority carriers are attracted to the grain boundary, while for an accumulation layer, the minority carriers are repelled, an effect which is basically the same as in high-low doping transitions (see Section 3.4.1). The effect of the grain boundary on minority carrier recombination is best described as a grain boundary recombination velocity. Conceptually, this is the same as the surface recombination velocity and can be calculated if the capture cross

Principles of Cell Design

125

section and state distribution of the grain boundary states in the forbidden gap are known. Because of the dependence of the band bending on the injection level near the grain boundary, the recombination velocity is defined uniquely only for a grain boundary running parallel to the junction. For a vertical grain boundary the excess carrier concentration along the grain boundary is no longer a constant, resulting in a depth-dependent recombination velocity. The effect of horizontal and vertical grain boundaries has been studied theoretically by Green (1995, p. 298). For low surface recombination velocities at a horizontal grain boundary, the excess minority carrier profile will differ only slightly from the minority carrier profile in the case without the grain boundary. For this 'small perturbation case', the effect of the grain boundary is easily taken into account by increasing the dark current contribution of the base without the grain boundary by a factor K, given by

_L JL K=—

(3.24)

where % is the bulk lifetime, S the grain boundary recombination velocity, and Wb the base width. The recombination current of the region containing the horizontal grain boundary is increased by a factor K in comparison with the recombination current of the same region without the grain boundary, which is given by eq. 3.25. b

ib=qj

^nMdx 0

(325)

%b

For high grain boundary recombination velocities at a horizontal grain boundary in a quasi-neutral region, the grain boundary competes with the actual junction for the photogenerated carriers. The carriers generated closer to the grain boundary will have more chance of recombining at the grain than being collected by the junction. When the grain boundary falls within the junction depletion layer, more elaborate expressions are needed. As long as the electrical field of the junction depletion layer is much larger than the field associated with the grain boundary, the effects are minor.

126

J. Poortmans, J. Nijs and R. Mertens

For vertical grain boundaries, the situation is much more complex. For a full description of the effects, a 2-dimensional approach is needed (see e.g. Edmiston, 1996). The reduction of the field in the junction depletion layer by interaction with the field in the grain boundary depletion layer can lead to a drastic increase of the recombination current (see e.g. Beaucarne, 1998: see also Fig. 3.23 for a schematic 2dimensional picture of the potential distribution in the case of a vertical grain boundary crossing the junction). The arrows indicate that forward injection of electrons is enhanced in the region where the potential barrier is lowered by this effect. In terms of cell design, this means that the doping level near the junction should be sufficiently high to have a strong field in the depletion layer. Whether this is compatible with an appropriate diffusion length in the base of the cell depends on the specific properties of the material used (for small-grain polycrystalline Si, see Beaucarne, 1998). /

I grain boundary

Figure 3.23 Two-dimensional picture of the potential distribution for a vertical grain boundary near the junction depletion region. (After Edmiston, 1996 and Beaucarne, 1998.)

3.5.4 Designs to cope with low diffusion lengths The presence of grain boundaries and other deviations from a perfect crystalline structure will in most instances result in less efficient collection of the photogenerated carriers. There are a number of design possibilities to cope with this situation. In this section, distinction is made between three approaches: graded base design and the development of cell structures that go beyond the basic p-n junction (p-i-n cells and parallel multijunction cells).

127

Principles of Cell Design

Graded base design In a graded base design, the properties (composition, doping) of the base region are varied over the base so as to provide an additional force besides diffusion to enhance collection of the photogenerated carriers. Detailed expressions for the photogenerated and dark current have been derived (see Hovel, 1975 and references therein) for the case of a constant electrical field in a p-type base under the assumption of constant carrier mobility and lifetime. The full expressions are lengthy, but the main result is that the effective diffusion length of electrons is increased as compared with the normal diffusion length Le according (Zheng, 1998) S{x)

ku = K

(3.26a)

+1

where £(x) is the local electrical field and £c the critical field strength, given by kT £,.

(a)

(3.26b)

=•

qK

(b)

Figure 3.24 Schematic band-gap representations of (a) the case where a graded base is realised by grading the doping over the base. In this example the base is p-type and the lowest base doping is near the junction region; (b) the case where a graded base is realised by grading the band gap over the base.

The most straightforward implementation of the graded base concept is shown in Fig. 3.24a, where the electrical field is realised by means of a graded dopant profile. The advantage of this approach is, however, diminished by the dependence of the transport parameters (mobility and lifetime) on the doping level. These degrade at

128

J Poortmans. J. Nijs and R. Mertens

higher doping levels anu una uuavio uiv uututHugv KJX in^ nv^iu. ncvci uicicss, an efficiency of 16.4% was realised by grading the base doping in a thin-film crystalline Si solar cell with a thickness of about 30 pm (Zheng, 1998). A better approach is to grade the band gap of the base so that the lowest band gap is reached at the depletion layer, as shown in Fig. 3.24b. Such a band-gap profile can be realised by intentionally changing the growth conditions during base growth, but may also result from interdiffusion phenomena. As an eminent example of the effects of band-gap grading, the reader is referred to the paper of Gabor (1996) on the effect of grading the band gap in CuInSe2/CuGaSe2 solar cells. p-i-n cells A second approach for coping with low diffusion lengths consists in broadening the zone where the electrical field of the junction is effective. This is achieved by including an intrinsic (i) part in the transition zone between the p- and n-type regions. The resulting band scheme is shown in Fig. 3.25. The photogenerated carriers in the depleted zone are immediately separated and accelerated away from each other by the electrical field. This type of design is particularly advantageous when the mobility and the diffusion constant are low and the absorption coefficient high. The design criterion to decide between a p-n junction and p-i-n cell type can be written as I »T£*P>JH—x

kj 0-21)

where Siev is the field in the extended depletion layer between the p+ and n+ region and ris the lifetime of the excess carriers. If this condition is satisfied, si p-i-n cell will deliver a higher output current than its p-n counterpart. A drawback of the p-i-n concept is related to the fact that the superposition principle is no longer valid. Under forward bias, the internal field decreases, resulting in a lower photocurrent. As a result, the fill factor will be lower than for the p-n junction. The p-i-n approach is widely adopted for a-Si:H and microcrystalline Si cells (Shah, 1997). In the former material the mobility is low (of the order of 1 cm2 V"1 s"1). The reason for the better performance of p-i-n microcrystalline Si solar cells as compared with their p-n counterparts remains a matter of discussion, because there is no clear understanding at the moment of the diffusive and drift components of the currents in this cell type.

129

Principles of Cell Design

p+-layer

intrinsic layer

n+-layer

Figure 3.25 Schematic band-gap representation of the p-i-n design. The electrical field in the intrinsic part ensures efficient separation of the generated electron-hole pairs.

Parallel multijunctions A third way of ensuring efficient collection of the photogenerated carriers in a material with low minority carrier lifetime and diffusion length is the so-called parallel multijunction approach, shown in Fig. 3.26a (see also Green, 1994). Fig. 3.26b shows a pictorial view of the tolerance of this cell type for active grain boundaries. In addition, the requirements on surface passivation become less stringent for the multijunction cell.

(a)

(b)

Figure 3.26 (a) Schematic cross section of a parallel multijunction cell, consisting of alternating n and p type layers, connected in parallel; (b) pictorial representation of the immunity of the parallel multijunction for grain boundaries. The shaded region shows the zone where the grain boundary competes with the p-n junction for the carrier collection. (Reprinted with permission from The Centre for Photovoltaic Devices and Systems.)

The most important design criterion in this cell is to make the thickness of the individual layers smaller than the diffusion length, in order to keep the collection

130

J. Poortmans, J. Nijs and R. Mertens

efficiency close to unity. Although at first sight the reader might expect the thickness of the individual layers also to play a strong role in the reduction of series resistance losses, the designer has large freedom in optimising their thickness thanks to reinjection effects between different layers (Honsberg, 1994). The increase of the number of depletion regions in the multijunction cell raises concerns about recombination in the junction region, which might dominate the dark current behaviour (Stocks and Blakers, 1995). However, careful design of the junctions by inclusion of an intrinsic region in the depletion layer and optimisation of the field strength are predicted to overcome these difficulties (Green, 1996). Efficiencies as high as 15% are predicted for 10 /im Si-films with a lifetime of only 50 ns (Shi, 1994). First experiments with CVD-grown epitaxial multilayers, grown on inactive Si carrier substrates, confirm the high potential efficiency of this type of cell (see Zheng, 1996) with an efficiency of 17.6% for an active layer thickness of only 17 /im.

3.5.5 Practical example of bulk recombination in Si In the base of crystalline Si cells, the minority carrier lifetime is mainly determined by Shockley-Read-Hall recombination centres. The nature and the density of these centres strongly depend on the Si substrate material. In monocrystalline Si, the number of defects is smaller than in multicrystalline Si because grain boundaries interrupt the periodicity of the lattice and create states in the energy gap of the semiconductor through which recombination can possibly occur. Multicrystalline Si materials are also often characterised by much higher intragrain defect concentrations than single crystalline silicon. Metallic impurities Metallic impurities in silicon gives rise to states in the forbidden energy gap through which recombination can occur. This effect is often expressed by a threshold concentration in silicon for which the cell efficiency is reduced by 10%. These concentrations vary from 10" atoms cm"3 for impurities such as tantalum and molybdenum up to 1017 atoms cm"3 for copper. Fortunately, impurities with the smallest threshold concentration also have the smallest segregation coefficient (this is the ratio of then concentration of the impurity in solid silicon to the concentration in the liquid phase). Segregation also explains why the efficiency of some multicrystalline Si cells is reduced less by the presence of contaminants in the feedstock than that of monocrystalline Si cells: the rejection of metallic impurities into the liquid

Principles of Cell Design

131

occurs more efficiently during the solidification of a multicrystalline ingot because of the larger solid-liquid interface and the lower solidification rates. Moreover, the impurities can be removed to the grain boundaries leaving extensive high-purity intragrain crystal regions. For some metallic impurities the presence of oxygen or carbon plays an important role and the trapping process at grain boundaries via oxygen or carbide bonds enhances the electrical activity of grain boundaries (Pizzini, 1985). Carbon and oxygen are important impurities in Si. The concentration of these two elements in Si substrates depends on the preparation method. Oxygen is unavoidably introduced by the quartz crucible during the Czokralski process used to prepare monocrystalline Si and carbon is particularly important in multicrystalline ingots prepared in graphite moulds. It is well known that oxygen and carbon mutually interact as they segregate onto defects. The segregation process depends on the relative concentration of both elements as determined by the ingot preparation method. The segregation of oxygen and carbon influences the electrical activity of extended defects such as grain boundaries. Although extended defects and isolated oxygen and carbon atoms in silicon are often electrically inactive, interaction among them as well as with metallic impurities can induce electrical activity. Lifetime-restoring treatments Gettering and hydrogen passivation are generally used in order to increase the minority-carrier lifetime in multicrystalline silicon having electrically active defects. Gettering requires high temperatures (typically 900 C) to allow efficient removal of metallic impurities to an intentionally damaged layer at the front or backside of the cell. This layer can eventually be removed during subsequent processing. Hydrogen passivation occurs at much lower temperatures (typically 300-400 C) and is therefore done at the end of the process. Gettering Gettering can be aimed at either the elimination of defects or the removal of metallic contamination. Internal gettering is provided by intrinsic or extrinsic defects created in the interior of the material. In the case of external gettering, the gettering sites are generated by an external processing step such as a phosphorus diffusion or an aluminium-alloyed back-surface-field region. For solar cells in particular external gettering is important.

132

J. Poortmans, J. Nijs and R. Mertens

The three major steps in a gettering process are: 1. The dissolution or release of the contaminants in the active device area; 2. The diffusion of the contaminants; 3. The stable capturing of the contaminants at a predefined gettering region. In most cases this region is obtained by a heavy phosphorus diffusion with a large concentration of damage sites and therefore a high solubility for contaminants. Gettering can be particularly useful in the case of multicrystalline Si substrates. Narayanan et al. (1986) have shown that a phosphorus pretreatment can increase the diffusion length in multicrystalline substrates. Martinuzzi et al. (1988) studied phosphorus-gettering techniques in detail and found that an optimum temperature for gettering exists. This optimum, lying around 900 C, results from the competition between external gettering, which extracts impurities, and internal gettering, that decorates the grain boundaries and increases the recombination at defects in the material. The efficiency of gettering steps depends on the method used and phosphorus gettering cannot remove impurities from precipitates and clusters. Other experimental results have indicated that high-quality semicrystalline silicon substrates with initial diffusion lengths exceeding 150 /im do not improve substantially following a phosphorus-gettering treatment. On the other hand, it has been shown that phosphorous gettering and subsequent gettering by aluminium or boron are additive Hydrogen passivation It is well known that hydrogen atoms introduced into silicon act as terminators for the dangling bonds associated with the crystal defects in the multicrystalline material. Three ways to introduce atomic hydrogen have been proposed: ion implantation, plasma hydrogenation in a hydrogen glow discharge, and hydrogenation using a plasma- deposited silicon nitride layer. Hydrogenation can be performed through the front- or backside. There are contradictory results (see e.g. Elgamel, 1998) concerning the efficiency of frontside hydrogenation by hydrogen plasma and silicon nitride in the literature. Whereas some groups have reported significant improvements in cell efficiency resulting from frontside hydrogenation, others found none. It is generally agreed that frontside hydrogenation is critical in the case of oxide-passivated Si emitters. For these cells, the plasma power must be adjusted to avoid damage to the oxide during plasma treatment. If the plasma power is too high hydrogenation may actually result in an efficiency decrease. It has also been found that, in the case of cells which are covered by nitride on both sides, efficiency greatly improves after heat treatment at 700 C. Heat treatments after hydrogen implantation reverse the

Principles of Cell Design

133

degradation of the silicon/silicon dioxide interface which occurs during the electrode firing. This recovery of the surface passivation and the bulk lifetime suggests that some form of hydrogenation will remain necessary in most future high-efficiency cell processing lines, even if the initial diffusion lengths exceed the wafer thickness. Moreover, as for gettering, hydrogen passivation results in a narrower efficiency distribution in production.

3.6 Design and fabrication of the metal contacts This section will address the design of the metallisation scheme. First, the design of the most widespread two-sided contacting scheme will be discussed. The trade-off between series resistance loss and optical losses by reflection on the metal fingers determines the final design of the front grid. The importance of the grid design is even greater in concentrator cells, more details of which can be found in Chapter 12. The second part of this section will deal with one-sided contacting schemes.

3.6.1 Two-sided contact design Optimisation of the front contact pattern Most cell designs are based on a two-sided contacting scheme, a schematic illustration of which is shown in Fig. 3.27a and b. The carriers are collected by metal contacts at both sides of the cell. Fig. 3.28a shows a top view of the front-surface finger structure. The narrow fingers collect the current and transport it to the busbars, which are connected to the external leads. Fig. 3.27b shows a cross-section of the twodimensional contacting scheme. Although the carrier flow inside the cell is mostly vertical, the flow in the upper (emitter) layer of the cell is in the horizontal direction towards the collecting fingers. The main design parameters are those relating to the finger grid at the frontside of the cell: the finger spacing S and the finger width Wf. The treatment given below is generally applicable to homojunction cells (e.g. n+-p Si solar cells), heterojunction cells (e.g. CdS/InP, CdS/CuInSe2, CdTe/ITO) and MIS cells. The carrier flow in the base layer of the cell is assumed to be purely in the vertical x-direction and homogeneously distributed over the v-plane. Although this is a reasonable approximation, deviations occur because of the absence of generation under the metal fingers and the influence of resistance drops in the upper layer on the forward injected current.

134

J. Poortmans, J. Nijs and R. Mertens

Detailed optimisations based on the real two-dimensional carrier flow inside a Si highefficiency n+-p Si homojunction cell can be found in the work of Aberle,1994 (see also Chapter 2). wf

•

<

A*Finger

1

<—

t t t Jb Base K \

\

V k k

Busbar

(a)

(b)

Figure 3.27 (a) Typical front contact lay-out. 5 is the finger spacing and Wf is the finger width. In the simple case, the finger width is constant; (b) cross section of cell showing the vertical carrier flow in the base of the cell and the horizontal majority carrier flow in the emitter

When the carrier fluxy'fc towards the junction is homogeneously distributed and the upper layer is characterised by the sheet resistance pSheet> t n e resistive losses in the upper layer per unit length of the finger are given by 5/2

5/2

J l\x)dR=

J

0

0

fbb2x2p, „dx =

JbPshea"

(3.28)

This value should be compared with the maximum power delivered by the cell (again per unit length of metal finger), which is given by eq. 3.29, where /mp and Vmp are the current and voltage generated at the maximum power point. The relative loss, defined as the ratio of the contact loss to the maximum power, is then given by eq. 3.30. (3.29)

P =V I • mp

mp mp

Relative loss = -

,S2L 4V

(3.30)

Equation 3.30 illustrates the importance of the finger spacing and the sheet resistance of the upper layer. Achieving low values for the sheet resistance is

Principles of Cell Design

135

obviously a promising way to reduce the resistive losses. Decreasing this by going to deeper emitters or thicker transparent conductive layers will, however, increase other losses. In both cases part of the upper layer will no longer contribute to the current but instead act as a 'dead layer'. Screen printing In this technique, widely used for industrial non-concentrating Si solar cells, a metalcontaining (mostly Ag) paste is printed through a suitably patterned screen. Besides the metal particles and organic binders, the paste contains glass frit (dispersed glass particles). After printing, the paste is dried and subjected to a high-temperature cycle ('firing'). The back and front metallisations can be fired separately or together (cofiring). After firing, the Ag particles form a series of point contacts to the top of the emitter layer. Because of these point contacts, the front Si surface has to be heavily doped (>2 x 1020 cm"3) to keep the contact resistance below 10~3 Q. cm2. For the backside metallisation, Al is added to the paste. Contact resistance is less of a problem at the backside because of the larger metallised area. In addition, the Al-alloyed highly doped region at the backside provides an efficient back-surface field. The main advantage of screen printing is its simplicity and cost-effectiveness. Drawbacks which are often cited are the relatively large line width, the high contact resistance due to oxide precipitation from the glass frit, and the low (broad and flat) aspect ratio of the metal lines, which results in large shadowing losses. Recent efforts have focussed on these limitations and, in combination with the bulk and surface passivation caused by hydrogen out-diffusion from PECVD-nitride during a 'firingthrough process', significant progress has been made with the screen-printing technique for crystalline Si solar cells (Duerinckx et al., 1997 and 1998). The advent of low-cost methods for the realisation of a selective emitter (see e.g. Horzel et ai, 1997 and Ruby et al., 1997) provided an additional boost to the screen-printing technology. As a result, efficiencies between 17 and 17.5%, both on large-area monocrystalline (Nijs et al., 1996) and industrial-type multicrystalline (Shirasawa et al., 1994) Si cells have been reported. Plating/buried contact technology Plating is a technique whereby a metal, dissolved in an aqueous solution, is deposited on a substrate. This can be realised by passing a direct current through the solution (electroplating), but in the context of Si solar cells, the most widely used technique is electroless plating. Here the plating is performed by chemical deposition under such

136

J. Poortmans, J. Nijs and R. Mertens

circumstances that the metal deposition occurs selectively in regions where the Si surface is in direct contact with the solution. The low aspect ratio of the contacts is one of the drawbacks of the screen-printing technique. A solution to this problem is the so-called 'buried contact' technology, whereby the metal is electrolessly plated into deep grooves, obtained by laser scribing or mechanical cutting. In most instances a triple layer, consisting of Ni, Cu and Ag is used. A cross section of the final cell is shown in Fig. 3.28. plated finger in groove n +-emitter n ^emitter substrate

Figure 3.28

Schematic cross section of buried-contact structure.

The first versions of this technology made use of an oxide layer in between the grooves to avoid metal deposition (see e.g. Wenham, 1993). Subsequently this oxide layer was replaced by a nitride layer (Bruton, 1991) since this withstands high temperature steps and also provides better anti-reflection properties. The buriedcontact approach is inherently a selective emitter approach, because two different diffusions are carried out during the process. There is a shallow diffusion in between the grooves, but a very deep diffusion within the grooves. Together with the low shadowing losses resulting from the high aspect ratio of the fingers, this technology has produced efficiencies between 16 and 18% on mono- and multicrystalline largearea Si solar cells. Transparent conductive oxides If the sheet resistance of the emitter is high, the use of a finger contact pattern at the frontside of the cell is excluded. Under such circumstances, a high-conductance layer covering the front surface is necessary. The transparent conductive oxides (TCOs) are a special class of materials that exhibit good electrical conductivity, high optical transparency and a high band gap. This makes them especially useful as n-type 'window layers' or transport layers on top of the actual window layer. The best known TCOs are Sn0 2 , ZnO and ITO (which is ln 2 0 3 containing 0.2-2% Sn). Table 3.5 gives typical resistivities for some state-of-the-art TCOs.

137

Principles of Cell Design

Whereas the screen-printed and buried contact approaches are widely used in crystalline Si solar cells, TCOs are used in CdTe, CIS and a-Si:H cells. The common feature of all these cell types is their high sheet resistance, which excludes significant sideways current transport in the emitter window layer. Table 3.5

Resistivities of some state-of-the art TCOs

TypeofTCO In203:Sn

Resistivity/Q cm 4

1.5 x IV

4

Reference Wu, 1997

Sn0 2 ZnO:Al

3.3 x 10"

Wu, 1997

1.7 xHT 4

Zafar, 1995

Cd 2 Sn0 4

2.0 x 10"4

Wu, 1997

3.6.2 One-sided contact designs Although the two-sided contact design is the most straightforward approach at the cell level, there is strong interest in realising one-sided contacting schemes. An obvious advantage is the elimination of shadowing losses when all the contacts are at the backside. This is important in cells for concentrating systems, where shadowing losses are more important. Additionally, one-sided contact designs are particularly attractive at the level of module production. Monolithic integration of cells on a large insulating substrate could be an important factor in cost reduction because the series interconnection between the front contact of a cell to the backside of the next cell in a conventional module forms an important part of module production costs. A common feature of the one-sided contact designs are the more stringent requirements on the level of surface passivation and (or) bulk lifetime. The analysis and simulation of these structures is complex and requires the use of more advanced simulation tools. This will be illustrated by a number of examples taken from Si-based photovoltaics. In contrast with the two-sided contact case, it is in most cases not possible easily to decouple vertical and horizontal carrier. An eminent example of such two-dimensional model is the analytical model, developed by Swanson (1986) for the radial carrier flow near the contacts in a point-contact cell (see Fig. 3.30). In most cases, however, two-dimensional simulation tools are necessary. The reader is referred to the work of Hebling (1995), who relied on a two-dimensional numerical simulator (DESSIS) to optimise the contact grid for the interdigitated cell structure for thin-film crystalline Si solar cells on a non-conductive substrate (see Fig. 3.31).

138

J. Poortmans, J. Nijs and R. Mertens

Cell structures with all the contacts at the backside One can distinguish two versions of this cell type, the point-contact and interdigitated designs shown in Fig. 3.29. The main difference between the two is the reduced backside area with a high doping and metallisation in the point-contact design, which achieves higher open-circuit voltages because of reduced recombination at the surfaces (see Section 3.4.1).

I

sunlight (a)

Figure i.29

I

sunlight (b)

(a) Backside point-contacted solar; (b) backside interdigitated contact solar cell.

Common to both approaches is the extreme requirement on front-surface passivation. This is alleviated for application in concentrating solar cells (see Section 3.4.1). In addition, the bulk lifetime has to be very high in order to allow the carriers to diffuse to the backside. Because of this and the lower surface recombination achievable on rc-type Si as compared with p-type Si surfaces, n-type floatzone Si wafers are generally used. Using the point-contact approach, an efficiency of 22% has been achieved under 1-Sun conditions and 26.8% under 50-Sun concentration (Verlinden, 1995). The severe requirements on surface recombination velocity and bulk lifetime can be reduced by including a front-surface field to reduce front-surface recombination (Von Roos, 1978) or by going to the emitter wrap-through design shown in Fig. 3.30 (Schoenecker, 1997, Van Kerschaever, 1997). In the latter case, the base diffusion length necessary to ensure efficient collection of the minority carriers is only half the substrate thickness Modified versions of the EWT concept that reduce the number of holes to be made in the substrate are under development (Einhaus, 1997).

Principles of Cell Design

139

/; -Si

p ~Si

p +-region

p -region TSffl—BBBBBT

I p -contact Figure 3.30

+

n -contact

TCw

I p ^contact

Schematic cross section of the emitter-wrap-through (EWT) cell.

Cell structures with all contacts at the frontside Although the advantage of no shadowing loss is lost if all the contacts are at the frontside, several groups are working on this concept in the context of developing a thin-film crystalline Si solar cell technology on an insulating substrate (see Fig. 3.31).

Implanted Si02-layer Si-substrate

Figure 3.31

Frontside-contacted interdigitated cell (after Hebling, 1995).

The monocrystalline Si version of the structure in Fig. 3.31 was realised on SIMOX-wafers by growing a 50 /xm epitaxial Si layer by Chemical Vapour Deposition at high temperature. An efficiency of 19.2% was obtained, whereas an efficiency of 11% was obtained for a thin-film Si structure on graphite, both for smallarea cells.

140

J. Poortmans, J. Nijs and R. Mertens

3.7 Conclusions This chapter has given an overview of techniques and principles which will reappear regularly in the following chapters, where their application will be focussed on one type of material (Si, III-V compounds, polycrystalline thin-film solar cells) or one type of application (concentrators, space solar cells). Chapter 4 provides an excellent example where a lot of the concepts discussed in this chapter flow together in one design, the PERL design for high-efficiency Si solar cells. In the PERL design, concepts such as the selective emitter, local back-surface field, well passivated frontand backside surfaces, texturing to reduce reflectance and increase the optical path length and an efficient back side reflector are applied to bring the efficiency of the cell near its theoretical limit. These techniques and principles are the result of intensive research over the last forty years that has required the simultaneous understanding of electrical and optical effects. The remaining challenge is to find a cost-effective way to apply these principles to construct a low-cost solar cell with high and stable efficiency, a unifying idea striven for in the majority of the following chapters of this book.

References Aberle A., Glunz S. and Warta W. (1992), 'Impact of illumination level and oxide parameters on Shockley-Read-Hall recombination at the Si-Si0 2 interface', J. Appl. Phys. 71,4422-4431. Aberle A. G., Heiser G. and Green M. A. (1994), 'Two-dimensional minority carrier flow in high-efficiency silicon solar cells at short-circuit, open-circuit and maximum power point operating conditions', Solar Energy Mater. Solar Cells 34, 149-160. Abouelsaood A. A., Ghannam M. Y., Poortmans J. and Mertens R. (1997), 'Accurate modeling of light trapping in thin-film silicon solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 183-186. Andreev V. M., Chosta O. I., Khvostikov V. P., Paleeva E. V. and Shvarts M. Z. (1997), '3.75 MeV electron irradiation of III-V concentrator tandem cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 999-1002. Araujo G. L. and Marti A. (1994), 'Absolute limiting efficiencies for photovoltaic energy conversion', Solar Energy Mater. Solar Cells 33, 213-240.

Principles of Cell Design

141

Barbe M., Bailly F., Lincot D. and Cohen-Sala I. (1982), '13% conversion efficiency in cadmium telluride shallow homojunction solar cell—effects of temperature and intensity', Conf. Record 16th. IEEE Photovoltaic Specialists Conf., San Diego, IEEE Press, Piscataway, 1133-1137. Beaucarne G., Poortmans J., Caymax M, Nijs J. and Mertens R. (1999), 'Properties of p-n diodes made in polysilicon layers with intermediate grain size', Solid State Phenomena B (Diffusion and Defect Data) 67-68, 577-582. Bertness K. A., Kurtz S. R., Friedman D. J., Kibbler A. E., Kramer C. and Olson J. M. (1994), '29.5%-efficient GalnP/GaAs tandem solar cells', Appl. Phys. Lett. 65, 989-991. Brendel R., Hirsch M., Stemmen M , Rau U. and Werner J. (1995), 'Internal quantum efficiency of thin epitaxial silicon solar cells', Appl. Phys. Lett. 66, 1261-1263. Brendel R., Bergmann R. B., Lolgen P., Wolf M. and Werner J. (1997), 'Ultrathin crystalline silicon solar cells on glass substrates', Appl. Phys. Lett. 70, 390-392. Brendel R. (1997), 'A novel process for ultrathin monocrystalline silicon solar cells on glass', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 1354-1357. Campbell P. and Green M. A. (1986), 'The limiting efficiency of silicon solar cells under concentrated sunlight', IEEE Trans. Electron Devices 33, 234-239. Campbell P., Wenham S. R. and Green M. A. (1988), 'Light trapping and reflection control with tilted pyramids and grooves', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 713-16. Chen Z., Rohatgi A., Bell R. O. and Kalejs J. P. (1994), 'Defect passivation in multicrystalline-Si materials by plasma-enhanced chemical vapor deposition of Si02/SiN coatings', Appl. Phys. Lett. 65, 2078-2080. Chung B. C , Virshup G. F. and Schultz J. C. (1990), '27.6% (1-sun, air mass 1.5 G) monolithic two-junction AlGaAs/GaAs solar cell and 25% (1-sun, air mass 0) three-junction AlGaAs/GaAs/InGaAs cascade solar cell', Conf. Record 21st. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 179-183. Cohen M. J., Paul M. D., Miller D. L., Waldrop J. L. and Harris J. S. (1980), 'Schottky barrier behavior in polycrystal GaAs', J. Vac. Sci. Technol. 17, 889903. Coutts T. J. and Yamaguchi M. (1988), 'Indium phosphide-based solar cells: a critical review of their fabrication, performance and operation', Chapter 2 in Current Topics in Photovoltaics, Vol. 3, Coutts T. J. and Meakin J. D., eds., Academic Press, New York, 79-234. Del Alamo J., Van Meerbergen J., d'Hoore F. and Nijs J. (1981), 'High-low junctions for solar cell applications', Solid-State Electronics 24, 533-538.

142

J. Poortmans, J. Nijs and R. Mertens

de Vos A. (1976), 'Calculation of the maximum attainable efficiency of a single heterojunction solar cell', Energy Conversion 16, 67-78. Duerinckx F., Einhaus E., Van Kerschaever E., Szlufcik J., Nijs J., Mertens R., Roy M. and Narayanan S. (1997), 'Increase in efficiency and material yield by use of PECVD silicon nitride in a simple screen printing process on Solarex material', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 259-262. Duerinckx F., Szlufcik J., Nijs J., Mertens R., Gerhards C , Marckmann C , Fath P. and Willeke G. (1998), 'High efficiency, mechanically V-textured, screen-printed multicrystalline silicon solar cells with silicon nitride passivation', Proc. 2nd. World Conf. Photovoltaic Energy Conversion, Vienna, European Commission, Brussels, 1248-1253 Eades W. and Swanson R. (1985), 'Calculation of surface generation and recombination velocities at the Si-Si0 2 interface', J. Appl. Phys. 58,4267-4276. Edmiston S. A., Heiser G., Sproul A. B. and Green M. A. (1996), 'Improved modeling of grain boundary recombination in bulk and p-n junction regions of polycrystalline silicon solar cells', J. Appl. Phys. 80, 6783-6795. Einhaus R., Vazsonyi E., Szlufcik J., Nijs J. and Mertens R. (1997), 'Isotropic texturing of multicrystalline silicon wafers with acidic texturing solutions', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 167-170. Elgamel H. E. A (1998), 'High efficiency polycrystalline silicon solar cells using low temperature PECVD process', IEEE Trans. Electron Devices 45, 2131-2137. Evrard O., Duerinckx F., Vermeulen T., Horzel J., Poortmans J., Nijs J. and Mertens R. (1995), 'The effect of hydrogen passivation on the electrical transport properties and cell characteristics of SSP Si foils', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S. Stephens & Associates, Bedford, 13521355. Fossum J. G. (1976), 'Computer-aided numerical analysis of silicon solar cells', Solid-State Electronics 19, 269-277. Gabor A. M„ Tuttle J. R., Bode M. H., Franz A., Tennant A. L., Contreras M. A., Noufi R., Jensen D. G. and Hermann A. M. (1996), 'Band-gap engineering in Cu(In,Ga)Se2 thin films grown from (In,Ga)2Se3 precursors', Solar Energy Mater. Solar Cells 41-42,247-260. Gale R. P., Fan J. C. C , Turner G. W„ Chapman R. C. and Pantano J. V. (1984), 'Efficient AlGaAs shallow-homojunction solar cells', Appl. Phys. Lett. 44, 632634.

Principles of Cell Design

143

Gerhards C , Marckmann C , Tolle R., Spiegel M., Fath P., Willeke G. and Bucher E. (1997), 'Mechanically V-textured low cost multicrystalline silicon solar cells with a novel printing metallization', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, EEEE Press, Piscataway, 43-46. Ghannam M. Y. (1991), 'A new n + /w + - s t r u c t u r e w i m back side floating junction for high efficiency silicon solar cells', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 284-289. Green M. A. and Wenham S. R. (1994), 'Novel parallel multijunction solar cell', Appl. Phys. Lett. 65,2907-2910. Green M. A. (1996), 'Depletion region recombination in silicon thin-film multilayer solar cells', Progress in Photovoltaics 4, 375-380. Green M. A. (1995), Silicon Solar Cells: Advanced Principles & Practice, Centre for Photovoltaic Devices and Systems, University of New South Wales, Sydney. Hebling C , Gaffke R., Sterk S. and Warta W. (1995), 'Interdigitated grid for silicon thin-film solar cells on Si0 2 layers', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S. Stephens & Associates, Bedford, 1512-1515. Honsberg C. B., Edmiston S., Wenham S. R., Sproul A. B. and Green M. A. (1994), 'Capitalizing on two-dimensional minority carrier injection in silicon solar cell design', Proc. 1st. World Conf. Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1413-1416. Honsberg C. B., Ghozati S. B., Ebong A. U., Tang Y.-H. and Wenham S. R. (1996), 'Elimination of parasitic effects in floating junction rear surface passivation for solar cells', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 401^104. Horzel J., Szlufcik J., Nijs J. and Mertens R. (1997), 'A simple processing sequence for selective emitters', Conf Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 139-142. Hovel H. J. (1975), 'Solar cells', Ch. 2 in Semiconductors and Semimetals, Vol. 11, Willardson R. K. and Beer A. C , eds., Academic Press, New York. Jablonovitch E. and Cody G. (1982), 'Intensity enhancement in textured optical sheets for solar cells', IEEE Trans. Electron Devices 29, 300-305. Jain S. C. and Roulston D. J. (1991), 'A simple expression for band gap narrowing (BGN) in heavily doped Si, Ge, GaAs and GexSii_x strained layers', Solid-State Electronics 34,453-465. Jorgensen A. M., Clausen T. and Leistiko O. (1997), 'Novel light trapping scheme for thin crystalline cells utilizing deep structures on both wafer sides', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 175-178.

144

J. Poortmans, J. Nijs and R. Mertens

Keavney C. J., Haven V. E. and Vernon S. M. (1990), 'Emitter structures in MOCVD InP solar cells', Conf. Record 21st. IEEE Photovoltaic Specialists Conf., Kissimmee, IEEE Press, Piscataway, 141-144. King R. R., Gruenbaum P. E., Sinton R. A. and Swanson R. M. (1990), 'Passivated emitters in silicon solar cells', Conf. Record 21st. IEEE Photovoltaic Specialists Energy Conf., Kissimmee, IEEE Press, Piscataway, 227-232. Kluth O., Loffl A., Wieder S., Beneking C , Appenzeller W., Houben L., Rech B., Wagner H., Hoffmann S., Waser R., Selvan J. A. A. and Keppner H. (1997), 'Texture-etched Al-doped ZnO: a new material for enhanced light trapping in thinfilm solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 715-718. Kuhlmann B., Aberle A. G. and Hezel R. (1997), 'Characterization and optimization of the Al/SiC\/p-Si MIS contact in MIS-IL silicon solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 79-82. Kurtz S. R., Faine P. and Olson J. M. (1990), 'Modeling of two-junction, seriesconnected tandem solar cells using top-cell thickness as an adjustable parameter', J. Appl. Phys. 68, 1890-1895. Lolgen P., Leguyt C , Verhoef L. A., Sinke W. C , Eikelboom J. A., Steeman R. A. and De Moor H. H. C. (1994), 'Surface passivation by a floating junction', Proc. 12th.European Photovoltaic Solar Energy Conf, Amsterdam, H. S. Stephens & Associates, Bedford, 544-547. Poitevin H., Zurletto C , El-Ghitani H., Sarti D. and Martinuzzi S. (1988), 'Low temperature gettering effects in p-type polycrystalline silicon', Proc. 8th. European Photovoltaic Solar Energy Conf, Florence, Kluwer Academic Publishers, Dordrecht, 620-624. Martinuzzi S., El Ghitani H., Sarti D. and Torchio P. (1988), 'Influence of phosphorus external gettering on recombination activity and passivation of defects in polycrystalline Si', Conf. Record 20th. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 1575-1578. Mason N. B., Jordan D., Bruton T. M., Summers J. G., Hughes A. E., Whitfield R. and Bentley R.W. (1991), 'High efficiency silicon solar cells for concentrators', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 124-127. Metz A., Meyer R., Kuhlmann B., Grauvogl M. and Hezel R. (1997), '18.5% efficient first generation MIS inversion-layer silicon solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 31-34.

Principles of Cell Design

145

Metz A. and Hezel R. (1997), 'Record efficiencies above 21% for MIS-contacted diffused junction silicon solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 283-286. Moller H. J. (1993), Semiconductors for Solar Cells, Artech House, Boston. Narayanan S., Wenham S. and Green M. (1986), 'High efficiency polycrystalhne silicon solar cells using phosphorus pretreatment', Appl. Phys. Lett. 48, 873-875. Nijs J., Demesmaeker E., Szlufcik J., Poortmans J., Frisson L., De Clercq K., Ghannam M., Mertens R. and Van Overstraeten R. (1996), 'Recent improvements in the screen-printing technology and comparison with the buried contact technology by 2D-simulation\ Solar Energy Mater. Solar Cells 41-42, 101-117. Ohyama H., Aramoto T., Kumazawa S., Higuchi H., Arita T., Shibutani T., Nishio T., Nakajima J., Tsuji M., Hanafusa A., Hibino T., Omura K. and Murozono M. (1997), '16.0% efficient thin-film CdS/CdTe solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 343-346. Okamoto S., Nishida M., Shindo T., Komatsu Y., Yasue S., Kaneiwa M. and Nanmori T. (1997), '23.5% efficient silicon solar cell with rear microcontacts of c-Si//ZC-Si:H heterostructure', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 255-258. Pizzini S. (1985), 'Effects of impurities', in Silicon Processing for Photovoltaics, North Holland Physics Publishing. Ruby D. S., Yang P., Roy M. and Narayanan S. (1997), 'Recent progress on the selfaligned selective-emitter silicon solar cell', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 39—42. Said K., Poortmans J., Caymax M, Nijs J., Silier I., Gutjahr A., Konuma M., Debarge L., Christoffel E. and Slaoui A. (1998), 'Detailed analysis of the factors influencing the efficiency of thin-film crystalline Si^Ge* solar cells', Proc. 2nd. World Conf. on Photovoltaic Energy Conversion, Vienna, European Commission, Brussels, 36-41. Schmidt M., Brauer M., Bannach J. and Flietner H. (1994), 'Characterization of heterojunctions by photoinjection', Proc. 12th. European Photovoltaic Solar Energy Conf, Amsterdam, H. S. Stephens & Associates, Bedford, 519-522. Schoenecker A., De Moor, H. H. C , Burgers A. R., Weeber A. W., Hoornstra J., Sine W. C , Michiels P. P. and Steeman R. A. (1997), 'An industrial multicrystalline EWT solar cell with screen printed metallisation', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 796-799.

146

J. Poortmans, J. Nijs and R. Mertens

Shah A., Meier J., Torres P., Kroll U., Fisher D., Beck N., Wyrsch N. and Keppner H. (1997), 'Recent progress on microcrystalline solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 569-574. Sheng P. (1984), 'Optical absorption of thin film of a lambertian reflector substrate', IEEE Trans. Electron Devices 31, 634-636. Shi Z. and Wenham S. R. (1994), 'Polycrystalline silicon thin-film solar cells: the future for photovoltaics?, Progress in Photovoltaics 2, 153-162. Shirasawa K., Takahashi H., Inomata Y., Fukui K., Okada M., Takayama M. and Watanabe H. (1994), 'Large-area high-efficiency multicrystalline silicon solar cells', Proc. 12th. European Photovoltaic Solar Energy Conf., Amsterdam, H. S. Stephens & Associates, Bedford, 757-760. Shockley W. and Queisser H. J. (1961), 'Detailed balance limits of efficiency of p-n junction solar cells', J. Appl. Phys. 32, 510-519. Singh S. N. and Singh P. K. (1991), 'Modeling of minority-carrier surface recombination velocity at low-high junction of an n+-p-p* silicon diode', IEEE Trans. Electron Devices 38, 337-343. Sivoththaman S., Horzel J., Laureys W., Dureinckx F., De Schepper P., Szlufcik J., Nijs J. and Mertens R. (1997), 'Towards a fast and cost-effective production of industrial size silicon solar cells using rapid thermal processing and screenprinting', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 400^03. Stalmans L., Poortmans J., Caymax M., Nijs J., Vazsonyi E. and Mertens R. (1998), 'Porous silicon in crystalline silicon solar cells: a review and the effect on the internal quantum efficiency', Progress in Photovoltaics 6, 233-246. Stocks M. J. and Blakers A. W. (1995), 'Junction recombination in multilayer silicon solar cells', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S. Stephens & Associates, Bedford, 1219-1222. Swanson R. M. (1986), 'Point-contact solar cells: modeling and experiment', Solar Cells 17, 85-118. Takamoto T., Ikeda E., Kurita H. and Ohmori (1997), 'Over 30% efficient InGaP/GaAs tandem solar cells', Appl. Phys. Lett. 70, 381-383. Thorp D., Campbell P. and Wenham S. R. (1996), 'Absorption enhancement in conformally textured thin-film silicon solar cells', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 705708. Thorpe T. P., Fahrenbruch A. L. and Bube R. H. (1986), 'Electrical and optical characterization of grain boundaries in polycrystalline cadmium telluride', J. Appl. Phys. 60, 3622-3630.

Principles of Cell Design

147

Tiedje T., Yablonovitch E. and Cody G. (1984), 'Limiting efficiency of silicon solar cells', IEEE Trans. Electron Devices 31, 711-716. Tsai M.-J., Fahrenbruch A. L. and Bube R. L. (1980), 'Sputtered oxide/indium phosphide junctions and indium phosphide surfaces', J. Appl. Phys. 51, 26962699. Tsaur S. C , Milnes A. G.( Sahai R. and Feucht D. L. (1972), 'Theoretical and experimental results for GaAs solar cells', 4th. Symposium on GaAs, Am. Inst. Physics, College Park, MD, 156-167. Tuttle J. R., Berens T. A., Asher S. A., Contreras M. A., Ramanathan K. R., Tennant A. L., Bhattacharya R., Keane J. and Noufi R. (1995), 'Prescriptions for the fabrication of high-efficiency Cu(In, Ga)Se2-based thin-films and devices', Proc. 13th. European Photovoltaic Solar Energy Conf., Nice, H. S. Stephens & Associates, Bedford, 2131-2134. Van Kerschaever E., Nijs J., Mertens R. and Ghannam M. (1997), 'Two-dimensional solar cell simulations by means of circuit modelling', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 175-178. Vazsonyi E., Fried M., Lohner T., Adam M., Mohacsy T. and Barsony I. (1995), 'High-efficiency silicon PV cells with surface treatment by anodic etching', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S. Stephens & Associates, Bedford, 37—40. Verlinden P. J., Swanson R. M., Crane R. A., Wickham K. and Perkins J. (1995), 'A 26.8% efficient concentrator point-contact solar cell', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S. Stephens & Associates, Bedford, 1582-1585. Von Roos O. and Anspaugh B. (1978), 'The front surface field solar cell. A new concept', Conf. Record 13th. IEEE Photovoltaic Specialists Energy Conf, Washington D.C., IEEE Press, Piscataway, 1119-1120. Wenham R. (1993), 'Buried-contact silicon solar cells', Progress in Photovoltaics 1, 3-10. Winz K., Fortmann C. M., Eickhoff T. and Wagner H. (1997), 'Optical optimization of a-Si:H solar cells using planar junctions and diffuse rear reflectors', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 723-726. Wu X., Sheldon P., Coutts T. J., Rose D. H. and Moutinho H. R. (1997), 'Application of Cd 2 Sn0 4 transparent conducting oxides in CdS/CdTe solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 347-350.

148

J. Poortmans, J. Nijs and R. Mertens

Yamamoto A., Yamaguchi M. and Uemura C. (1984), 'High conversion efficiency n+-p InP solar cells', Proc. 1st. Int. Photovoltaic Solar Energy Conf., Kobe, 369372. Yamamoto K., Yoshimi M., Suzuki T., Okamoto Y., Tawada Y. and Nakajima A. (1997), 'Thin-film poly-Si solar cell with "Star Structure" on glass substrate fabricated at low temperature', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 575-578. Yang J., Banerjee A., Glatfelter T., Sugiyma S and Guha S. (1997), 'Recent progress in amorphous silicon alloy leading to 13% stable cell efficiency', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 563-566. Zafar S., Ferekides C. S. and Morel D. L. (1995), 'Characterization and analysis of ZnO:Al deposited by reactive magnetron sputtering', J. Vac. Sci. Technol. A 13, 2177-2182. Zechner C , Hahn G., Jooss W., Wibral M., Bitnar B., Keller S., Spiegel M., Fath P., Willeke G. and Bucher E. (1997), 'Systematic study towards high efficiency multicrystalline silicon solar cells with mechanical surface texturization', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 243-246. Zhao J., Wang A., Aberle A., Wenham S. R. and Green M. A. (1994), '717 mV opencircuit voltage silicon solar cells using hole-constrained surface passivation', Appl. Phys. Lett. 64, 199-201. Zhao J., Wang A., Altermatt P. and Green M. A. (1995), 'Twenty-four percent efficient silicon solar cells with double layer antireflection coatings and reduced resistance loss', Appl. Phys. Lett. 66, 3636-3638. Zheng G. F., Sproul A. B., Wenham S. R. and Green M. A. (1996), 'High-efficiency CVD multilayer thin-film silicon solar cells', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 465468. Zheng G. F., Zhang W., Shi Z., Thorpe D., Bergmann R. B. and Green M. A. (1998), 'High-efficiency drift-field thin-film silicon solar cells grown on electronically inactive substrates', Solar Energy Mater. Solar Cells 51, 95-104.

CHAPTER 4

CRYSTALLINE SILICON SOLAR CELLS MARTIN A. GREEN Photovoltaics Special Research Centre, University of New South Wales, Sydney, N.S.W. AUSTRALIA, 2052 [email protected] "Vast Power of the Sun Is Tapped by Battery Using Sand Ingredient." Front page headline, New York Times, 26 April 1954.

4.1 Overview Front page headlines in the New York Times and the Wall Street Journal in 1954 heralded to the world the demonstration of the first reasonably efficient solar cells, an event made possible by the rapid development of crystalline silicon technology for miniaturised electronics. Since that time, the majority of solar cells fabricated to date have been based on silicon in monocrystalline or large-grained polycrystalline form. There are two main reasons for this. One is that silicon is an elemental semiconductor with good stability and a well-balanced set of electronic, physical and chemical properties, the same set of strengths that have made silicon the preferred material for microelectronics. The second reason why silicon cells have been so dominant is that the success of silicon in microelectronics has created an enormous industry where the economies of scale directly benefit the presently smaller photovoltaics industry. Most silicon cells have been fabricated using thin wafers cut from large cylindrical monocrystalline ingots prepared by the exacting Czochralski (CZ) crystal growth process and doped to about one part per million with boron during ingot growth. A smaller but significant number use what are referred to as 'multicrystalline' wafers sliced from ingots prepared by a simpler casting (or, more generally, directional solidification) technique, which produces large-grained polycrystalline ingots. To produce a cell, these boron-doped starting wafers generally have phosphorus diffused at high temperatures a fraction of a micron into the surface to form the p-n junction required. Metal contacts to both the n- and the p-type side of the junction are formed by screen printing a metal paste that is then densified by firing at high temperature. Each cell is typically 10-15 cm either in diameter or along either side if square or rectangular.

149

150

M.A.Green

Figure 4.1 Exploded view of a standard silicon photovoltaic module. The different layers shown are laminated together under pressure at a temperature around 140-150 C where the transparent EVA (ethylene vinyl acetate) softens and binds the different layers together on cooling. Source: Green and Hansen (1998).

Cells generally are sold interconnected and packaged into a weatherproof, glassfaced package known as a module, as shown in an exploded view in Fig. 4.1. Each module typically contains 36 cells soldered together in series. Since each individual cell gives a maximum output of about 0.6 V in sunlight, this results in a module of over 20 V maximum output voltage, sufficient for fully charging a normal 12 V leadacid battery. The output current of each cell depends on its size and the sunlight intensity (solar irradiance) but generally would lie in the 2-5 A range in bright sunshine. The packaging consists of a glass/polymer laminate with the positive and negative leads from the series-connected cells brought out in a junction box attached

Crystalline Silicon Solar Cells

151

to the module rear. Such modules have proved extremely reliable in the field with all manufacturers offering a 10-25 year warranty on the module power output, one of the longest warranties on any commercial product (saucepans have been suggested as one of the few manufactured products with a comparable warranty period!). The efficiency of the cells in the module would typically lie in the 12-16% range, less than half the fundamental 'detailed-balance' limit of 33% for silicon (Tiedje et ai, 1984). Module efficiency is slightly lower than that of the constituent cells due to the area lost by frames and gaps between cells, with module efficiency generally lying in the 10-13% range. Over the last few years, commercial cells and modules of significantly higher performance have been available in multi-megawatt quantities using a more advanced cell processing technology (developed by the author's research group), discussed in more detail in the text. This technology produces cells of 17-18% efficiency and module efficiency in the 14-15% range. (Unless otherwise noted, all efficiencies quoted in this chapter are at standard test conditions, namely with a cell temperature of 25 C under 1000 W m~2 sunlight intensity with the standard global air mass 1.5 spectral distribution). This chapter discusses the historical and ongoing links between silicon solar cells and the broader microelectronics industry. Also discussed are standard and improved methods for preparing silicon cell substrates and for processing cells to extract as much performance as possible from such substrates at the lowest possible overall cost. The chapter also describes recent progress with supported silicon films. These provide a 'thin-film' approach for transforming silicon technology from the thoroughbred of the twentieth century to a much lower cost, market-leveraging workhorse of the twenty-first.

4.2 Silicon cell development The development of silicon photovoltaics is inextricably intertwined with the development of the general silicon electronics field and the subsequent founding of the microelectronics industry. The rapid increase in interest in the properties of doped silicon in the 1940s was triggered by the astonishing photovoltaic properties demonstrated by serendipitously formed p-n junctions, as described below. This increased interest led directly to the development of point contact and junction transistors and ultimately to integrated circuits. The earliest commercial silicon electronic devices were silicon point-contact or "cat's whisker" diodes which date from the early 1900s (Riordan and Hoddeson, 1997). These devices rectified electrical signals at a junction formed by pressing a thin

152

M. A. Green

metal wire against a piece of polycrystalline silicon (other semiconductors, such as silicon carbide, were also used). These cat's whisker diodes were key components in early radios. By the 1930s, thermionic valves had replaced these diodes in most applications. However, the evolving field of microwave technology created a renewed interest in the cat's whisker diodes in the mid-1930s. At Bell Laboratories in the USA, Russell Ohl guessed that impurities were the cause of the erratic behaviour often observed whereby the cat's whisker only operated correctly if located on a 'hot spot' in the silicon. Ohl therefore encouraged colleagues to grow samples of purer silicon, by melting the purest material available in a quartz capsule, and then cooling. In one specific ingot, the eighteenth in the series (Riordan and Hoddeson, 1997) which had been prepared by very slow cooling, Ohl and his colleagues found unusual properties, including a surprisingly large photovoltage of about half a volt when the ingot was illuminated by a flashlight. The silicon in this ingot showed two distinct types of properties, dubbed "positive" (p-type) and "negative" (n-type), depending on the polarity required for easy current flow between the material and a metal wire placed on the silicon surface, and also the polarity of voltage observed under illumination. It was quickly realised that the junction between the p-type and n-type regions, the p-n junction, was responsible for the unusual properties of the original ingot. The first silicon solar cells were formed by cutting the ingot to include sections with both a pandrc-typeregion and applying metal contacts (Ohl, 1941). These earliest silicon solar cells, shown in Fig. 4.2a, appear, based on available data, to have been only a fraction of percent efficient, but were still very much better in performance than earlier photovoltaic devices which had been based on selenium or cuprous oxide. The "grown-in" junctions in the earliest cells arose from the serendipitous distribution of p-type (boron) and n-type (phosphorus) impurities in the silicon resulting from the slow solidification process. Ohl realised that more controllable ways of forming the junction would be likely to give better performance. In the early 1950s, he was involved in experiments aimed at forming surface junctions by implanting helium at high energy into the surfaces of p-type polycrystalline silicon (Kingsbury and Ohl, 1952). Although this approach produced improved cells of efficiency estimated to be up to 1%, (Fig. 4.2b), this work was soon overtaken by independent improvements in silicon technology also made at Bell Laboratories, particularly in two areas. The first was the development of techniques for preparing single crystals of silicon using the Czochralski method. The second was the formation of junctions by the hightemperature diffusion of dopant impurities into the silicon surface. Combining these two techniques, researchers at Bell Laboratories were able to announce the first modern silicon solar cell in early 1954, in one of the early successes of the diffused junction approach (Chapin et al., 1954). Figure 4.2c shows the resulting cell structure.

153

Crystalline Silicon Solar Cells

top contact (+)

p-type

rear contact (-) grown junction (a)

p-type

(b)

p-type

Figure 4.2 (a) Silicon solar cell reported in 1941 relying on 'grown-in' junctions formed by impurity segregation in recrystallised silicon melts; (b) helium-ion bombarded junction device of 1952; (c) first modern silicon cell, reported in 1954, fabricated on single-crystalline silicon wafers with the p-n junction formed by dopant diffusion. Source: Green (1995).

154

M. A. Green

The impressive performance of these cells by previous standards, up to 6% energy conversion efficiency, created enormous interest at the time, as the newspaper headline at the beginning of this chapter suggests, and it also generated unbounded enthusiasm for the future of these devices. This enthusiasm proved to be premature, although the cells did find an almost immediate application in spacecraft. Space applications drove the rapid improvement in cell technology such that, by the early 1960s, cell energy conversion efficiency of about 15% under terrestrial sunlight had been demonstrated and cells had found a secure market niche in providing power for a rapidly increasing number of satellites (Wolf, 1976). The basic cell design which evolved (Fig. 4.3a), remained unchanged from the early 1960s for almost a decade. In the early 1970s, a reassessment of cell design at COMSAT Laboratories showed that a shallower diffusion combined with more closely spaced metal fingers could give a substantial improvement in the cell performance by improving the response to blue wavelengths (Lindmayer and Allison, 1973). The resulting cells, shown in Fig. 4.3b, known as "violet" cells due to their characteristic colour arising from the shorter wavelengths reflected, produced the first improvement in cell performance for over a decade. This improvement was augmented by the realisation that incorporating a thin heavily doped layer under the back contact, a so-called "back-surface field", gave unexpected benefits (Godlewski et al., 1973). This approach worked best if the rear doped layer was formed by alloying the underlying silicon with aluminium deposited over the rear of the cell. Not long afterwards, the idea of using anisotropic chemical etches to form geometrical features on the silicon surface was successfully demonstrated, also at COMSAT Laboratories (Haynos et al., 1974), and resulted in a further boost in cell performance, taking terrestrial cell performance to above 17% (Fig. 4.3c). The surface features consisted of square-based pyramids defined by slowly etching {111} crystallographic planes. These greatly reduced reflection from the cell surface, giving these 'black' cells the appearance of black velvet after antireflection coating. The improvements of the early 1970s came about primarily by improving the ability of the cell to collect carriers generated by the incoming photons. Since cells now appeared to be performing to close to their full potential in this area, it seemed that any further improvement in silicon cell performance would have to result from improved open-circuit voltage. Such improvement became the focus of work directed at increasing cell efficiency throughout the second part of the 1970s, largely as a result of a program directed by NASA-Lewis with this as a goal in order to improve space cell performance (Brandhorst and Bernatowicz, 1980).

Cryslailine Silicon Solar Cells

155

antireflection coatinc

top metal finger

i-type

thin n-type layer

p + layer textured surface (b) surface

(c) Figure 4.3 (a) Space silicon cell design developed in the early 1960s which became a standard design for over a decade; (b) shallow junction 'violet' cell; (c) chemically textured non-reflecting 'black' cell. Source: Green (1995).

156

M. A. Green

On the commercial front, the oil embargoes of the early 1970s had generated widespread interest in alternative sources of terrestrial energy. A small terrestrial photovoltaic industry came into existence largely as a result of the US Government's photovoltaic program. One component of this program (Christensen, 1985), arguably the most successful in terms of developing the industry and its products, involved a staged series of purchases of photovoltaic modules meeting increasingly stringent specifications. The first such purchase in 1975/6, known as "Block I", was remarkable for the diversity of both cell fabrication approaches and module encapsulation approaches used in the product supplied by four different manufacturers. One manufacturer, Spectrolab of Sylmar, California, supplied cells where the contacts had been applied using screen-printing (Ralph, 1975), the forerunner of the millions of cells of this type which were to follow. In the "Block II" purchases under this program (1976/7), the same company combined screen-printed cells with a laminated module design (Fig. 4.1), a combined approach that had been adopted by almost all commercial manufacturers by the early 1980s and, with relatively minor modification, remains the present commercial standard. The main features of a commercial screen-printed cell are shown in Fig. 4.4. The basic cell design is similar to that of a standard space cell of the 1960s (Fig. 4.3a), but incorporates the surface texturing of the 'black' cell of Fig. 4.3c as well, of course, as the screen-printing approach to applying the front and rear contacts. Since the Block II purchases of 1976/7, no major changes have been made in either the basic screen-printed approach to cell fabrication or to the cell encapsulation approach until quite recently. Considerable attention, however, has been directed towards reducing the cost of the silicon wafer, usually grown by the Czochralski technique, since this accounts for about 40% of the cost of a standard silicon module. The most successful approach has been the simplification of the ingot growth processes by using cruder directional solidification or 'casting' approaches to produce multicrystalline ingots (Ferrazza, 1996). The first multicrystalline silicon cells developed specifically for the terrestrial market were reported in 1976 (Lindmayer, 1976; Fischer and Pschunder, 1976) and commercial multicrystalline cells have been available since the late 1970s. These multicrystalline approaches involve basically a reversion to the earlier ingot-forming approaches for crystal rectifiers, techniques predating the microelectronics explosion. In 1998, multicrystalline silicon cells accounted for about 30% of the total market for photovoltaic product. Another major area of developmental emphasis has been to reduce the thickness of the silicon wafer by slicing it more thinly. This is resulting in a steady replacement of inner diameter sawing methods traditionally favoured by the microelectronics industry by wire cutting approaches, as described in more detail in Section 4.3.1.

Crystalline Silicon Solar Cells

157

150 urn P\ I

Figure 4.4

3mm

Screen-printed crystalline silicon solar cell (not to scale). Source: Green (1995).

On the research front, it had become apparent by the late 1970s that oxide passivation of the cell surfaces was the key to obtaining improved open-circuit voltage. The early 1980s saw a series of successively improved oxide-passivated cells fabricated by the author's group at the University of New South Wales (UNSW) taking silicon cell efficiency past 18%, then past 19% and finally 20%, the "four minute mile" of the photovoltaics area. The UNSW group has held the world record for silicon cell performance, almost without interruption, since this time. The UNSW-developed microgrooved PESC cell (passivated emitter solar cell) of Fig. 4.5a was the first silicon cell to exceed 20% energy conversion efficiency in 1985. The same basic approach has since been used by several other groups to produce cells of similar efficiency, with commercial quantities produced for solar car racing and for space. The approach is characterised by the use of a thin thermally grown oxide to "passivate" (reduce the electronic activity of) the top surface of the junction diffusion (the emitter of the cell), combined with the use of a shallow, high sheet resistivity phosphorus diffusion for this emitter. Another is the use of photolithography to produce relatively small contact area to this emitter region by defining openings in the "passivated oxide". Photolithography is also used to pattern the top contact fingers and to align these fingers to the oxide openings. The rear of the cells borrows the "alloyed-aluminium back-surface-field" approach from earlier space cells. In this approach, a layer of aluminium is deposited on the rear of the cell and

158

M.A.Green

Figure 4.5 (a) The microgrooved passivated emitter solar cell (PESC cell) of 1985, the first silicon cell to exceed 20% efficiency; (b) buricd-contact solar cell. Source: Green (1995).

alloyed into the cell at temperatures above the Si-Al eutectic. After cooling, this produces a layer of p-type Si heavily doped with Al at the rear of the silicon substrate. This reduces rear contact recombination rates. Some improvement of substrate quality also occurs during alloying by defect "gettering". A parallel PESC approach had, prior to 1998, given the best results for multicrystalline silicon cells, with efficiencies up to 18.6% demonstrated on such material by this approach (Rohatgi et al., 1996).

Crystalline Silicon Solar Cells

159

Cells of a similar quality to the first 20% efficient PESC cell have also found their way into high-volume terrestrial cell manufacture through the laser-grooved buriedcontact cell of Fig. 4.5b. This cell retains an alloyed aluminium rear and also incorporates the improvements in front-surface passivation first demonstrated by the PESC cell. To make the approach suitable for low cost production, however, the photolithographic metallisation of the PESC cell is replaced by a unique combination of laser grooving, to define the areas to be metallised, followed by electroless metal plating. The oxide in this case not only serves as top-surface passivation, but it also serves as a diffusion mask to confine the heavy diffusion to the laser grooved areas and as a plating mask for the subsequent plating of metal into these grooved areas. In commercial versions of this sequence, the oxide can be replaced by a high-temperature dielectric such as silicon nitride, as discussed in more detail in Section 4.4. The buried-contact approach now produces the highest performance terrestrial cells that are produced in any appreciable volume, with efficiency in the 17-18% range routinely obtained using standard low-cost commercial silicon wafers. The next major laboratory improvement in silicon cell design came in the use of oxide passivation along both the front and rear surfaces, as first demonstrated in the rear point contact solar cell developed by Stanford University. As shown in Fig. 4.6, this cell has an unusual design in that both positive and negative contact are made at the rear surface of the cell. Although this might, at first sight, appear to be a regression to a similar design to that used in the first modern silicon cell of Fig. 4.2c, there is a substantial difference in the way the two types of cells operate. For the modern rear contact cells, which take advantage of the excellent quality of silicon presently available, carrier diffusion lengths are several times the cell thickness, allowing carriers photogenerated near the top surface of the cell to diffuse to the rear contacts. In the earlier device, the junctions at top and rear surfaces are electrically connected by the junction around the cell edge. Most carriers in this earlier cell are collected by the top junction and flow around the edge of the cell to the rear contact, an approach that is only feasible for small area cells. The rear point contact cell demonstrated 22% efficiency in 1988 and has since been commercialised, finding use in photovoltaic systems which rely on concentrated sunlight and for high value-added applications such as solar car racing and high-altitude aircraft flights (Verlinden et al, 1997). The next improvement in silicon cell efficiency came, again at UNSW, by combining the earlier developments in the PESC cell sequence with the front and rear oxide passivation first demonstrated in the rear point contact cell. This is possible in a number of ways as shown in Fig. 4.7. In the PERC cell (passivated emitter and rear cell) of Fig. 4.7a, the first to be successfully demonstrated, rear contact is made to the

M. A. Green it busbar

Figure 4.6 Rear point contact solar cell which demonstrated 22% efficiency in 1988 (cell rear shown uppermost). Source: Green (1995).

silicon substrate through holes in the rear passivating oxide. This approach works reasonably well provided the substrate is sufficiently heavily doped for contact resistance between the metal and substrate not to be an issue (below about 0.5 Q cm resistivity for p-type substrates). The PERC cell is often suggested as a relatively low cost way for making silicon cells above 20% efficiency, since it is the simplest of the approaches of Fig. 4.7. Historically, the next improvement was demonstrated by the PERL cell (passivated emitter, rear locally diffused cell) of Fig. 4.7b. In this case, local diffusion is used in the area of the rear point contact to provide a minority carrier-reflecting region between this contact and the substrate and to reduce contact resistance. This approach produced the first 24% efficient silicon cell in 1994 (Zhao et ai, 1995) and holds the current world record of 24.5% (Zhao et ai, 1998). The PERL cell has been used in reasonably large quantities in solar car racing and in space cells. The third cell of Fig. 4.7c is the PERT cell (passivated emitter, rear totally diffused). To date this has not given as good a performance as the PERL cell but it offers some fabrication simplifications. The PERT cell has also been used as the basis of space cell production.

Crystalline Silicon Solar Cells

linger

"inverted- pyramids

rear contact

linger

oxde

"inverted" pyramids

rear contact

°><de

Figure 4.7 A family of four related high efficiency solar cell structures: (a) the passivated emitter and rear cell (PERC cell); (b) the passivated emitter, rear locally diffused cell (PERL cell) which took efficiency above 24% in the early 1990s; (c) the passivated emitter, rear totally diffused cell (PERT cell); and (d) the passivated emitter, rear floating junction cell (PERFcell). Source: Green and Hansen (1998).

162

M. A. Green

The final structure shown in Fig. 4.7d, the PERF cell (passivated emitter rear floating junction) offers perhaps the best long-term potential for high performance. This structure has produced the highest open-circuit voltage silicon cells to date with open-circuit voltage up to 720 mV demonstrated under standard test conditions (Wenham et ai, 1994), together with efficiencies above 23%. One feature of these more recent cell designs is the very effective trapping of light within the cell. By depositing metal over the entire rear surface of the cell but ensuring it is displaced from the silicon substrate by an intervening layer of oxide, very high rear reflectance for light striking this rear reflector structure from within the cell is obtained. When combined with appropriate geometrical structure on the front surface of the cell, weakly absorbed light that is reflected from this rear surface can be trapped quite effectively within the cell after total internal reflection from this front surface. This greatly extends the response of the cell to infrared wavelengths. Cells that convert such wavelengths with an efficiency approaching 50% have been demonstrated (Green et ai, 1992).Although oxide passivation remains the most effective technique for passivating cell surfaces yet demonstrated in experimental devices, recent work using wider band-gap amorphous or microcrystalline silicon layers has also produced encouraging results, as has work with specially deposited silicon nitride. A cell using an amorphous silicon emitter layer has been reported to give good performance (Tanaka et ai, 1993), although fundamentally limited to being less efficient than an oxide-passivated cell due to the poor electronic quality and the strong absorption in the amorphous silicon material. Similarly, cells with rear passivation using microcrystalline silicon layers instead of thermal oxide have produced quite encouraging results (Okomoto et ai, 1997) although they are also inherently not capable of matching the optical performance of oxide passivated devices due to parasitic absorption in the microcrystalline layer. Excellent surface passivation properties have also been reported for silicon nitride deposited by a remote plasma approach (Aberle et ai, 1997). How will cell design evolve in the future? Some insight is provided by Fig. 4.8, which shows the calculated intrinsic energy conversion efficiency bounds on single-junction silicon solar cells, with and without 'lambertian' light trapping. In 'lambertian' light trapping schemes, the light direction within the cell is randomised (Green, 1995) allowing path-length enhancements to be quite readily calculated (about 50 in idealised situations). The best laboratory cells have demonstrated close to 85% of the achievable efficiency, according to this figure. In the best experimental devices, performance losses of the order of 5% arise from less than ideal values of each of the short-circuit current, open-circuit voltage and fill factor parameters. The short-circuit current losses are most easily identified and reduced. These come from metal finger

Crystalline Silicon Solar Cells

163

coverage of the top surface, top-surface reflection loss, and less than perfect light trapping in the experimental cells. The voltage loss arises from finite surface and bulk recombination in excess of the lower limit imposed by intrinsic Auger recombination processes (Green, 1984). The fill factor loss comes not only from ohmic series resistance loss within the cell, but also from the same factors producing the opencircuit voltage loss. To eliminate the latter, parasitic recombination must be sufficiently reduced so that the dominant recombination component at the cell's maximum power point is Auger recombination. This is a more challenging requirement than the corresponding criterion at open-circuit voltage (Green, 1984). 3J

'•"'Tambertian

^^-——

/

3D -

/planar, cne pass

1

10 7

n \j

i

r Tvrnr|

10

1 — | | ) ITlfl

T~ I I I M i l l

100

1CO0

'

n - m n

10C00

Thickness (urn)

Figure 4.8 Limiting efficiency of a silicon solar cell as a function of cell thickness with and without lambertian light trapping (global AM1.5 spectrum, 100 mW cm""2, 25 C). Source: Green (1995).

As opposed to the case of laboratory devices, most manufacturers of commercial cells would be very pleased to be producing consistently cells of half the limiting efficiency of Fig. 4.8. Some of the difference between laboratory and commercial cell performance is due to poorer quality of silicon substrate material. A large component, however, is due to limits imposed by the present screen-printing process predominantly used for commercial cell fabrication. The penalty for the processing simplicity offered by this approach is a compromised cell design, since a heavily doped emitter layer appears unavoidable. Improved designs such as the buried-contact cell offer the seemingly contradictory advantages of both higher cell performance and lower overall manufacturing costs.

164

M. A. Green

It seems that eventually it should be feasible to produce low-cost commercial silicon cells of efficiency above 20% with such improved cell designs by paying attention to the passivation of both front and rear surfaces, by thinning the cells to reduce bulk recombination and by modifying the crystal growth processes to produce low-cost silicon customised for photovoltaics, particularly in its ability to withstand high-temperature processing without loss of electronic quality. An interesting result highlighted by Fig. 4.8 is the way that light trapping allows high performance, in principle, from silicon cells that are only 1 \im thick. This provides a justification for expecting very high performance, eventually, from the thin, supported silicon cells discussed in Section 4.7. To approach the limiting performance, the demands on bulk quality become less severe as the cell becomes thinner (Green, 1995). However, those upon light trapping and surface passivation become more severe. Various approaches have been suggested which have the potential, in principle, for exceeding even the efficiency limits of Fig. 4.8. These include the use of tandem cells, the use of high-energy photons to create more than one electron-hole pair (Werner et al, 1994), or the use of sub-band-gap photons in schemes such as incorporation of regions of lower band gap (Healy and Green, 1992), multiple quantum wells (Barnham and Duggan, 1990) or mid-gap impurity levels (Wolf, 1960). Experimentally, the tandem cell approach appears the most likely to have impact in the long term, once the problems with lattice-matching a top cell to silicon with a suitable band gap are overcome. Promising results have been demonstrated with aSi/c-Si tandem cells, although a better performing top cell will be required in the longer term to retain a performance advantage over that offered by the rear cell alone.

4.3 Substrate production 4.3.1 Standard process Not only is silicon solar cell technology capable of benefiting directly from the economies of scale of the silicon microelectronics industry, but also it is capable of using scrap material from this industry because the requirements for material quality in photovoltaics are less demanding than in the more general microelectronics field. Until the photovoltaic industry requires a larger volume of silicon than the microelectronics industry, it will be difficult for approaches customised for photovoltaics to compete with the costs of reject silicon from microelectronics. Accordingly, given the present size relativities, most silicon cells are made from

Crystalline Silicon Solar Cells

165

standard silicon source material originally intended for microelectronics. Over the last ten years, the size relativities have not changed enormously, since both industries have been steadily growing. Explosive growth in the photovoltaics industry, such as that stimulated by urban residential rooftop applications of photovoltaics in 1997, will increasingly upset this delicate balance. For microelectronics, the starting point for producing the requisite high quality "semiconductor grade" silicon is a lower grade of silicon known as "metallurgical grade", produced by the reduction in an arc furnace of quartzite by carbon, the latter generally in the form of wood chips. This metallurgical grade silicon is of about 98% purity and is produced in large quantities for the steel and aluminium industries. A relatively small quantity is refined for microelectronics by conversion to a volatile intermediary that can be purified by fractional distillation. The purified intermediate compound is then decomposed to re-extract the silicon in a highly purified form. Generally the metallurgical grade silicon is converted by hydrochloric acid to trichlorosilane which is then purified to 99.9999999% (nine "nines") purity by fractional distillation. Silicon is then extracted from the trichlorosilane by reducing the latter by hydrogen at high temperature. In this process electrically heated silicon rods are exposed to a trichlorosilane/hydrogen mixture which reacts on the surface of the rods, depositing silicon onto them and hence building up their cross section. These rods grow with a fine-grain polycrystalline silicon microstructure. After the rod diameter has increased to the required size, the process is stopped and the rods mechanically broken into smaller chunks, which maintain "nine-nines" purity. These chunks then become the starting point for the growth of ingots of good crystalline quality. As previously mentioned, crystalline ingots are generally grown by the Czochralski process. In this process, the purified silicon chunks are melted in a quartz crucible along with small pieces of silicon heavily doped with boron. This produces a boron-doped melt into which a seed crystal is dipped and slowly withdrawn (Fig. 4.9a). For high quality crystal growth, good temperature uniformity and slow and steady growth are required. Typically ingots are grown to about 10-15 cm in diameter and 1-2 metres in length, weighing 50-100 kg. The crystallographic orientation of the seed is transferred to the grown crystal. Generally, for photovoltaics, the crystal is grown with a preferred orientation so that the wafers which are sliced from the crystal perpendicular to the growth axis have surfaces parallel to {100} crystallographic planes. Prior to slicing these ingots into wafers, the ingots are generally subject to a centreless grinding operation to remove the slight fluctuations in diameter along the length of the ingot that occur during crystal growth.

M. A. Green

Figure 4.9

(a) Czochralski (CZ) growth; (b) squared-off CZ ingot. Source: Green and Hansen (1998).

Alternatively, the ingots can be "squared-off by sawing off large sections parallel to the growth axis (Fig. 4.9b), giving "quasi-square" wafers after wafering. The large pieces of silicon sawn off in this approach are then generally recycled by re-melting as feedstock for the CZ growth.

Crystalline Silicon Solar Cells

Figure 4.10

167

(a) Inner diameter wafer sawing; (b) continuous wire sawing. After Dietl et al. (1981).

The technique traditionally used in microelectronics for sawing wafers from ingots has been based on the use of inner diameter saws. In this technique, shown Fig. 4.10a, thin metal sheet blades are given dimensional solidity by being held in tension around their outer perimeter. The cutting surface is a diamond-impregnated region surrounding a hole within the tensioned metal sheet. This technique gives excellent dimensional tolerance, although there are limitations arising from the thickness of the silicon wafers that is possible to produce while still maintaining high yield. Other limitations arise from the wastage of silicon as "kerf loss during cutting. Generally, about 10-15 wafers per centimetre of ingot length are achieved by this process. An alternative technique increasingly being used in photovoltaics is based on wire sawing (Fig. 4.10b). In this case, tensioned wire is used to guide an abrasive slurry through the ingot. Advantages are thinner wafers and less surface damage for these wafers as well as lower kerf or cutting loss, allowing the sawing of over 20 wafers per centimetre.

168

M. A. Green

In the 1970s and early 1980s, several other options for preparing silicon feedstock were investigated as part of a large US government PV program encouraged by the Carter administration (Christensen, 1985). A great diversity of alternative routes to producing pure silicon were investigated. These ranged from those involving radically different approaches to those exploring only minor changes from the sequence outlined above, such as the use of different compounds of silicon as the intermediate during the purification process. One such process, based on the use of silane as the intermediate (Christensen, 1985), is now being used commercially, although the product is used exclusively for the microelectronics industry. Parallel development has been conducted outside this program, notably in Germany (Aulich, 1996) and Japan. In Japan, Kawasaki Steel have been investigating an alternative route to preparing cheap silicon feedstock from metallurgical grade precursors and were scheduled to begin pilot production with this sequence in 1998 (Sakaguchi et al., 1997). To produce ingots from the pure silicon feedstock, a modification of the CZ process which produces "tricrystalline" silicon has also been used for photovoltaics (Endros et al, 1997). Wafers cut from the crystals have a different {111} equivalent orientation for each third of their surface, with the differently orientated regions separated by a twinning plane. Claimed advantages of higher growth rates and greater mechanical strength are probably not large enough to offset disadvantages of not being able to chemically texture such wafers and the poor electronic quality near the twinning planes. Another alternative to the standard Czochralski process for producing crystalline ingots is the floatzone (FZ) process. Although some studies have predicted superior economics when compared with the Czochralski process for cell production due to the elimination of consumables such as quartz crucibles, the FZ process, as commercially implemented, is capable of accepting feedstocks only in the form of high quality cylindrical rods. This makes it unsuitable for using low-cost off-grade material. However, the casting and directional solidification processes used to produce multicrystalline silicon are generally extremely tolerant of poor quality feedstock material. These techniques will be discussed in more detail in the following section.

4.3.2 Multicrystalline silicon ingots In 1998, about 30% of the world's photovoltaic production was based on multicrystalline silicon wafers. Several companies have developed commercial processes for producing the precursor multicrystalline silicon ingots (Ferrazza, 1996). Advantages over the Czochralski process are lower capital costs, higher throughput and a higher tolerance to poor feedstock quality.

Crystalline Silicon Solar Cells

169

(b)

Figure 4.11 (a) Directional solidification of silicon within a mould; (b) sawing of large ingot into smaller sub-sections. Source: Green and Hansen (1998). The technique involves controllably solidifying molten silicon in a suitable

container to give silicon ingots with large columnar grains generally growing from the bottom of the crucible upwards (Fig. 4.1 la). Pioneers with this approach for modern photovoltaics in the mid-1970s were Wacker Chemitronic of Germany (Authier, 1978) and Solarex of the USA (Lindmayer, 1976). In the 1980s, other manufacturers including Eurosolare/Crystallox, Kyocera, Bayer, Crystal Systems and Sumitomo Sitex had developed processes capable of producing good quality multicrystalline material. Techniques differ between these manufacturers in the choice of crucible material, the method of loading the crucible with silicon and the method for controlling the cooling of the melt. A good summary can be found elsewhere

170

M. A. Green

(Ferrazza, 1996). The size of a nominally rectilinear ingot can be very large, up to 60 cm x 60 cm x 20 cm, and these ingots can weigh several hundred kilograms (Khattak and Schmid, 1997). The large ingots are sawn into smaller sections as shown in Fig. 4.11b, eventually to give wafers generally 10-15 cm along the sides. These smaller sections can be sawn by the standard inner-diameter or continuous wire sawing processes. The resulting multicrystalline wafers are capable of producing cells of about 80% of the performance of a monocrystalline cell fabricated on a CZ wafer. However, because of the higher packing density possible due to their square or rectangular geometry, this performance difference is largely masked at the module level with multicrystalline module performance lying in the range demonstrated by modules made from monocrystalline cells. An interesting variation on this approach is the continuous casting process such as developed by Sumitomo Sitex. In this case, electromagnetic fields are used to constrain the molten silicon to produce essentially a continuous ingot of multicrystalline silicon (Sarti et al., 1997).

4.3.3 Sheet and ribbon silicon Although there is the potential for substantial cost reductions in both the cost of preparing the silicon feedstock and in forming crystalline or multicrystalline ingots from it, one unavoidable cost with the silicon wafer approach is the cost of sawing the ingot into wafers. Several studies have suggested that the lower bound on this cost may be something of the order of US$0.20/watt (Christensen, 1985; Bruton et al, 1997). This has provided the rationale for investigating approaches that produce silicon directly in the form of self-supporting sheets without the need for sawing (Bergin, 1980; Shulz and Sirtl, 1984). Commercially, the most advanced sheet or ribbon approach is based on the edgedefined film-fed growth (EFG) technique of Fig. 4.12. As originally developed in the early 1970s, this involved the pulling of a thin sheet of silicon ribbon from a strip of molten silicon formed by capillary action at the top of a graphite dye (Fig. 4.12a). Substantially higher throughput was obtained with the more symmetrical configuration shown in Fig. 4.12b, where the ribbon is pulled in the form of a hollow nonagon. Individual wafers are then cut from the sides of the nonagon, normally by laser scribing wafers from each of the sides. The material produced is multicrystalline with elongated grains and of a similar quality to the standard directionally solidified multicrystalline material. Commercial cells made from EFG material have been available sporadically since the early 1980s with a large 25 MW/yr facility recently announced.

171

Crystalline Silicon Solar Cells

iyciystalline ribbon

molten silicon macs by capilary action

carbcnde

A

A

/^L ;en silicon

(a)

(b)

Figure 4.12 (a) Edge-defined, film-fed growth (EFG) method; (b) growth of a nonagonal ribbon of silicon using the EFG method. Source: Green and Hansen (1998).

172

M. A. Green

An even older ribbon growth process is the dendritic web approach of Fig. 4.13 first described by Westinghouse in the 1960s. In this approach, close thermal control is used to cause two dendrites spaced several centimetres from each other to solidify first during the growth step. When these are drawn from the melt, a thin sheet of molten silicon is trapped between them. This quickly solidifies to form a ribbon. After a substantial research and demonstration program by Westinghouse in the 1970s and early 1980s, this approach is now under development by Ebara Solar (Narashima et ai, 1997). A somewhat related approach is the string ribbon approach. In this case, the molten silicon is trapped between two graphite strings that are drawn from the melt. This relaxes the requirement on thermal control, compared with the previous dendritic web approach. The string ribbon approach is under development by Evergreen Solar (Janoch etal, 1997'; Wallace etai, 1997). Another interesting approach that was developed in the 1980s relied on direct casting of silicon wafers using a centrifugal casting approach to overcome surface tension problems within the closely spaced faces of a horizontally aligned graphite mould (Maeda and Hide, 1987). Despite initially promising results, this approach appears to be no longer under active development. A compact but thorough review of most of the above ribbon processes including references is given elsewhere (Shulz and Sirtl, 1984). A thorough bibliography of work prior to 1980 has also been published (Bergin, 1980). Somewhat related to the above ribbon approaches are other sheet approaches which produce silicon films on substrates from which they are subsequently detached. The most developed version of this technology is the VEST technology developed by Mitsubishi (Hamamoto et ai, 1997). In this approach, a potentially reusable silicon substrate is oxidised, then vias are etched in the oxide and then a silicon film is deposited on top of the oxide. This film is subsequently laser-recrystallised. The thickness of this seeding layer is then increased by the subsequent high-temperature epitaxial growth of the silicon layer. After reaching a target thickness of 50-80 |im, the film is detached from the substrate which is then potentially reusable. Promising efficiencies over 16% have been obtained from this approach for substrates that are only 60-70 |xm in thickness, but are still self-supporting. Other researchers have suggested similar techniques to produce even thinner films. Some have suggested the use of a silicon wafer treated to produce a layer of porous silicon along the surface as the source of the crystallographic template for the subsequent growth of a silicon layer, which is then detached (Wenham and Green, 1995; Brendel, 1997). If this detached layer is too thin to be self-supporting, it could be transferred to structurally strong components such as the glass layer in a structural superstrate design. Another

Crystalline Silicon Solar Cells

173

variant involves forming vias through the oxide to a {100) orientated substrate and the subsequent use of liquid phase epitaxy to grow a mesh of silicon on the substrate which again is detached after processing (Weber et al., 1997). In this case, layers of about 70 |Xm thickness are envisaged, although cell processing on the unusual geometries that result would pose obvious challenges. silicon dendrites or carbon string

Figure 4.13 Schematic illustrating either the dendritic web growth process or the string ribbon approach. Source: Green and Hansen (1998).

4.4 Cell processing 4.4.1 Standard process In the previous section, standard and non-standard ways of forming the silicon substrate were described. The major commercial substrates are those formed by the wafering of monocrystalline and multicrystalline ingots, with a much smaller quantity of EFG ribbon substrates produced commercially. Since monocrystalline silicon wafers are the norm, processing for these will be described first with the processing of other types treated as variations on the monocrystalline processing approach.

174

M. A. Green

At present, no photovoltaic manufacturer prepares polysilicon source material. Manufacturers generally purchase off-specification material from the microelectronics industry or, alternatively, bypass the crystal growth step by purchasing silicon wafers. Processing starts by chemically cleaning the starting wafers and etching their surfaces, generally in a sodium hydroxide etchant, to remove saw damage from the wafers. For monocrystalline wafers, the next step is crystallographic texturing, again using sodium hydroxide but in a more dilute solution. The composition and temperature of this solution determines the texturing quality (King and Buck, 1991), including the size of the pyramidal features resulting from the texturing and the percentage of wafer surface area successfully covered by such features. Texturing is a demanding step in cell processing and the quality of texturing varies enormously between different manufacturers. Cell performance, however, is not critically dependent on texturing quality. The next major stage of processing is the diffusion of the cell junction. This is generally achieved by spraying or spinning a compound containing phosphorus onto the cell surface, followed by heating at high temperature to allow phosphorus dopant atoms to seep into the cell surface by thermal diffusion. Typically, the depth of diffusion is less than 1 jum. The same thermal diffusion process is widely used in microelectronics but processing for photovoltaics generally involves cruder equipment and techniques, since the aim is to produce cells at the lowest possible cost without unduly sacrificing cell performance. Although the diffusion is required over only one surface of the wafer and processing techniques are generally chosen to encourage such a result, phosphorus invariably seeps into both wafer surfaces to some extent. To break the connection between the phosphorus diffused into front and rear surfaces, an 'edge junction isolation' step is required to remove the thin phosphorus layer around the edge of the wafer. This isolation is often achieved by 'coin stacking' the wafers so that only their edges are exposed and then placing the stack in a plasma etcher to remove a small section of silicon from the wafer edge, hence breaking the conductive link between front and rear surfaces. The screen printing of metal contacts onto the front and rear surfaces completes cell processing. Silver paste consisting of a suspension of fine particles of silver and glass frit in an organic medium together with appropriate binders (Hoernstra et al, 1997) is squeezed through a patterned screening mesh onto the cell surface. After application, the paste is dried at low temperature and then fired at a higher temperature to drive off the remaining organics and to allow the silver regions to coalesce. The glass frit is important in promoting adhesion to the silicon substrate. Often pastes are doped with phosphorus to help prevent the screened contact from penetrating the thin phosphorus skin that it is intended to contact.

Crystalline Silicon Solar Cells

175

The paste for the top surface is printed in a characteristic finger pattern to minimise the resistive losses in the cell while allowing as much light as possible into it. Sometimes the rear contact is patterned, not to allow light into the cell, but merely to reduce the amount of paste required and hence reduce the cost of this processing step. Sometimes small quantities of aluminium are added to the paste used on the rear surface to give a heavily doped p-type 'back-surface field' region underlying the rear metal contact or, alternatively, separate screening and firing of an Al paste over the entire rear surface is used to more fully optimise this feature (Nijs et ah, 1996). This screen-printing method for applying the metal contact was borrowed in the early 1970s from the hybrid microelectronics industry (Ralph, 1975). This ensured the ready availability of both screen-printing equipment and the paste-firing furnaces suited to this application. Labour and equipment costs associated with this step tend to be very low. However, the pastes themselves can be expensive and an even larger cost penalty is paid for the simplicity of this approach by the forfeiture of the inherently available power output from the silicon wafer, as discussed later. A quarter wave antireflection coating can be applied to the cell at this stage. Generally, titanium dioxide is used as the antireflection coating material due to the simplicity of depositing this compound and its almost ideal refractive index for this application. Some manufacturers deposit the antireflection coating before the metal paste-firing step and fire the paste through this coating. The cells are then ready for testing under a solar simulator. Cells are usually graded based on their short-circuit current or current at a nominal operating voltage, e.g., 450 mV. Generally, cells are sorted into 5% performance bins. This sorting is required to reduce the amount of mismatch within the completed module. To a large extent, the output current of the module is determined by that of the worst cell in the module, resulting in large power losses within mismatched modules. Even worse, low output cells can become reverse-biased under some modes of module operation and destroy the module by localised over-heating. Very similar processing is applied to multicrystalline silicon wafers. In this case, most of the grains will have incorrect orientation for effective texturing by anisotropic etching, although such texturing is sometimes used for the relatively small benefit that can be obtained. However, a quarter-wave interference antireflection coating has been mandatory for good performance from multicrystalline materials. One disadvantage of anisotropic texturing of these materials is that different grains etch at different rates, giving a very uneven surface due to steps at grain boundaries. This can pose hazards for continuity of the subsequently screened metal lines. Accordingly, some manufacturers prefer to etch multicrystalline silicon with an isotropic etch to maintain a smooth surface.

176

M. A. Green

The rippled surface that is a natural consequence of the EFG ribbon growth process poses similar continuity hazards for screen-printed metallisation. To accommodate this rough surface, a novel technique has been developed whereby the metal paste is squeezed through an orifice and then drops to the cell surface, much the same as squeezing toothpaste from its tube onto a toothbrush.

4.4.2 Limitations of the screen-printing approach There are four main limitations arising from the screen-printing approach to applying the front contact which cause the simplicity in processing to be at the expense of cell performance. As noted above, performance can be reduced well below that inherently achievable. One limitation is that the phosphorus diffusion has to be heavier than desirable purely from the point of view of cell performance, to allow reliable low resistance contact between the screen-printed metal and the diffusion. Typically, sheet resistivities of this diffusion less than 60 ohms/square are required (Green, 1995; De Clercq et al., 1997). Such diffusions generally reduce the quality of the silicon in the region near the cell surface where blue wavelengths in sunlight are strongly absorbed. A screen-printed cell does not therefore respond well to blue wavelengths in sunlight, wasting at least 10% of the possible current output through this deficiency. The remaining three limitations relate to the geometry and conductivity of the metal lines it is possible to produce by the standard screen-printing process. Since the paste thickness shrinks to about one-third of its original thickness during firing (silver constitutes only 25-30% by volume of the original paste, with up to 5% glass frit), it is very difficult to achieve metal lines with high aspect ratio (height/width). High aspect ratios are the key to designing metal grids which result in low overall losses (Serreze, 1978). The nature of the screening meshes that have sufficient ruggedness for use in commercial production means it is very difficult to achieve fine lines using screen printing in production. Typically, 150 \im is the minimum width that can be cost-effectively achieved. This limitation means that there will generally be high shading losses in screen-printed cells due to the large percentage (10-15%) coverage of the front surface by the metal. Additionally, the relatively poor conductivity of the fired silver paste—about 2 times lower than that of pure silver for large features such as busbars but up to 6 times lower for finer features such as fingers (de Moor et al, 1997)—fundamentally limits ability to optimise metal contact design, in much the same way as does the low aspect ratio previously discussed. Recent work describes improved laboratory cell performance based on experimental screens formed by cutting patterns in thin metal sheets using a laser (Nijs et al.,

Crystalline Silicon Solar Cells

177

1996). This approach is reported to allow reduced linewidths, although the authors may be overly optimistic about the potential in a production setting (de Moor et al., 1997). With a standard sequence under laboratory conditions, cell efficiency is limited to less than 15% even with these improved linewidths (Nijs et al, 1996). In a more complex sequence involving inherently costly steps such as the deposition of a plasma nitride antireflection coating and a complicated rear Al screening and removal step, efficiency approaching 17% has been confirmed. Some caution is required in accepting the authors' enthusiasm as to how transferable these sequences may be to a production setting. The same authors (Nijs et al, 1996) also analyse cost and performance relative to established commercial sequences such as the buried-contact sequence described below but these analyses appear to be skewed by overly optimistic assumptions about screen-printing metallisation parameters.

4.4.3 Buried-contact solar cells As mentioned in Section 4.2, buried-contact cells were developed as a way of incorporating some of the efficiency improvements demonstrated in the mid-1980s into low-cost commercial cell production sequences. This aim has been successfully realised with recent independent costing studies showing that the buried-contact cell not only produces the highest commercial silicon cell efficiency, but also the lowest cost approach for fabricating commercial silicon cells, of any of those at any reasonable state of development (Bruton et al, 1997). The processing of buried-contact cells begins similarly to that outlined for screenprinted cells. In the commercially most successful buried-contact sequence (Jordan and Nagle, 1994), the incoming wafers are cleaned and textured as with conventional wafers and then diffused. A silicon nitride antireflection layer is grown by chemical vapour deposition over the entire top surface of the cell. Grooves are next formed in this surface through the antireflection coating and prior diffusion. A standard neodynium YAG laser readily produces grooves of about 20 /xm width. The depth depends on the laser power, but desirably lies in the range 20-60 /m\. After etching to clean the grooves, a second diffusion, which is restricted by the nitride to the regions that have been laser-grooved, is performed. Aluminium is then evaporated onto the rear of the wafer and sintered. Electrolessly plated nickel followed by similarly applied copper and silver is then deposited. Again, the insulating nitride restricts the plating to the grooved areas and to the rear of the wafer which has already been metallised by aluminium.

178

M. A. Green

When applied to the same commercial silicon wafers as used in the screen-printing process, cell efficiencies in the 17-18% range are obtained. Figure 4.14 compares the performance of a screen-printed and a buried-contact cell fabricated on the same quality starting material (BP Solar, 1991). A performance advantage of 20-30% is demonstrated by the buried-contact approach, largely as a result of improved shortcircuit current density but with other significant contributions coming from improved fill factor and open-circuit voltage. In addition to this performance advantage under standard test conditions, field studies have shown that buried-contact cells give up to 15% more energy per rated watt as a result of an even larger performance margin at low light intensities and under the bluer light associated with cloudy conditions (Mason etal., 1997). 4

V

^ \

tn

I,. c

I

o

1 •]

^ ^ ^ » Buried Conlacl _ _ _ . Screen-Printed

\ \ \ \ \ \ \ \ \ \ \ \ \

0 . 0.0

0.2

0.4

0.6

Voltage, Volts

Figure 4.14 Output characteristics of buried-contact cells compared with screen-printed cells. After BP Solar (1991).

4.5 Cell costs There have been many studies of the costs of the different stages of silicon cell production using different basic assumptions, particularly in relation to the production volume assumed in the study and the cost of polysilicon source material. Probably the most recent and most authoritative is one conducted under the auspices of the European Union Photovoltaic Program (Bruton et al., 1997). This study involved representatives of seven major European photovoltaic manufacturers and research laboratories, and is valuable for the breadth of representation and the diversity of approaches explored. Since the groups involved are known for their strong views on the virtues of the different sequences studied, the study also involves an undoubtedly

179

Crystalline Silicon Solar Cells

hard-won political and technical consensus of those intimately involved with these issues. The key assumptions of the study were manufacturing volume of 500 MWp of solar cells per annum and the availability of silicon source material at US$25 per kg. A number of different technologies were compared. Important comparisons were between EFG ribbon, multicrystalline and crystalline wafer technologies, between screen-printed, buried-contact, metal-insulator-semiconductor and PERL cell processing sequences, in various combinations of wafers and processing, and between two different module encapsulation approaches. However, the results from all possible combinations were not studied (or, at least, not published), but only the seven selected combinations shown in Table 4.1. Table 4.1 Summary of published results of a European Commission study of manufacturing costs for 500 MWp per year factory ID

Wafer8

Process

Cell efficiency* study (present)

#1 #2 #3 #4 #5 #6 #7

DS CZ CZ CZ CZ CZ EFG

SP SP LGBC MIS/A MIS/B PERL SP

15% (12.6-14.8%) 16% (13.9-15.6%) 18% (16.5-17.5%) 17% (N/A) 17% (12.2%) 20% (N/A) 14.4% (12%)

Estimated cost (ECU/Wp)' 0.91 1.25 1.15 1.28 1.34 1.78 0.71

Key variable Wafer Wafer/process Process Process/module Module Process Wafer

°DS: directional solidification; CZ: Czochralski growth; EFG: edge-defined film-fed growth; SP: screenprinted; LGBC: laser grooved, buried-contact; MIS/A: metal-insulator-semiconductor; MIS/B: as for MIS/A but with resin-fill packaging; PERL: passivated emitter, rear locally diffused (less appropriate acronym LBSF used in study). 'The cell efficiencies assumed in the study in some cases differ appreciably from present average production values, deduced by the present author from manufacturers' data sheets or the results from large field installations. C1ECU « US$1.2. Source: Bruton etal., 1997.

Several key results can be deduced from this Table. When comparing screenprinted cells on ribbon (EFG), multicrystalline (DS) and monocrystalline (CZ) wafers, the ribbon produces the lowest cost of 0.71 ECU/WP followed by the multicrystalline wafers at ECU0.91/Wp and the monocrystalline wafers at ECU1.25/Wp. The advantage of the ribbon stems almost entirely from the fact that it does not need to be sawn, as previously mentioned.

180

M. A. Green

Comparing between the different processing approaches on single crystal wafers, the cheapest is the buried-contact at 1.15 ECU/Wp, followed by the screen-printed at 1.25 ECUAVp, followed by the metal-insulator-semiconductor at 1.28 ECUAVp, followed by the PERL at 1.78 ECUAVp. The buried-contact achieves its cost advantage over the screen-printing approach by virtue of the increased efficiency giving more power per unit processing area. Such advantages would transfer to the less expensive substrate approaches studied, suggesting the best possible combination of wafer, process and moduling approach would result in manufacturing costs well below 0.70 ECUAVp. In the module area, the standard laminated module approach is calculated to be slightly cheaper than an alternative resin-fill approach. Compared to the predictions of this study, present manufacturers fabricate screen-printed monocrystalline and multicrystalline cells and buried-contact monocrystalline cells in roughly 10-20 MWp per year production capacities with large-volume selling prices of modules in 1999 of about US$4AVp (a similar amount in ECUAVp). Present encapsulated cell efficiencies, deduced by the present author mainly from manufacturers' data sheets or from recent field performance, are also shown in Table 4.1, indicating the various levels of extrapolation in cell performance assumed for the different cell technologies in the study.

4.6 Opportunities for improvement 4.6.1 Commercial cells The large differential between the efficiencies of a typical screen-printed commercial cell (15%) and the best laboratory silicon cell (24%) shows the enormous potential for further efficiency improvement in commercial devices. Part of this potential has been recently realised with the commercialisation of the buried-contact cell with cell efficiencies in the 17-18% range obtained in production. There remains scope for a further substantial performance improvement. One reason for the difference between laboratory and the best commercial cells is the difference between the CZ wafers used in commercial production and the FZ wafers used for the best laboratory cells. CZ grown wafers are invariably contaminated with oxygen and carbon during growth to a much higher level than FZ wafers, due to use of quartz crucibles and graphite heaters in the CZ process. These impurities give rise to a much more subtle dependence on processing conditions, in the CZ material, of an important silicon material property for producing high performance cells, the minority carrier diffusion length. For example, applying the

Crystalline Silicon Solar Cells

181

high temperature processing associated with PERL type sequences to CZ silicon gives a large spread in results depending on the supplier of CZ material and hence most probably on the oxygen and carbon content (Wettling et al, 1996). Additionally, the quality of CZ material as compared with FZ falls off quite rapidly as the boron content is increased, possibly because boron/oxygen complexes form within the material. This reduces flexibility in cell design since it eliminates the possibility of using low resistivity CZ substrates. In microelectronics, high oxygen content resulting from the CZ process is regarded as an asset. Oxygen increases the mechanical strength of the wafers as well as allowing gettering of surface regions, where the operational microelectronic devices are confined, by the precipitation of oxygen defects away from the wafer surfaces. A simple option for improving the suitability of CZ material for photovoltaics may be merely to change the crucible material used in the CZ process. For example, experiments have been conducted with silicon nitride coated crucibles as a way of reducing oxygen content within the material while increasing that of nitrogen (Shimura, 1989). However, relatively little exploration of such possibilities has been undertaken, probably because the benefits would not, in any case, be seen with the standard screen-printing approach. More sophisticated cell processing sequences would be required (such as offered by the buried-contact approach) to obtain the full benefits from such improved temperature tolerance. As opposed to the case of CZ silicon, much experimentation has been conducted with directionally solidified multicrystalline silicon. It may well be that multicrystalline silicon eventually exceeds the standard CZ material in its performance potential for photovoltaics due to the eventually better high temperature tolerance of the material. For example, the recent demonstration of 19.8% efficiency upon multicrystalline silicon (Zhao et al, 1998) puts the performance of this material right in the middle of the performance range observed with a similar sequence using CZ wafers (Wettling et al, 1996). The trend towards thinner cells that arises primarily from efforts to reduce the costs of the silicon wafer may actually help to improve the cell efficiency. Thin wafers give the opportunity for back-surface fields or other rear-surface passivation approaches to be used to improve cell performance, primarily through increased voltage output. Again, the buried-contact processing sequence would be capable of realising such potential performance advantages due to its high open-circuit potential, which is largely untapped in wafers of standard thickness. Bifacial cell designs offer another way of effectively improving cell efficiency. Recent studies suggest that module output can be improved by approximately 20% in standard open back configuration without any special effort if use can be made of light scattered onto the rear of the module (Chieng and Green, 1993). However, as cells are

182

M. A. Green

now being increasingly used in the residential market where rear illumination of the module is unlikely, only part of the market would benefit from this improvement. Japanese groups in particular are showing increasing interest in amorphous and microcrystalline silicon-based surface passivations as demonstrated by the 'HIT' cell structure (Tanaka, et al., 1993; Sawada et al., 1994), as shown in Fig. 4.15. This cell structure has demonstrated an open-circuit voltage capability similar to that of the buried-contact approach. However, it is inherently incapable of giving a similar current output due to light absorption in the 'transparent' conducting oxide layer required to give lateral conductivity to the amorphous silicon emitter as well as the less than 100% collection efficiency from the latter region.

Figure 4.15 HIT (Heterojunction with Inlrinsic Thin Layer) cell on textured crystalline silicon substrate. After Green and Hansen (1998).

Given better feedstock material or cells below 150 u.m in thickness, improved rearsurface passivation approaches such as demonstrated by the PERC and PERL cells of Fig. 4.7 could become appropriate. A promising start has been made with the double sided buried-contact cell which applies high quality oxide passivation to both top and rear surfaces (Green, 1995). Other options may be rear passivation layers based on amorphous, microcrystalline or polycrystalline silicon (Okamoto et al., 1997), or on specially deposited silicon nitride (Aberle et al., 1997).

Crystalline Silicon Solar Cells

183

4.6.2 Laboratory cells For laboratory cells, an appropriate reference point for performance is the AM 1.5 detailed-balance efficiency limit of 33% for material of the band gap of silicon. However, it has been shown that another intrinsic process, Auger recombination, provides a more severe fundamental limit for silicon than the radiative recombination processes assumed in the detailed-balance limit (Green, 1984; Tiedje et al., 1984). Unlike the detailed-balance limit, the Auger limit for a silicon cell is dependent on the cell thickness, as shown in Fig. 4.8 (Green, 1995). This difference arises because the detailed-balance calculation includes photon recycling which makes nett recombination rates independent of cell volume. With lambertian light trapping, the optimum cell performance in the Auger limit is 29% for a cell of about 80 |im thickness. Such a cell would have an open-circuit voltage of about 760 mV, higher than the highest value ever demonstrated for silicon of 720 mV. The voltages of these best performing experimental devices demonstrating 720 mV were limited by surface recombination rather than bulk recombination. Figure 4.16 shows the results of efficiency calculations with various amounts of surface recombination added, characterised in terms of the open-circuit voltage limit that this recombination would impose if it were the only recombination process in the cell. Increasing surface recombination reduces the value of the obtainable efficiency as well as pushing the optimum cell thickness to larger values.

1

10

100 Thickness (um)

1000

10000

Figure 4.16 Limiting efficiency of silicon cell with lambertian light trapping as a function of surface recombination velocity, characterised in terms of the voltage limit imposed by this recombination. Source: Green (1999).

184

M. A. Green

Figure 4.16 makes it clear that, to improve silicon cell efficiency much beyond 26%, improved surface passivation (both surfaces) is essential beyond the 720 mV capability presently demonstrated. If this improved quality cannot be achieved, an alternative possibility is to maintain the same quality of surface passivation as presently demonstrated and reduce the effective threshold energy of the photovoltaic process within the bulk regions of the cell. Techniques such as alloying sections of the cell with germanium to reduce its band gap (Healy and Green, 1992) or doping with a photoactive impurity to give impurity photovoltaic effects in the bulk region (Keevers and Green, 1994) have been suggested and shown, in some cases, to have theoretical advantages. However, no experimental performance advantage has been demonstrated by either technique to date. A well-proven approach for improving solar cell efficiency is the use of the tandem cell structure. Efforts to produce tandem cells with silicon have not yet given good results due to the inability to find a suitable wide band-gap partner that is latticematched to silicon (Corkish, 1991). For low quality cells, amorphous silicon/ polycrystalline silicon tandems have given improved results over either cell type alone (Yamamoto et al, 1997; Shah et al, 1997). As the quality of the polycrystalline lower cell in this combination improves, however, it is doubted that this situation will continue (see Section 4.7). A higher performance top cell will eventually be required, which could be provided by a crystalline compound cell if difficulties with lattice mismatch to the silicon substrate can be overcome. Fuller use of the available photon energy by incorporating efficient impact ionisation process has been suggested as a way of boosting cell performance by generating more than one electron-hole pair from one high energy photon (Kolodinski et al, 1993; Werner et al, 1994). However, such processes are quite weak in silicon with increases in current density limited to less than 0.1mA cm-2 (Green, 1987). Manipulating the details of the band gap of silicon, for example by alloying with germanium; may improve prospects. However, since the high-energy photons of most interest for this process are absorbed very close to the surface of silicon, such approaches may interfere with the ability to obtain well-passivated surfaces. Limited experimental work with shallow germanium implants has not given any nett performance benefit (Keevers et al, 1996). More advanced concepts such as the multiple quantum wells discussed in Ch. 10 might also be appropriate (Barnham and Duggan, 1990). A recent area of interest has been the use of ZnS/Si multiple quantum wells. Not only is there a good lattice match between ZnS and Si, but recent studies suggest is may be easier to obtain direct bandgap-like properties from such multiple quantum wells than the Si/Ge alternative which has been the focus of most past study. There is still some question as to whether or not

Crystalline Silicon Solar Cells

185

a multiple quantum well device offers a performance advantage above the detailedbalance limit (Araujo et al., 1994), but this approach almost certainly offers performance advantage above the Auger limit for silicon, in principle.

4.7 Silicon-supported thin films There has long been an interest in transferring the strengths demonstrated by crystalline silicon wafer technology to cells based on silicon thin films. Historically, work can be divided into two phases: (i) that before the 1980s when the benefits of light trapping were not fully appreciated; and (ii) that after the mid-1980s where light trapping has been regarded as an essential feature of any silicon thin-film cell design. The early work laboured under what is now known to be a misconception that quite thick layers (>20 |xm) of silicon would be required to give reasonable performance due to silicon's poor absorption characteristics arising from its indirect band gap (see Fig. 4.8). However, since light trapping can increase the effective optical thickness of a silicon cell by 10-50 times, this means that layers of only 1 jjm or so thickness are still inherently capable of producing similar performance to much thicker layers. Approaches to producing supported silicon films can be divided into hightemperature and low-temperature strategies depending on whether or not the substrate is heated to high temperature during the silicon deposition or subsequent processing.

4.7.1 High-temperature supported films One of the earliest silicon supported film approaches was the 'silicon-on-ceramic' approach (Christensen, 1985) whereby a ribbon of ceramic material was dipped into a molten silicon bath or pulled across the surface of a silicon melt so that one side was coated with silicon. This produced silicon of modest quality and the approach suffered from difficulties in making rear contact to the cells, since the ceramics used were insulating. This approach was discontinued in the early 1980s. Early work by Ting and Shirley Chu involved the deposition of silicon onto a range of foreign substrates by high temperature chemical vapour deposition (Chu, 1977). Operational cells were obtained using a number of substrate materials. The best results were obtained by depositing the silicon layers on multicrystalline silicon substrates prepared from metallurgical grade silicon. Given the previous studies that have shown that sawing of wafers represents one of the major costs in any wafer-type approach, the overall economics of such an approach using a wafer substrate are questionable, regardless of

186

M. A. Green

the quality of this substrate. Other early work involved the deposition of silicon onto ceramic substrates by high temperature CVD and the subsequent increase in crystal size by melting and directional solidification (Minagawa et al., 1976). In the post-1980 era, efforts in silicon supported film were revitalised by the US company AstroPower (Barnett et al, 1985). In this company's approach, silicon is deposited onto a conducting ceramic substrate by a technique which has not been disclosed but which has been reported to involve the deposition of silicon from solution in molten metal as one step (Barnett et al, 1989). The resulting ceramicbased wafers can then be processed in the same way as a multicrystalline silicon wafer. Efficiency up to 16.6% has been confirmed for films that are apparently in the 50-100 [im thickness range (Bai et al, 1997). Efforts are underway to include more effective light trapping into these devices and to produce an integrated module upon insulating ceramic (Ford et al., 1997). More recently, promising laboratory results have also been obtained by a German collaborative effort using much thinner films. These films were formed by first depositing and recrystallising a thin silicon layer upon a silicon carbide coated graphite substrate followed by the deposition of an epitaxial layer of silicon of about 30 ixm thickness upon this recrystallised layer. Cell efficiency above 11 % has been confirmed for a cell of this 30 ^m thickness (Ltidemann et al, 1997).

4.7.2 Low-temperature approaches One of the first papers addressing silicon photovoltaic thin films described the deposition of silicon by low temperature chemical vapour deposition onto an aluminium substrate (Fang et al., 1974). A surprisingly large grain size was obtained, attributed to eutectic reaction with the aluminium. In more recent times, laser crystallisation has been used in the active matrix liquid crystal display industry to produce relatively small-grain polycrystalline silicon films from amorphous silicon precursors, generally deposited by low-pressure chemical vapour deposition. Grain sizes are typically less than a micron or so, so that these films would probably not be suitable for photovoltaics. Also, thicknesses for the active matrix display industry tend to be only about 100 nm, which would be too thin for photovoltaic application. From 1989, a group at Sanyo explored the use of low-temperature solid-phase crystallisation of amorphous silicon as a technique for producing thin-film polycrystalline silicon cells. Good results have been obtained with 9.2% (unconfirmed) efficiency reported in 1995 (Baba et al., 1995). These cells were approximately 1 cm2 in area deposited onto a textured metallic substrate and heated at

Crystalline Silicon Solar Cells

187

approximately 600 C for many hours to enable the crystallisation of the originally amorphous films. After crystallisation, the HIT structure developed by Sanyo is used to complete the cell processing at low temperature. Two groups have reported good results using silicon thin films deposited directly in microcrystalline form onto glass substrates. The University of Neuchatel has reported unconfirmed efficiencies of about 7% for 3 /zm thick microcrystalline cells deposited at 500 C (Shah et al., 1997). The cell has a p-i-n structure with the intrinsic region comprising most of the device thickness. The cell is designed for this intrinsic region to be depleted during normal device operation to create a high electric field to aid carrier collection, as with a standard amorphous silicon cell. Finally, Kaneka Corporation (Yamamoto et al., 1997) has reported efficiencies over 10% with a similar device structure, shown in Fig. 4.17. Nearly the same efficiency was obtained when the total device thickness was varied over the 1.5-3.5 /urn range. Both the above groups have reported even higher efficiencies when amorphous silicon cells are used in a tandem configuration on top of the microcrystalline device. Given the relatively small amount of effort so far dedicated to this area, these results are extremely encouraging, and show the enormous potential of such low temperature approaches.

/ poly-S[

Figure 4.17 Structure of 9.4% efficient thin-film microcrystalline solar cell developed by Kaneka. Cell thickness is typically 1-3 //m. After Yamamoto el al. (1997).

188

M. A. Green

4.7.3 Multilayer solar cells The parallel multilayer solar cell shown in Fig. 4.18 provides an effective cell design when dealing with low quality silicon such as obtained by low temperature approaches (Green and Wenham, 1994). This cell differs from a tandem cell because all junctions are connected in parallel. The active region of any cell is the depletion region straddling the metallurgical junction and the region within a diffusion length on either side of this depletion region. The challenge in producing a thin supported silicon device of high performance is therefore in having material quality sufficiently good for this active volume to be wide enough to result in a large amount of current collection. In the microcrystailine work previously reported, attempts have been made to enlarge this active region by expanding the junction region by making this region as lightly doped as possible. The multilayer approach provides an alternative (or complementary) way of achieving the same result. By having multiple p-n junctions dispersed throughout the material, it is possible to make the whole volume of material electronically active regardless of material quality. The approach is particularly appropriate when the parallel layers are very heavily doped allowing unique thin-film cell and module designs where the lateral conductivities of the doped layers are sufficiently high to allow lateral current flow without appreciable resistance loss. This removes the need for transparent conducting oxides used in other thin-film technologies to provide this lateral conductance.

Figure 4.18 Parallel multilayer cell schematic. The red and while layers correspond to different doping polarities. Each layer is thinner than the minority carrier collection distance. Source: Green and Hansen (1998).

Crystalline Silicon Solar Cells

189

Figure 4.19 Fabrication of UNSW multilayer cells: 1. Glass superstratc; 2. Multilayer deposition; 3. First polarity groove; 4. Second polarity groove; 5. Metallisation. The cell is designed to be illuminated from the glass side (the underside in this schematic) although bifacial operation is feasible. Opposite polarity regions in adjacent cells are connected, providing automatic series connection within the module. Source; Green and Hansen (1998).

At UNSW, the parallel muitijunction approach has been combined with the buriedcontact approach to produce the device fabrication sequence shown in Fig. 4.19. After deposition of a multilayer stack on a low temperature substrate such as glass, laser grooving and groove doping of one polarity is applied to connect all the layers of this polarity together in parallel. A second laser grooving and doping step involving the other polarity follows. By aligning to the first step, series interconnection of the cells is also achieved in a very elegant process. Pacific Solar in Sydney is working on commercialising this process, although few details have yet been published (Pacific Solar, 1997).

4.8 Summary Although crystalline silicon devices have dominated the commercial marketplace for photovoltaics for more than two decades, there still remains scope for considerable improvement in both the performance and cost of these cells. Recent studies suggest that manufacturing costs well below US$1 AVp are obtainable in manufacturing volumes of 500 MWp per year, without major changes in present processing sequences. This suggests module costs will steadily decrease from present values of about US$4/Wp as manufacturing volumes continue to increase. The average energy

190

M. A. Green

conversion efficiency of product sold should also increase due largely to the increasing market share of high efficiency approaches, particularly the buried-contact cell approach. The trend towards thinner silicon wafers to decrease wafer cost is compatible with ongoing increases in cell efficiency provided cell structures as effective as the buried-contact approach are adopted and improved methods are demonstrated in production for passivating the rear surface of the cell. Substrates with more consistent high temperature performance than standard CZ grown silicon may be required to allow the full performance potential of the silicon wafer approach to be obtained. Particularly promising progress has been made in this area with multicrystalline silicon over recent years. This enormous potential for both performance and cost reduction will make these bulk silicon approaches an increasingly challenging target for the thin-film approaches currently under development. In this context, excellent recent progress has been made with supported silicon film. Films processed at low temperature on substrates such as glass have made exceptional gains in the laboratory over the last 2-3 years and offer great promise for stable low cost thin-film cell technology for the future. Innovative cell design such as the parallel multilayer cell will allow the particular properties of thin-film silicon to be used to their full advantage.

References Aberle A. G., Lauinger T. and Hezel R. (1997), 'Remote PECVD silicon—a key technology for the crystalline silicon PV industry of the 21st. century?', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 684-689. Araujo G. L., Marti A., Ragay F. W. and Wolter J. H. (1994), 'Efficiency of multiple quantum well solar cells', Proc. 12th. European Photovoltaic Solar Energy Conf., Amsterdam, H. S. Stephens & Associates, Bedford, 1481-1484. Aulich H. A. (1996), 'Review of solar-grade silicon research programs carried out in Germany during the 80s', 9th. Sunshine Workshop on Crystalline Silicon Solar Cells, unpublished. Authier B. (1978), 'Polycrystalline silicon with columnar structure', Festkorperprobleme XVIII, F. Vieweg & Son, Wiesbaden. Baba T., Shima M , Matsuyama T., Tsuge S., Wakisaka K. and Tsuda S. (1995), '9.2% efficiency thin-film polycrystalline solar cell by a novel solid phase crystallisation method', Proc. 13th. European Photovoltaic Solar Energy Conf., Nice, H. S. Stephens & Associates, Bedford, 1708-1711.

Crystalline Silicon Solar Cells

191

Bai Y., Ford D. H., Rand J. A., Hall R. B. and Barnett A. M. (1997), '16.6% efficient Silicon-Film™ polycrystalline silicon solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 35-38. Barnett A. M., Hall R. B., Fardig D. A. and Culik J. S. (1985), 'Silicon-film solar cells on steel substrates', Conf. Record 18th. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 1094-1099. Barnett A. M., Dumian F. A., Ford D. H., Kendall C. L., Rand J. A., Rock M. L. and Hall R. B. (1989), 'Thin silicon-film solar cells on ceramic substrates', Proc. 4th. Int. Photovoltaic Solar Energy Conf, Sydney, Inst. Radio and Electronics Engineers, 151-158. Barnett A. M., Ford D. H., Checchi J. C , Culik S., Hall R. B., Jackson E. L., Kendall C. L. and Rand J. A. (1997), 'Very-large-area Silicon-Film™ solar cells', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 999-1002. Barnham K. W. J. and Duggan G. (1990), 'A new approach to high-efficiency multiband-gap solar cells', J. Appl. Phys. 67, 3490-3493. Bergin D. O. (1980), 'Shaped crystal growth—a selected bibliography', J. Crystal Growth 50, 381-396. BP Solar (1991), Data Sheet, BP Saturn Solar Cells. Brandhorst H. W. and Bernatowicz D. T. (1980), 'Space solar cells—high efficiency and radiation damage', Conf. Record 14th. IEEE Photovoltaic Specialists Conf, San Diego, IEEE Press, Piscataway, 667-671. Brendel R. (1997), 'A novel process for ultrathin monocrystalline silicon solar cells on glass', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 1354-1357. Bruton T. M., Luthardt G., Rasch K.-D., Roy K., Dorrity I. A., Garrard B., Teale L., Alonso J., Ugalde U., Declerq K., Nijs J., Szlufcik J., Rauber A., Wettling W. and Vallera A. (1997), 'A study of the manufacture at 500 MWp p.a. of crystalline silicon photovoltaic modules', Proc. 14th. European Photovoltaic Solar Energy Barcelona, H. S. Stephens & Associates, Bedford, 11-16. Chapin D. M., Fuller C. S. and Pearson G. O. (1954), 'A new silicon p-n junction photocell for converting solar radiation into electrical power', J. Appl. Phys. 25, 676-677. Chieng C. and Green M. A. (1993), 'Computer simulation of enhanced output from bifacial photovoltaic modules', Prog, in Photovoltaics 1, 293-299. Christensen E. (1985, ed.), Flat Plate Solar Array Project, Final Report, Jet Propulsion Laboratory.

192

M. A. Green

Chu T. L. (1977), 'Silicon films on foreign substrates for solar cells', J. Crystal Growth 39, 45-60. Corkish R. (1991), 'Some candidate materials for lattice-matched liquid phase epitaxial growth on silicon', Solar Cells 31, 537-548. De Clercq K., Frisson L., Szlufcik J., Nijs J. and Mertens R. (1997), 'Design for manufacturability of a silicon solar cell process based on screen printing by means of manufacturing science techniques', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 135-138. Dietl J., Helmreich D. and Sirtl E. (1981), "Solar' silicon', in Silicon, Vol. 5 of Crystal Growth Properties and Applications, pp. 43-107. Endros A. L., Einzinger R. and Martinelli, G. (1997), Tri-silicon: a novel substrate for thin wafer solar cells', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 112-114. Fang P. H., Ephrath L. and Nowak W. B. (1974), Appl. Phys. Lett. 25, 583. Ferrazza F. (1996), 'Growth and post-growth processes of multicrystalline silicon for photovoltaic use', in Polycrystalline Semiconductors IV—Physics, Chemistry and Technology, Pizzini S., Strunk H. P. and Werner J. H., eds., Vol. 51-52, Solid State Phenomena, Scitech Publ., Zug, Switzerland. Fischer D., Keppner H., Kroll U., Torres P., Meier J., Platz R., Dubail S., Selvan J. A. A., Vaucher N. P., Ziegler Y., Tscharner R., Hof C , Beck N., Goetz M., Pernet P., Goerlitzer M , Wyrsch N., Vuille J., Cuperus J. and Shah A. (1997), 'Recent progress of the "Micromorph" tandem solar cells', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 2347-2350. Fisher H. and Pschunder W. (1976), 'Low cost solar cells based on large area unconventional silicon', Conf. Record 12th. IEEE Photovoltaic Specialists Conf, Baton Rouge, IEEE Press, Piscataway, 86-82. Ford D. H., Rand J. A., Barnett A. M., DellDonne E. J., Ingram A. E. and Hall R. B. (1997), 'Development of light-trapped, interconnected, silicon-film modules', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 631-634. Godlewski M. P., Baraona C. R. and Brandhorst H. W. (1973), 'Low-high junction theory applied to solar cells', Conf. Record 10th. IEEE Photovoltaic Specialists Conf, Palo Alto, IEEE Press, Piscataway, 40-49. Green M. A. (1984), 'Limits on the open-circuit voltage and efficiency of silicon solar cells imposed by intrinsic Auger processes', IEEE Trans. Electron Devices, ED31,671-678.

Crystalline Silicon Solar Cells

193

Green M. A. (1987), High Efficiency Silicon Solar Cells, Trans Tech Publications, Aedermannsdorf, Switzerland. Green M. A., Zhao J., Wang A. and Wenham S. R. (1992), '45% efficient silicon photovoltaic cell under monochromatic light', IEEE Electron Devices Lett. 31, 317-318. Green M. A. and Wenham S. R. (1994), 'Novel parallel multijunction solar cell', Appl. Phys. Lett. 65, 2907-2909. Green M. A. (1995), Silicon Solar Cells: Advanced Principles and Practice, Bridge Printery, Sydney. Green M.A. and Hansen J. (1998), Catalogue of Photovoltaic Drawings, Photovoltaics Special Research Centre, University of New South Wales, Sydney. Green M. A. (1999), 'Limiting efficiency of bulk and thin-film silicon solar cells in the presence of surface recombination', Prog, in Photovoltaics 7, 327-330. Guttler G. and Queisser H. J. (1970), 'Impurity photovoltaic effect in silicon', Energy Conversion 10, 51-55. Hamamoto S., Morikawa H., Naomoto H., Kawama Y., Takami A., Arimoto S., Ishihara T. and Namba K. (1997), 'Development of the VEST cell process for low cost fabrication', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 2328-2332. Haynos J., Allison J., Arndt R. and Meulenberg A. (1974), 'The COMSAT nonreflective silicon solar cell: a second generation improved cell', Int. Conf. on Photovoltaic Power Generation, Hamburg, 487. Healy S. A. and Green M. A. (1992), 'Efficiency enhancements in c-Si solar cells by the incorporation of a region alloyed with germanium', Solar Energy Mat. Solar Cells 28, 273-284. Hoernstra J., Weeber A. W., de Moor H. H. C. and Sinke W. C. (1997), 'The importance of paste rheology in improving fine line, thick film screen printing of front side metallization', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 823-826. Janoch R., Wallace R. and Hanoka J. I. (1997), 'Commercialization of silicon sheet via the string ribbon crystal growth technique', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 95-98. Jordan D. and Nagle J. P. (1994), 'New generation of high-efficiency solar cells: development, processing and marketing', Prog, in Photovoltaics 2, 171-176. Keevers M. J. and Green M. A. (1994), 'Efficiency improvements of silicon solar cells by the impurity photovoltaic effect', J. Appl. Phys. 75,4022^*033.

194

M. A. Green

Keevers M. J., Zhang G. C , Saris F. W., Zhao J. and Green M. A. (1995), 'Screening of optical dopants in silicon solar cells for improved infrared response', Proc. 13th. European Photovoltaic Solar Energy Conf., Nice, H. S. Stephens & Associates, Bedford, 1215-1218. Khattak C. P. and Schmid F. (1997), 'Growth and characterization of 200 kg multicrystalline silicon ingots by HEM', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 111-114. King D. L. and Buck E. M. (1991), 'Experimental optimization of an anisotropic etching process for random texturization of silicon solar cells', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, SOSSOS. Kingsbury E. F. and Ohl R. S. (1952), 'Photoelectric properties of ionically bombarded silicon', Bell Syst. Tech. J. 31, 802-815. Knobloch J., Glunz S. W., Biro D., Warta W., Schaffer E. and Wettling W. (1996) 'Solar cells with efficiencies above 21% processed from Czochralski grown silicon', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 405^108. Kolodinski S., Werner J. H., Wittchen T. and Queisser H. J. (1993), 'Quantum efficiencies exceeding unity due to impact ionization in silicon solar cells', Appl. Phys. Lett. 63, 2405-2407. Lew Yan Voon L. C., Ram-Mohan L. R. and Soref R. A. (1997), 'Electronic and optical properties of (001) Si/ZnS heterostructures', Appl. Phys. Lett. 70, 1837— 1839. Lindmayer J. and Allison J. (1973), 'The violet cell: an improved silicon solar cell', COMSAT Tech. Rev. 3, 1-22. Lindmayer J. (1976), 'Semi-crystalline silicon solar cells', Conf. Record 12th. IEEE Photovoltaic Specialists Conf, Baton Rouge, IEEE Press, Piscataway, 82-85. Liidermann, R., Schaefer S., Schiile C. and Hebling C. (1997), 'Dry processing of mcsilicon thin-film solar cells on foreign substrates leading to 11% efficiency', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 159-162. Mason N. B., Bruton T. M. and Heasman K. C. (1997), 'Factors which maximise the kWh/kWp performance of PV installations in northern Europe', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 2021-2024. Maeda Y. and Hide I. (1987), 'Solar cells using spinning Si polycrystals', Tech. Digest, 3rd. Int. Photovoltaic Sci. Eng. Conf, 87-92.

Crystalline Silicon Solar Cells

195

Minagawa S., Saitoh T., Warbisako T., Nakamura N., Itoh H. and Tokuyama T. (1976), 'Fabrication and characterization of solar cells using dendritic silicon thin films grown on alumina ceramic', Conf. Record 12th. IEEE Photovoltaic Specialists Conf., Baton Rouge, IEEE Press, Piscataway, 77-81. de Moor H. H. C , Hoornstra J., Weeber A. W., Burgers A. R. and Sinke W. C. (1997), 'Printing high and fine metal lines using stencils', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 404^07. Narashima S., Crotty G., Krygowski T., Rohatgi A. and Meier D. L. (1997), 'Backsurface field and emitter passivation effects in the record high efficiency n-type dendritic web silicon solar cell', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 235-238. Nijs J. Demesmaeker E., Szlufcik J., Poortmans J., Frisson L., de Clercq K., Ghannam M., Mertens R. and van Overstraeten R. (1996), 'Recent improvements in the screen-printing technology and comparison with the buried-contact technology by 2D-simulation', Solar Energy Mat. Solar Cells 41-42, 101-118. Ohl R. S. (1941), 'Light sensitive electric device', US Patent 2,402,622 (27 March), 'Light-sensitive electric device including silicon', US Patent 2,443,542 (27 May). Okamoto S., Nishida M., Shindo T., Komatsu Y., Yasue S., Kaneiwa M. and Nanmori T. (1997), '23.5% efficient silicon solar cell with rear micro contacts of c-Si/mcSi:H heterostructure', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 255-258. Pacific Solar Pty. Ltd. (1997), Annual Review, Sydney. Ralph E. L. (1975), 'Recent advancements in low cost solar cell processing', Conf. Record 11th. IEEE Photovoltaic Specialists Conf, Scottsdale, IEEE Press, Piscataway, 315-316. Riordan M. and Hoddeson L. (1997), Crystal Fire: The Birth of the Information Age, Norton, New York. Rohatgi A., Narasimhi S., Kamra S., Doshi P., Khattak C. P., Emery K. and Field H. J. (1996), 'Record high 18.6% efficient solar cell on HEM multicrystalline material', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 741-744. Sakaguchi Y., Yuge N., Nakamura N., Baba H., Hanazawa K., Abe M. and Kato Y. (1997), 'Purification of metallic grade silicon up to solar grade by NEDO melt purification process', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 157-160.

196

M. A. Green

Sard D., Durand F., Choudhury A., Marfaing Y., Einhaus R. and Luque A. (1997), 'Electromagnetic cold crucible continuous casting for multicrystalline silicon solar cells', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 849-852. Sawada T., Terada N., Tsuge S., Baba T., Takahama T., Wakisaka K., Tsuda S. and Nakano S. (1994) 'High-efficiency a-Si/c-Si heterojunction solar cell', Proc. 1st. World Conf. on Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1219-1225. Schulz M. and Sirtl E. (1984), 'Silicon sheet', in Landolt-Bornstein: Numerical Data and Functional Relationships in Science and Technology, Springer-Verlag, Berlin, Vol. 17c, Section 6.1.2.5.3, pp. 52-54 and pp. 442-444. SerrezeH. B. (1978), 'Optimizing solar cell performance by simultaneous consideration of grid pattern design and interconnect configurations', Conf. Record 13th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 609-614. Shah A., Meier J., Torres P., Kroll U., Fischer D., Beck N., Wyrsch N. and Keppner H. (1997), 'Recent progress on microcrystalline solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 569-574. Shimura F. (1989), Semiconductor Silicon Crystal Technology, Academic Press, New York. Tanaka M., Taguchi M., Takahama T., Sawada T., Kuroda S., Matsuyama T., Tsuda S., Takeoka A., Nakano S., Hanfusa H. and Kuwano Y. (1993), 'Development of a new heteroj unction structure (ACJ-HIT) and its application to polycrystalline silicon solar cells', Prog, in Photovoltaics 1, 85-92. Tiedje T., Yablonovitch E., Cody G.D. and Brooks B. G. (1984), 'Limiting efficiency of silicon solar cells', IEEE Trans. Electron Devices ED-31, 711-716. Verlinden P. J., Sinton R. A., Wickham K., Crane R. A. and Swanson R. M. (1997), 'Backside-contact silicon solar cells with improved efficiency for the '96 world solar challenge', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 96-99. Wallace R. L., Hanoka J. I., Narasimha S., Kamra S. and Rohatgi A. (1997), 'Thin silicon string ribbon for high efficiency polycrystalline solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, Piscataway, 99-102. Weber K. J., Catchpole K., Stocks M. and Blakers A. W. (1997), 'Lift-off of silicon epitaxial layers for solar cell applications', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 107-110.

Crystalline Silicon Solar Cells

197

Wenham S. R., Robinson R., Dai X., Zhao J., Wang A., Tang, Y. H., Ebong A., Honsberg C. B. and Green M. A. (1994), 'Rear-surface effects in high efficiency silicon solar cells', Proc. 1st. World Conf. on Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1278-1282. Wenham S. R. and Green M. A. (1995), 'Thin film silicon formation using porous silicon as a contamination barrier', Australian Provisional Patent Application PN6061, 19 October 1995. Werner J. H., Brendel R. and Queisser H. J. (1994), 'New upper efficiency limits for semiconductor solar cells', Proc. 1st. World Conf. on Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1742-1745. Wolf M. (1960), 'Limitations and possibilities for improvement of photovoltaic solar energy converters', Proc. Inst. Radio Engineers 48, 1246-1263. Wolf M. (1976), 'Historical development of solar cells', in Solar Cells, Backus C. E., ed., IEEE Press, Piscataway. Yamamoto K., Yoshimi M., Suzuki T., Okamoto Y., Tawada Y. and Nakajima A. (1997), 'Thin film poly-Si solar cell with "Star Structure" on glass substrate fabricated at low temperature', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 575-580. Zhao J., Wang A., Altermatt P. and Green M. A. (1995), '24% efficient silicon solar cells with double layer antireflection coatings and reduced resistance loss', Appl. Phys. Lett. 66, 3636-3638.

CHAPTER 5

AMORPHOUS SILICON SOLAR CELLS CHRISTOPHER R. WRONSKI Centerfor Thin Film Devices, The Pennsylvania State University, University Park, PA 16802 Crwece @ engr.psu. edu and DAVID E. CARLSON BP Solarex, 3601 Lagrange Parkway, Toano, VA 23168 [email protected]

One advantage of being disorderly is that one is constantly making exciting discoveries. A. A. Milne, Winnie-The-Pooh, 1926.

5.1 Introduction In the last few years significant progress has been made in improving the efficiencies of amorphous silicon (a-Si)-based solar cells and in the scale of production of a-Si PV modules, which is currently about 30 peak megawatts (MWP) per year. These advances, together with the already established large-scale mass production of liquid crystal displays based on hydrogenated amorphous silicon (a-Si:H) thin-film transistors, indicate a coming of age for amorphous silicon technology. The first small-area amorphous silicon solar cells were fabricated with initial efficiencies of 12% (Carlson and Wronski, 1976). Today one square foot panels are being fabricated with stabilised efficiencies greater than 10% and modules are commercially produced with areas up to 12 square feet (Wronski, 1996; Guha, 1996; Forrest, 1997). The progress in a-Si solar cell technology is due to concurrent advances in the areas of new materials, novel cell designs and the large-area deposition techniques suitable for mass production. This progress has predominantly been a consequence of the lack of long-range order, such as is present in crystalline silicon (c-Si), which on one hand drastically changes the photovoltaic properties of the a-Si materials but on the other offers great flexibility in the manufacture of different solar cell structures as well as large-area monolithic modules.

199

200

C. R. Wronski and D. E. Carlson

Several outstanding features of amorphous silicon-based technology have allowed continuous advances to be made in the performance and manufacture of a-Si solar cells. Cells can be fabricated using a-Si alloyed with hydrogen, germanium and carbon to form semiconductors with band gaps between about 1.3 and 2.0 eV (Dawson et ah, 1992). These band gaps allow the fabrication of not only singlejunction, but also tandem and triple-junction, stacked cells designed in such a way as to maximise the absorption of the solar spectrum (Yang and Guha, 1992; Yang et al., 1998). These a-Si alloys are excellent candidates for thin-film solar cells since, unlike crystalline silicon, they have optical absorption coefficients similar to those found in direct band-gap semiconductors (Collins and Vedam, 1995). In addition, unlike any other amorphous semiconductors, these a-Si-based materials can be doped both p-and n-type (Spear and LeComber, 1975), which allows high-quality junctions and nohmic contacts to be used in solar cell fabrication (Carlson and Wronski, 1976; Wronski, et al., 1976). Last but not least, the fabrication processes of these cells are run at temperatures less than 300 C, which allows uniform thin films and cells to be reproducibly deposited over large areas. Because the long-range order present in c-Si is absent in a-Si there are significant differences between the semiconductor properties of the two materials. The properties that are important in solar cells and their contributions to cell performance are reviewed in Section 5.3. These include the properties of intrinsic and doped materials as well as the reversible light-induced changes that occur under sunlight. Despite the amorphous nature of the materials, their photovoltaic properties do depend on their microstructure. This is determined by growth kinetics and hence by the various fabrication conditions discussed in Section 5.4. The principal mechanisms that determine the operation of efficient a-Si-based solar cells, and which are different from those in crystalline Si cells, are briefly described in Section 5.5. Next, a review is presented of how these cells have been optimised by incorporating different a-Si:H alloy materials into various cell structures and multijunction stacked cells. Section 5.6 describes the different cell structures utilised by various organisations and discusses the characteristics and performance of single-junction, tandem and triple-junction cell structures. Section 5.7 discusses the development and commercialisation of a-Si:H modules fabricated on glass, stainless steel and plastic substrates. Issues related to the manufacturing costs of such modules are presented in Section 5.8 and their long-term reliability is discussed in Section 5.9. Environmental issues and the challenges for the future are reviewed in Sections 5.10 and 5.11 respectively.

Amorphous Silicon Solar Cells

201

5.2 Background Amorphous silicon deposited from silane was first investigated by Chittik et al. (1969). Although subsequent work was carried out on this material, its potential as a useful optoelectronic material was not recognised until 1976. In that year, Carlson and Wronski (1976) reported the first results on 2% efficient a-Si solar cells and shortly afterwards on ones with 5% efficiency (Carlson et al., 1976). This sparked worldwide interest in a-Si and a-Si-based solar cells, and led to numerous fundamental studies on the materials and improving the performance of a-Si solar cells. Once the importance of hydrogen in these materials had been recognised (Brodsky et al., 1977), a wide range of deposition techniques and conditions were employed in attempts to improve the properties of hydrogenated amorphous silicon (a-Si:H) for solar cells. These included not only the intrinsic optoelectronic and photovoltaic properties of the material, but also the equally important, large changes that occur upon exposure to sunlight and which are perfectly reversible on annealing at ~150C for a few hours (Staebler and Wronski, 1977). These light-induced changes, known as the StaeblerWronski effect (SWE), manifest themselves in both thin-film materials and solar cells. The early discovery of the SWE had an enormous effect on the development of a-Si solar cells and their technology by having a major impact on cell design. Engineering approaches were developed for minimising the effects of the SWE on the degraded steady-state (i.e. stabilised) cell efficiencies by making the cells as thin as possible (Hanak and Korsun, 1982). Reduction in cell thickness, and the corresponding lower absorption of sunlight in the cell, leads to lower short-circuit currents (i^), which reduces the possible power conversion efficiency. This problem was greatly reduced with the development of efficient optical enhancement (Yablonovitch and Cody, 1982; Deckman et al., 1984) obtained by introducing textured, rather than smooth, optical reflectors. Such optical enhancement, which was first successfully applied to a-Si-based solar cells, is now extensively used in all types of thin-film solar cells. Another key element in the development of efficient a-Si solar cells with thinner absorber layers was the introduction of amorphous silicon-germanium alloys, which have band gaps significantly lower than those of a-Si:H. This allowed not only single- but also tandem and triple-junction cells to be fabricated (Yang and Guha, 1992; Yang et al., 1984). With successful engineering, both the initial and the degraded steady-state efficiencies of these cells have been steadily improved.

202

C. R. Wronski and D. E Carlson

5.3 Amorphous silicon based materials 5.3. J Introduction The materials used in amorphous silicon-based solar cells are prepared by plasmaenhanced chemical vapour deposition (PECVD), which relies on the decomposition of gases containing silane (S1H4). These materials are in fact silicon-hydrogen alloys, typically containing 5-20 at.% hydrogen. The key function of the hydrogen in these materials is to passivate the broken Si bonds that are introduced by the absence of the long-range order present in c-Si. The differences in the structure between c-Si and hydrogenated amorphous silicon (a-Si:H), and the passivation of dangling bonds by hydrogen, is illustrated in Fig. 5.1. The diagram on the left shows the tetrahedrally bonded crystal structure that extends throughout the lattice of c-Si. On the right is a schematic diagram of the a-Si:H network, which has a Si-Si nearest neighbour configuration similar that of c-Si but no long-range order, and consequently a large number of dangling (broken) bonds. Luckily, as indicated in the figure, the vast majority of these bonds are passivated by the hydrogen in the a-Si:H-based materials.

Figure 5.1 The tetrahedrally bonded crystal structure of c-Si is shown on the left. The absence of longrange order and passivation of Si dangling bonds by hydrogen is illustrated on the right.

This passivation greatly reduces the density of the -10 cm" dangling-bond defects present in unhydrogenated a-Si. The incorporation of hydrogen also leads to materials with significantly wider band gaps than c-Si, but which, because of their disorder, have at the same time much higher optical absorption. The atomic hydrogen present during the growth of the materials also plays a very important role in determining the growth kinetics and the resulting microstructure. With high dilution

Amorphous Silicon Solar Cells

203

of the feedstock silane gas with hydrogen it is even possible (and sometimes advantageous) to obtain microcrystalline Si:H (/JC-Si:H) alloys (Hirose, 1984; Tanaka and Matsuda, 1987; Collins and Fujiwara, 1997). Hydrogenated amorphous silicon can also be grown with germanium and carbon to produce a-SiGe:H and a-SiC:H alloys, materials, which are extremely useful as components of a-Si:H-based solar cells. However, because the properties of a-Si-based materials depend on the growth and deposition conditions, unlike c-Si there is no unique form of a-Si:H. While this allows a wide range of materials to be developed for solar cell applications, it has on the other hand made it difficult to obtain consistent material parameters, since studies are generally carried out on materials grown under different conditions.

5.3.2 Band gaps and optical absorption The disorder in a-Si:H-based materials transforms the nature of the optical absorption associated with the indirect band gap of c-Si to that of direct band-gap semiconductors (Collins and Vedam, 1995). Incorporation of hydrogen into the a-Si:H network not only removes defects, and defect states in the forbidden gap, but also widens the gap. Hence when the hydrogen content in a-Si:H is increased from about 5% to 20%, the band gap increases from -1.6 eV to -1.8 eV (Zanzucchi et al., 1977). The band gaps of a-SiGe:H and a-SiC:H alloys depend on the concentrations of the Ge and C as well as that of hydrogen. The alloys used in solar cells have Ge up to about 60 at.% and C up to about 20 at.% with band gaps which are down to about 1.3 eV for the a-SiGe:H materials and up to about 2.0 eV for the a-SiC:H materials (Dawson et al, 1992; Lu et al, 1994; Ganguly and Matsuda, 1996). By incorporating different amounts of hydrogen into a-Si:H, it is possible to change not only its band gap but also the density of states in the gap; materials prepared with 5-15 at.% of hydrogen generally have densities of dangling-bond states on the order of 1015 to 1016 cm"3. The optical absorption of such a-Si:H materials is shown in Fig. 5.2 for three RF PECVD films prepared under similar conditions but at different substrate temperatures Ts, with the corresponding optical gaps Ug (Tauc et al., 1966) also shown in the figure. The changes in the gaps are clearly reflected in the systematic horizontal shifts of the absorption spectra, where regions of a greater than 10 cm" correspond to optical transitions between the valence and conduction bands. The regions of a between about 103 and 10 cm-1, which are exponential in nature, arise from the absorption in valence-band tail states, created by the disorder in these amorphous materials (Roxlo et al., 1983). The densities of the valence-band tail states are significantly higher than those of the conduction-band tails (Tiedje, 1984), even in

204

C. R. Wronski and D. E. Carlson

I 1.0

1.2 1.4 Energy/eV

1.6

Figure 5.2 Optical absorption a versus photon energy for a-Si:H films deposited by RF PECVD at three substrate temperatures, T„ showing also the resulting band gaps Ug. After Wronski (1996).

the recently developed materials with a highly ordered network (Koh et al., 1998). The shoulders in the absorption spectra at photon energies less than -1.4 eV, with values of a below -10 cm"1, are due to optical transitions originating from defect states near and around the middle of the gap (Jackson et al., 1983; Wronski et al., 1997). The energy band diagram and the distribution of gap states in a-Si:H are shown schematically in Fig. 5.3. Unlike c-Si, the a-Si materials have a continuous distribution of localised states in the gap through which electrons and holes cannot move freely. Near the conduction and valence band edges are the two sets of band tail states, whose densities decrease exponentially from the main conduction and valence band edges. There are also deep-lying gap states whose densities, generally represented by gaussian distributions, consist of: neutral dangling-bond states (D°) in the middle of the gap; negatively charged defect states (D~) below the middle of the gap; and positively charged defect states (D+) above the middle of the gap (Branz and Silver, 1990; Powell and Dean, 1993; Jiao et al, 1996a). These deep-lying states are very important in determining the collection of photogenerated carriers in a-Si solar cells. The absorption that is useful in creating free carriers in solar cells is at values of a greater than about 103 cm-1, which corresponds to photons with energies greater than the band gap of the material. Even though a wide range of band gaps is available from

Amorphous Silicon Solar Cells

1

205

Conduction Band

/ ""CTTTair

I

=^H

z>°* ^ > D°

^ = = *!!>-<-' >

D

^BJail Valence Band

\ =m

DENSITY OF STATES Figure 5.3 A plot of the densities of states versus energy for a-Si:H, illustrating the continuous distribution of the different localised states.

the a-Si:H alloy materials, the values that can be utilised in efficient solar cells are limited by the increase in the deep-lying defect states that occurs with the introduction of Ge and C. The a-Si:H alloy materials that currently allow photogenerated carriers to be efficiently collected in solar cell structures have band gaps between about 1.3 and 2.0 eV, a sufficiently wide range to offer the flexibility required for constructing efficient single-junction, as well as multijunction, solar cells.

5.3.3 Intrinsic properties The intrinsic materials obtained under the usual deposition conditions contain impurities such as O, N, C, typically in the concentration range 1018—10 cm" (Carlson, 1984). Interestingly, significant reduction in the levels of these impurities to values as low as 10 l5 -10 l6 cirf 3 does not appear to improve the semiconductor properties or the stability of the materials (Kamei et al., 1996). The semiconductor properties and stability do, however, depend on the mechanisms that determine the growth processes, the incorporation of hydrogen and the resultant microstructure (Tanaka and Matsuda, 1987; Collins, 1994; Collins and Fujiwara, 1997). By optimising growth conditions, it has been possible to obtain a-Si:H materials with properties that are outstanding for an amorphous semiconductor (Lee et al., 1996). These intrinsic (undoped) materials have Fermi levels near the middle of the gap and free-carrier transport via the delocalised states of the conduction and valence bands. However, because of the lack of long-range order, the free-carrier mobilities in these bands are significantly lower than those in c-Si, being only about 10 and 1 cm2 V - ' s~

206

C. R. Wronski and D. E. Carlson

for electrons and holes respectively (Tiedje, 1984). The exponential distributions of gap states near the bands, such as seen in Fig. 5.3, may not directly affect the extended-state mobilities but they can have an effect on solar cell performance. The gap states with the largest effect on the transport of photogenerated carriers are those located near midgap, which act as very efficient recombination centres. Keeping their densities down to the previously discussed levels of ~1015-1016 cm -3 is key to efficient solar cell operation. This allows the hole and electron lifetimes to be as high as 10~8 -KT 6 s, and the low space-charge densities associated with such a low concentration of defect states allows the junction electric field to extend over the entire thickness of efficient solar cells (Carlson and Wronski, 1979). Incorporation of Ge and C introduces deep-lying gap states into the a-SiGe:H and a-SiC:H alloys. This reduces the carrier lifetimes but does not affect the extendedstate free carrier mobilities. In the case of a-SiC:H, even low carbon incorporation has a large effect on the microstructure so that the densities of defects, particularly after prolonged illumination in sunlight, are too high for this material to be useful as an absorber layer in solar cells. However, this is not the case for a-Si:Ge:H alloys, which retain good semiconductor properties and densities of midgap states around 1016 cm"3 even with about 60 at. % of Ge.

5.3.4 Doping Amorphous Si:H-based materials are the only amorphous materials that can be doped both n- and p-type. High-quality undoped materials are intrinsic (t) or slightly n-type, with Fermi levels at or near midgap. Because the midgap state density is low, the introduction of donors and acceptors can readily move the Fermi levels towards the conduction and valence band, respectively, so the initially high-resistivity materials can become very conductive. As in the case of c-Si, n-type doping is achieved by incorporating phosphorus into the materials, and p-type doping by incorporating boron (Spear and LeComber, 1975). The incorporation of high densities of these dopants, however, also introduces defect states near midgap that limit the doping efficiency and drastically reduce free-carrier lifetimes (Wronski et al., 1982). As a result, doped materials cannot be used as active absorber layers in solar cells, as they can in c-Si p-n solar cells. Instead, a-Si:H-based solar cells take advantage of the excellent transport properties of their intrinsic materials and utilise both p-i-n and n-i-p cell structures. These cells are actually heterojunction structures which have ptype a-SiC:H or p-type /ic-Si:H p/i junctions and ohmic contacts formed of either ntype a-Si:H or n-type ,uc-Si:H layers. The p-type a-SiC:H and the p-type jUc-Si:H

Amorphous Silicon Solar Cells

207

result in excellent contacts since their quasi-Fermi levels are located about 0.4 eV and 50 meV from their respective valence bands. The n-type a-Si: H and n-type /xc-Si:H provide excellent ohmic contacts to the Mayers since their Fermi levels are about 0.2 eV and 50 meV from their respective conduction bands. By incorporating such ptype and n-type materials into p-i-n or n-i-p cells, it is possible to obtain built-in voltages well over 1 V (Lee et al, 1997).

5.3.5 The Staebler—Wronski effect The light-induced changes in a-Si:H-based materials known as the Staebler-Wronski effect (SWE) are not only of great scientific interest, but also of great technological importance, because of their effect on the long-term stability of a-Si:H-based solar cells and modules. The SWE was first observed as changes produced by sunlight in the carrier transport of thin films, which were found to be completely reversible on annealing the materials for a few hours at -150 C (Staebler and Wronski, 1977). These changes result from the introduction of metastable defects whose rate of creation and densities depend on the intensity of illumination and the temperature. The reversible changes that occur between an annealed initial state A and a 'lightsoaked' state B have become one of the most investigated phenomena in a-Si:H-based materials and solar cells (Wronski, 1984; Fritzsche, 1997; Stutzmann, 1997; Wronski, 1997). However, progress in obtaining a definitive understanding and control of the SWE has been relatively slow. This is largely because of the complex nature of the defects and the differences in the microstructure of a-Si:H materials prepared under different growth conditions. Figure 5.4 shows an example of the now well-known decreases in photoconductivity and dark conductivity that occur on prolonged illumination of a-Si:H thin films. Figure 5.4 shows the change in photoconductivity (upper curve) of a-Si:H film annealed at 150 C, state A, as it is illuminated with 100 mW cm"2 of white light. Also shown is the corresponding decrease of the dark conductivity during the illumination period (lower curve). These changes are a direct consequence of the defects created in the a-Si:H by the light. These metastable defects act as additional recombination centres that reduce the carrier lifetimes and hence the photoconductivity. Their presence in the gap also pulls the Fermi level towards midgap, which reduces the free electron concentration and decreases the dark conductivity by several orders of magnitude. Though advances have been made in the understanding of the SWE, there is as yet still no consensus on the exact nature of the light-induced defects or the mechanism

208

C. R. Wronski and D. E. Carlson

io"-

•"

>0-9.

illumination

\

O

dark conductivity

50

100

ISO

2O0

250

Time/min

Figure 5.4 Dark conductivity and photoconductivity of an a-Si:H thin film shown as a function of the illumination time with sunlight. After Staebler and Wronski (1980).

responsible for their creation. There is general consensus, however, that the hydrogen that plays a key role in eliminating dangling-bond defects in a-Si:H alloys also plays a key role in their light-induced creation (Lee et al., 1996; Carlson and Rajan, 1998). For a long time, the widely held view was that the only defect states produced by light were associated with the neutral dangling bond, D°. However, there is now extensive evidence indicating the importance of microstructure, other than that associated purely with hydrogen, and showing that light-induced changes in the charged defect states are just as important as, if not more important than, those in the D° states (Wronski et al., 1997; Lu et al, 1999). Significant progress has been made over the years, not only in improving the initial (state A) properties of a-Si-based materials, but also in reducing the SWE. This has been achieved by optimising growth conditions to improve the microstructure of the materials through incorporation of hydrogen into the network. As a result it is possible to obtain solar cells with not only higher initial efficiencies but, more importantly, better performance after they reach degraded steady state under illumination with 1 Sun. In addition, these materials and their solar cells require much shorter times to reach the degraded steady state in sunlight, under 100 hours as compared to thousands of hours in the past (Yang and Chen, 1994; Lee et al., 1996). This makes fundamental studies of the SWE, as well as those on solar cell improvements, more amenable to detailed investigations.

Amorphous Silicon Solar Cells

209

5.4 Growth and microstructure The a-Si:H-based materials used in solar cells are usually deposited by PECVD at substrate temperatures from about 100 to 300 C. The decomposition of the feedstock gases may be carried out in a variety of reactor geometries, and with plasmas generated over an extremely wide frequency range, including DC; RF (13.56 MHz); VHF (60-100 MHz); and microwave (2.45 GHz) (Uchida, 1984; Watanabe et al., 1986; Tanaka and Matsuda, 1987; Shah et al, 1996; Collins and Fujiwara, 1997). However, in all cases the growth process can be considered to occur in three stages. The first stage is the dissociation of S1H4 into a partly ionised reactive mixture. Next, while the mixture is transported to the surface of the growing film, there are continuous chemical reactions between the different species. The species arriving at the surface are adsorbed on the growing film, where they can react with both the film itself and the radicals in the gas phase. The resulting by-products (mainly hydrogen and unreacted silane radicals) desorb from or are etched off the surface by the reactive species arriving at it. The main precursor in the growth is the SiH3 radical, but other neutral species, such as Si, SiH and SiH2, also reach the growing surface and have a pronounced effect on the structural, optoelectronic and photovoltaic properties of the materials. Hydrogen coverage of the growing surface is desirable since it is a critical factor for surface mobility of the precursor species. The high mobility allows the radicals to find more stable sites for forming a dense random network, leading to superior material. At a given substrate temperature, therefore, there is a trade off between any increase in surface diffusion and the desorption of hydrogen that leaves behind unpassivated dangling bonds. The quality of amorphous silicon-based films is determined by deposition parameters such as the substrate temperature, the pressure, the flow rate of the source gases, the plasma frequency, the power and the electrode spacing. As we have just noted, the substrate temperature is a critical parameter, and since it controls hydrogen incorporation it can be used to tailor the band gap of the materials. While decreasing the substrate temperature can increase the optical band gaps, the accompanying changes in the growth processes must be taken into account. These include both the changes in microstructure arising from the lower diffusivity of species on the surface, and an undesirable tendency to incorporate polyhydrides such as SiH2 and S1H3. The optimum substrate temperature range of -180-250 C maximises the surface mobility of the surface radicals while at the same time allowing adequate hydrogen surface coverage for passivating Si dangling bonds (Tanaka and Matsuda, 1987). The properties of the materials also depend on the pressure of the source gases. At low pressures, the growing surface can suffer severe ion bombardment. At high

210

C. R. Wronski and D. E. Carlson

pressures, on the other hand, because of the increased collision frequency between electrons and radicals in the plasma, there is a tendency to create powder in the gas phase, which introduces dihydride and polyhydride complexes into the deposited material. The flow rate of the source gases is an important deposition parameter since it determines the residence time of the different molecules in the plasma and hence affects the growth kinetics. The frequency used also affects the nature of the plasmas, and in particular the ion bombardment intensity, which becomes significantly lower at VHF and microwave frequencies. The nature of the plasmas and growth processes also changes with the introduction of the alloy-forming gases, GeH4 and CH4, the ntype dopant PH3, and the p-type dopants B2H6 or trimethyl boron. However, in all the types of depositions using PECVD, hydrogen plays a key role in reducing defects and improving the quality of the a-Si materials. 10000

1000 • <

""^ 200 •o

100

10

5

10

20

40

100

R=[H2y[SiH4] Figure 5.5 Film thickness dn for the phase transition from a-Si:H to //c-Si:H plotted versus the hydrogen dilution ratio R. After Koval et al. (1999).

The beneficial effect of hydrogen in the feedstock gases has been successfully enhanced by diluting them with molecular hydrogen. This was used first in the deposition of a-Si:H alloys (Ganguly and Matsuda, 1996), and more recently in the fabrication of intrinsic a-Si:H (Lee et al, 1996). Dilution with hydrogen has a profound effect on film growth, beginning with the nucleation and coalescence of the thin films and then controlling the bulk as well as the growing surface (Collins, 1994). With a relatively low hydrogen dilution ratio R (= [ItyMSiFL,]) of 10, not only does a-Si:H film growth become dependent on the substrate but also the microstructure becomes thickness-dependent. The materials, which are initially amorphous, eventually become microcrystalline during growth. The thickness at which the transition from the amorphous to microcrystalline phase occurs depends on both the substrate and the dilution ratio R. This is illustrated in Fig. 5.5, which shows the

Amorphous Silicon Solar Cells

211

thickness at which this phase transition is detected by real-time spectroscopic ellipsometry (Koh et al, 1998) for different values of R. These films were deposited at 200 C using RF discharge decomposition of H2 and SitL, onto substrates that are undiluted (R = 0) a-Si:H layers. The best amorphous materials are protocrystalline (Lu et al, 1994; Koval et al, 1999), corresponding to a-Si:H which is close to the //cSi:H regions, but to the left of the transition regime shown on the phase diagram in Fig. 5.5. These materials not only exhibit improved photovoltaic properties but also degrade less and reach a degraded steady state sooner (Lee et al, 1996). The low-temperature PECVD process used in the fabrication of a-Si:H materials and cells offers a number of technological advantages. Not only can it be readily scaled up to produce photovoltaic modules with very large areas, but it allows an extremely high degree of uniformity to be achieved over these areas. Moreover, PECVD is a deposition process that allows controlled changes in composition to be made with very high precision during film growth. Such controlled growth has been successfully applied to improving pli interfaces as well as band-gap profiling of alloy materials in solar cells. However, the deposition rate of the best materials using PECVD is only about 1 A s"1, which limits the mass production of a-Si:H solar cells. The challenge is to increase the deposition rate of the a-Si materials substantially while maintaining their outstanding properties.

5.5 5.5.1

Solar cells Principles of solar cell operation

The operation of all solar cells is based on common physical principles. However, since efficient a-Si based solar cells rely on material properties that are distinctly different from those of crystalline Si, the mechanisms that determine and limit their performance will be briefly reviewed. An indication of these differences is given by the basic cell structures used. In order to take advantage of the excellent properties of intrinsic (undoped) a-Si:H and a-SiGe:H, p-i-n and n-i-p heterojunction cell structures are used rather than the classic nip structures of c-Si cells. Normally, the p-i-n cell structures are p(a-SiC:H)-/(a-Si:H or a-SiGe:H)-n(a-Si:H or /ic-Si:H), and the n-i-p cells are n(a-Si:H and jUc-Si:H)-i(a-Si:H and a-SiGe:H)-p(|ic-Si:H). In both cases, high built-in potentials, well in excess of 1 V, are obtained with heavily doped n- and p-layers that serve as ohmic and window contacts. Figure 5.6 shows the energy-band diagrams of a /?(a-SiC:H)-/(a-Si:H)-«(a-Si:H) solar cell at thermodynamic equilibrium in the dark and at the maximum power point under AM 1.5 illumination.

212

P Layer

C. R. Wronski and D. E. Carlson Intrinsic Layer

(a)

N Layer

P Layer

Intrinsic Layer

N Layer

(b)

Figure 5.6 Calculated energy-band diagrams of a p(a-SiC:H)-/(a-Si:H)-n(/«:-SI:H) cell with a 320 nm thick /-layer with a band gap of 1.76 e V and 10"crrfJ densities of neutral dangling bond and charged defects (a) in the dark under zero bias, showing the built-in potential Vi„; (b) at the maximum-power point under AM 1.5 illumination, showing the different recombination paths for carriers; Ut„ and U

The built-in potential, which is determined by the separation of the Fermi levels iff and t7NF in thep- and n-contacts, is in this case 1.3 V. This appears in Fig. 5.6a as the band bending across the /-layer, where the electric field is clearly large, as indicated by the gradient of the band bending. Built-in potentials like this not only allow high V^ values to be obtained, but also result in built-in electric fields greater than 104 V cm"1 across cells with ('-layers less than 1 /xm thick. These fields help to overcome some of the limitations on collection of photogenerated carriers imposed by their low electron and hole mobilities, as well as their low diffusion lengths, all of which are much smaller than in c-Si. The efficient collection of carriers generated by sunlight in the /-layers now no longer depends on purely diffusive processes, but also relies on the electric fields to make the average time taken to transit the /-layers shorter than their recombination lifetimes (Carlson and Wronski, 1979). Such field-assisted carrier collection is, however, very sensitive to the thickness L of the /-layer, being approximately proportional to 1/L2 rather than \IL. This makes it difficult to maintain efficient carrier collection when the /-layer is made thicker in order to increase the amount of sunlight absorbed in the cell. The corresponding decrease in the average electric field and carrier collection has an effect not only on 4c, but even more so on the fill factor since under load (forward bias) the fields are significantly reduced. Consequently, the fill factor is the cell parameter that is most

213

Amorphous Silicon Solar Cells

sensitive to the cell thickness as well as to the number of states in the gap, since they determine the carrier lifetimes and the space-charge densities responsible for the electric field distributions across the Mayers. Although the neutral, dangling band-gap states (D° in Fig. 5.3) are located near the middle of the gap and act as efficient recombination centres, there are also significant contributions to the recombination kinetics from the D+ and D" states even though they are located closer to the band edges (Wronski et al., 1997; Lu et al., 1999; Jiao et al, 1996b). Despite their displacement from midgap, the D+ and D" states become important under solar illumination, when there are large densities of photogenerated carriers in the Mayers. Under these conditions, as indicated in Fig. 5.6b, the electron and hole quasi-Fermi levels, U^ and U^, are close to the band edges, and the recombination of carriers proceeds through all the gap states between U^ and (7fp. For the case shown, the separation between U^ and U^ is about 0.9 eV so both D+ and D~

I Intrinsic layer thickness

-

0.0

0.2

3300A (FF=.73) - 5500A (FF=.68) 0.4

0.6

0.8

1.0

VOLTAGE (V)

Figure 5.7 Current-voltage characteristics under AMI .5 illumination for a-Si:H p-i-n solar cells with 330 and 550 n m /-layers.

states act as recombination centres and their space charge is a key factor in determining the electric field across the Mayer. The large decrease in the electric field that occurs under load can be inferred from Fig. 5.6b, where there is no longer any significant band bending across the Mayer, as there is in Fig. 5.6a. It is therefore not surprising that the fill factor is strongly dependent on the thickness of the Mayer. This is illustrated in Fig. 5.7, which shows the measured light I-V characteristics for /?(aSiC:HH(a-Si:H)-n(/ic-Si:H) cells with Mayer thicknesses of 330 and 550 nm, similar to those in Fig. 5.6. It is interesting that in this case the carrier collection under short-circuit conditions is still efficient, as shown by the increase of ix with thickness. However, the increased thickness lowers the effective collection efficiency under load, and the FF decreases from 0.73 to 0.68.

214

C. R. Wronski and D. E. Carlson

In addition to carrier recombination in the bulk of the /-layer, additional recombination is possible in the pli interface region adjacent to the p-contact, as indicated in Fig. 5.6b. Such pli regions tend to have defect densities greater than those in the bulk and, even though they extend over a relatively thin region, they can have a large effect on both carrier recombination and electric field distribution. Furthermore, due to their location close to the p-layer they affect not only the fill factor but also the open-circuit voltage (Lee et al., 1998). They also affect other cell characteristics that depend on generation-recombination processes, such as the forward-biased dark currents. This is illustrated in Fig. 5.8, which shows the measured dark l—V characteristics of two p-i-n solar cells, similar to the thin one of Fig. 5.6. The logarithmic dark current, which is due to generation-recombination via near-midgap defect states, is reduced by an order of magnitude by improving the pli interface by high hydrogen dilution, with a corresponding increase in V^ of 30 mV (Wronski et al., 1997; Lee et al., 1998). With such improved interface regions, the currents are no longer dominated by the pli interface, becoming limited by the bulk. ~r~

10° I

~~I

a-Si:H p-i-n solar cell (0.32 urn) 10"2

-

*L „;

0.0

•

Standard

•

Hydrogen treated pli interlace

>•;

0.5

^r/' J*/

1.0

Voltage/V

Figure 5.8 The dark forward current-voltage characteristics of two p-i-n solar cells that are identical except that one has a hydrogen-treated (improved) pli interface region and the other does not. After Wronski et al. (1997).

The mechanisms determining the operation of a-Si solar cells clearly depend strongly on the light-induced defects associated with the SWE. The introduction of defects after prolonged exposure to sunlight reduces the free-carrier lifetimes and increases the space charge, which leads to a redistribution of the electric fields across the /-layers. This leads to changes in the quantum efficiencies as a function of wavelength, but fortunately these result in only a small decrease in /sc in high-quality

Amorphous Silicon Solar Cells

215

solar cells, particularly when the Mayer is less than -300 nm thick. The defects generated in the bulk of the Mayers do have a large effect on the fill factors, and their light-induced changes are the major contributor to the loss in cell efficiencies. Lightinduced defects in the pli interface regions have the most pronounced effect on lowering the open-circuit voltages, but with good pli interfaces high open-circuit voltages that do not degrade in sunlight are obtained. In summary, the limitations on the operation of a-Si solar cells imposed by the low mobilities of the free carriers are compensated by the high optical absorption and low space-charge densities of the Mayers, which allow the electric fields to sweep out the carriers before they recombine. Both the recombination lifetimes and the electric fields depend strongly on the densities and types of the defect states. Thus in large part they determine the thicknesses of cells in which the photogenerated carriers can be sufficiently well collected to give high values of fill factors.

5.5.2

Optimisation of solar cells

From the beginning, the developers of a-Si-based solar cells sought improvements in the performance, not only of the initial cells, but also of the degraded, steady-state cells. At first the effort focussed on single-junction a-Si:H cells, but this quickly expanded to include optimisation of tandem and multijunction cells. Particular attention was paid to the degraded steady-state efficiencies obtained after prolonged exposure to AM 1.5 sunlight. The development proceeded along several tracks, which included improvements in materials and solar cell structures as well as engineering approaches for minimising the effects of the SWE. It also relied heavily on the flexibility that a-Si alloys offer in terms of band gaps and their ability to generate high open-circuit voltages with p-type a-SiC:H and /ic-Si:H materials. Improvements in V^ were obtained with better p-type contacts and improved pli interface regions. Improvements in ix were obtained by using lower band-gap materials to increase absorption in the intrinsic layers. For single-junction solar cells, the a-Si:H band gaps giving the highest efficiencies are around 1.7 eV. With these band gaps, open-circuit voltages of 0.9 V can be obtained while at the same time a 1 /jm thick Mayer absorbs a fraction of AM 1.5 sunlight that is sufficient to generate a short circuit current of -18 mAcnf 2 . The challenge, however, is to maximise ix by increasing the thickness of the Mayers while at the same time retaining the collection of carriers at a level necessary for high values of fill factors. Improved materials with low defect densities allow the thicknesses of Mayers to be extended while still having high carrier collection efficiencies, but thus far these thicknesses are still significantly

216

C. R. Wronski and D. E. Carlson

less than 1 /im. The acceptable thickness of high-performance cells is further limited by the SWE, since the light-induced defects degrade the carrier transport and further limit the carrier collection efficiencies. A major breakthrough in achieving thin, high-efficiency cells was achieved with the development of optical enhancement based on textured substrates and reflectors (Yablonovitch and Cody, 1982; Deckman et al., 1984). The optical enhancement arises from the large-angle scattering caused by the surface texture, which produces multiple internal reflections that allow weakly absorbed light to undergo many passes through the Mayer. This greatly increases the already high optical absorptivity at longer wavelengths, so that significantly higher quantum efficiencies can be obtained 1.0

1

•

.

.9 .8

FlatAg

! • '

\ \

s f .6 8 -

Flat Cr - ^

\ ^ '•. \ \ \

Detached" reflector

•.

VC

B .5 j> t> .4 •o

* £

3

Tuned reflector

-

\

*.

\ \

• Theory detached reflector

.1

\

\

'••

* Theory uned reflector i

.4

.5

i

i

.6

.7

>i .8

Wavelength/^ m

Figure 5.9 Experimental collection efficiency as a function of wavelength for an a-Si:H p-i-n solar cell with four different reflectors: flat Cr, flat Ag, textured tuned or textured detached. The results are normalised to the peak quantum efficiency at 0.55 /an. Shown as symbols are the theoretical results for a tuned and a detached reflector. After Deckman etal. (1984).

at these wavelengths without any increase in the cell thickness. Figure 5.9 shows the improvement that can be obtained by depositing an n(a-Si:H)-i(a-Si:H)-/?(a-SiC:H) cell on a textured, rather than flat, metal reflector. When the flat Cr or Ag reflector is replaced with a textured detached or tuned reflector, the quantum efficiency at 0.7 /an is improved from -10 to 20% of the peak value (observed at 0.55 fan) to -0.5 of the maximum for the tuned reflector, and -0.6 for the detached reflector. The tuned reflectors consist of metal evaporated on appropriately textured glass onto which the n-i-p cell is deposited. The detached reflectors have the same metal combination but with the addition of a conductive oxide (Sn0 2 , ZnO) several thousands of Angstroms thick for 'detaching' the n-i-p cell from the metal. Using textured substrates it has been possible to obtain short-circuit currents of about 18 mA cm" with a-Si:H of Ug -1.7 eV and Mayers much thinner than 1 /an.

Amorphous Silicon Solar Cells

217

150

<

AM-1.5, lOOmW/cm2

100

I

Voc(V) 12.53 Isc (mA) 130.1 F.F. 0.735 r, (%) 12.0

c

2

5

50 .i

"0

•

i

•

I

.

.

i

•

5

L.

10

15

Voltage/V Figure 5.10 Current-voltage characteristic of an integrated-type single-junction a-Si:H module under extended AM 1.5 illumination. The 12% efficient module has an area of 100 cm2 and consists of 14 cells in series. After Tsuda et al. (1993).

The high short-circuit currents possible with thin absorber layers greatly reduce the requirements on the carrier transport needed to obtain good fill factors in both the initial and the degraded states. By taking advantage of such optical enhancement, as well as ongoing improvements in materials and cell components, initial efficiencies in excess of 12% can now be obtained in single-junction a-Si:H cells and modules. The light I-V characteristics of a 100 cm2, single-junction module with an efficiency of 12% is shown in Fig. 5.10 (Tsuda et al, 1993). This panel consists of 14 seriesconnected cells, each having a FF in excess of 0.73 and a V^ of 0.9 V. Steady improvements in the stability of a-Si:H high-performance cells were also achieved by developing and then incorporating more stable a-Si:H materials into novel cell structures. The recent use of protocrystalline a-Si:H obtained by hydrogen dilution of the silane feedstock has led to a significant improvement in the stability of a-Si:H cells. This is illustrated in Fig. 5.11 for a p(a-SiC:HH(a-Si:H)-n(/ic-Si:H) cell having a 180 nm /-layer of such protocrystalline a-Si:H (Koval et al, 1999). Not only does the FF reach a degraded steady state in less than 100 hours of AM 1.5 sunlight illumination, but its degradation to a value of -0.70 corresponds to a light-induced diminution of only about 6%. The large interest in the improvement of not only the initial, but also the degraded, steady-state efficiencies, started the early development of multij unction cells along a two-track approach. In the first approach, improved stability of the degraded steadystate efficiency was sought by replacing a single thick a-Si:H cell by two thinner ones. In such a double-junction a-Si:H/a-Si:H tandem cell, the smaller thickness of the

218

C. R. Wronski and D. E. Carlson

0.68

0.1

1

10

100

AM 1.5 illumination time (hrs) Figure 5.11 Fill factor versus AM 1.5 illumination time at 25 C for a 180 n m thick p-i-n cell with a protocrystalline a-Si:H /-layer. After Koval el al. (1999).

individual cells lowers the requirements on carrier transport for efficient operation of each. The bottom cell is made somewhat thicker in order to compensate for the lower absorption of the longer wavelengths of the sunlight not absorbed in the top cell. The increase in thickness required to achieve equal short-circuit currents from the two cells is, however, not great since the optical enhancement, discussed earlier, is more efficient for the bottom cell. This approach has been successfully implemented in both cells and modules to obtain initial efficiencies greater than 10% and degraded steadystate efficiencies greater than 9% (Ichekawa et al., 1993). The second approach to tandem cells is the more traditional one where, in order to generate power more efficiently from the different parts of the solar spectrum, the stacked cells have different band gaps. Indeed, a-Si:H materials cover such a wide range of band gaps that solar modules composed of three-junction stacked cells are also practical. Thus not only a-Si:H/a-SiGe:H tandem cells but also a-Si:H/aSiGe:H/a-SiGe:H triple-junction cells have been developed. The optimisation of such cells centres on the ability to split the absorption of sunlight in the component cells in such a way as to maximise their carrier collection efficiencies, while at the same time generating equal currents in all the component cells. Because of the flexibility provided by the wide range of available band gaps, there are a number of different combinations for the absorber layer band gaps that lead to high-efficiency a-Si:H alloy triple-junction cells. Extensive work on a-Si:H/a-SiGe:H tandem cells has been carried out but, despite continuous progress, their performance is limited by the quality of the narrow bandgap a-SiGe:H materials, which is still markedly inferior to that of a-Si:H. To

Amorphous Silicon Solar Cells

219

overcome this limitation, special cell designs have been developed for maximising cell performance and at the same time minimising light-induced degradation. This includes band-gap profiling and constructing cells whose initial performance is not optimised, but rather account is taken of their individual degradation so as to achieve a maximum stabilised overall performance (Yang and Guha, 1992; Guha et al., 1989; Yang et al., 1994). This has led to tandem cells with absorber layers having a wide range of thicknesses and profiled band gaps.

0

0.5

1.0 Voltage/V

(a)

1.5

2.0

350

550 750 Wavelength/nm

950

(b)

Figure 5.12 (a) The current-voltage characteristic of an a-Si:H/a-SiGe:H tandem solar cell in its initial state of 12.61% efficiency, (b) The quantum efficiency as a function of wavelength in the AM 1.5 degraded steady state when the efficiency is 11.1%. After Yang and Guha (1992).

Figure 5.12 shows the characteristics of an a-Si:H/a-SiGe:H tandem solar cell in which the a-Si:H has a band gap of 1.75 eV and the a-SiGe:H one of 1.45 eV in its narrowest region. Figure 5.12a shows the initial current-voltage characteristic under AM1.5 illumination, when the cell efficiency is 12.61%. The short-circuit current of 10.67 mA cm" is limited by the smaller of the currents generated in the individual cells, the open-circuit voltage of 1.65 V is the sum of the voltages in the two individual cells, and the FF of 0.716 is determined by the degree of current matching between the two cells, their ability to collect carriers, and the quality of the interconnect between the two cells. Figure 5.12b shows the quantum efficiency as a function of wavelength for the cell after it has been degraded with AM 1.5 sunlight for 600 hours at 50 C to an efficiency of 11.1%, and how the currents in the two cells are generated by different parts of the solar spectrum. The decrease in cell efficiency between the initial and the degraded state is due to the decreases in V^ from 1.65 to 1.61 V, in the FF from 0.716 to 0.655, and in ix from 10.67 to 10.61 mA cm"2.

C. R. Wronski and D. E. Carlson

220

Triple-junction stacked cells offer even more efficient utilisation of the solar spectrum by using thin component cells to successfully absorb and collect the blue, green and red photons in cells with corresponding smaller band-gap j-layer materials (Yang et al., 1998). The highest efficiencies have been obtained using cell combinations in which: the top cell, which captures the blue photons, has an a-Si:H ilayer with a gap of ~ 1.8 eV; the middle cell, ideally suited for absorbing the green photons, has an i'-a-SiGe:H layer with 10-15 at.% Ge, and a gap of -1.6 eV; and the bottom cell, capturing the red and infrared photons, has an i-layer of a-SiGe:H with 40-50 at.% Ge and a gap of -1.4 eV. In the highest efficiency cells, the electrical and optical losses in the interconnects between adjacent cells have been minimised by using thin, highly doped n-p tunnel/recombination junctions. Also the currents in the bottom a-SiGe:H cells are optimised by band-gap profiling and highly efficient optical enhancement with textured detached reflectors. 2.5

o.o

f -2.5 <:

•

I . -5.0

1.0

J\

j Voc-2.344 V Jsc-8.99 mA/cm2 FF=0.722 Eff-15.2% A=0.25 cm2

/ ^

0.8

/ / / /

0.6

a

/

27.16^ mA/cms

1 9.02

0.4

y 9-isV.J

8.99\

0.2 -7.5 : -10.0-0.5

I

0.0

I

I

0.5

.

1....

1

1.0 1.5 Volts (V) (a)

-J

J

2.0

A.

.-

2.5

— i ^ i i _i^±»_

0.0 350

450

i

550 650 750 Wavelength (nm) (b)

.

i"—-~-

850

950

Figure 5.13 (a) Current-voltage characteristic and (b) quantum efficiency as a function of wavelength of a Uni-Solar a-Si:H/a-SiGe:H/a-SiGe:H triple-junction n-i-p cell with an STP efficiency of 15.2%. The short-circuit currents generated in the individual cells are indicated in (b), as is the total short-circuit current of 27.16 mA cm-2. After Yang et al. (1998).

Figure 5.13 shows the current-voltage and quantum efficiency characteristics of such an a-SiGe:H/a-SiGe:H/a-Si:H n-i-p stacked cell on stainless steel (Yang et al, 1998). The initial efficiency of this cell is 15.2%, as compared with the 12.61% of the tandem cell in Fig. 5.12. The introduction of the third cell increases the open-circuit voltage but lowers the short-circuit current relative to the tandem structures. This can be seen in Fig. 5.13a, where the open-circuit voltage of 2.34 V obtained from the three cells in series is much higher than the 1.65 V in Fig. 5.12a, but the short-circuit current of 8.99 mA cm-2 is significantly lower than the 12.6 mA cm-2 of the tandem cell since now each component cell absorbs a smaller fraction of the solar spectrum.

221

Amorphous Silicon Solar Cells

Figure 5.13b shows the utilisation of the solar spectrum in the three component cells and the excellent matching of the currents generated in the different cells. The total current of 27.16 mA cm"2 that would be generated if the cells were in parallel is higher than for the tandem cell. It is important to note that these currents are generated in cells that are thinner than those of other cell structures, thus minimising the requirements on carrier transport for efficient operation and assuring high fill factors as well as reducing the deleterious effects of the SWE. As noted above, significant progress has been made in the last two years in improving the efficiencies of amorphous silicon solar cells, particularly the degraded steady-state performance.

5.6 Solar cell structures 5.6.1 Introduction Amorphous silicon solar cells have been fabricated in a variety of different device structures. These include single-junction p-i-n and n-i-p devices as well as Schottky barrier cells, MIS cells and several different types of multijunction cells. A variety of different substrates such as glass, metal foils and plastics have been used. The major constraint on the substrate material is that it must be able to withstand temperatures of at least 150 C and not contaminate the a-Si films during deposition. Table 5.1

Amorphous silicon device structures and performance Efficiencies (%) Device structure (initial/stable)

Steel / Ag / ZnO / n-i-p 1 n-i*-p 1 n-i*-p 1ITO Steel / Ag / ZnO / n-i-p 1 n-i*-p 1 ITO Glass / Sn02 / p-i-n 1 p-i-n 1 ZnO / Ag Glass / Sn0 2 / p-i-n 1 ZnO / Ag Glass / SnQ2/p-i-n 1 p-i*-n 1 ZnO / Ag

15.2/13.0 14.4 / 12.4 12.5/9.0 12.0/8.9 11.6/10.0

Organisation USSC

ussc Fuji Electric Sanyo BP Solarex

Note: i* represents an a-SiGe /-layer.

Table 5.1 lists some of the a-Si device structures that have been made by different organisations and the best initial and stabilised conversion efficiencies achieved in the laboratory with these structures. The highest stabilised conversion efficiencies (13.0%) have been obtained with triple-junction structures (Guha et al., 1999). Even higher efficiencies (18.3%) have been recently achieved with a-Si/crystalline silicon

222

C. R. Wronski and D. E. Carlson

hybrid structures such as the Sanyo HIT cell (Tanaka et al., 1992). Since these devices are not thin-film structures, we shall not consider them here. Rather, we will review some of the more important commercial a-Si device structures in detail.

5.6.2 Single-junction structures Most of the low-wattage a-Si PV products used in consumer applications have been fabricated using single-junction p-i-n structures on glass substrates (Carlson and Catalano, 1989). As shown in Fig. 5.14, these single-junction devices are deposited on glass substrates coated with ~600 nm of a textured tin oxide that acts as the front electrical contact. The tin oxide is grown with a sub-micron surface texture that is determined by the grain size of the crystallites, and this texture improves the device performance by scattering and trapping the incident light.

,

I :„.,..,;..,...

light .

i •

glass

;

; .

.,

.

, "1

^silicon dioxide 'textured tin oxide - p-layer • a-Si /-layer ' n-layer "^aluminium

Figure 5.14 substrate.

A schematic of a typical commercial single-junction device structure fabricated on a glass

The p-i-n diode is formed by first depositing about 10 nm of a boron-doped amorphous silicon-carbon alloy (the a-SiC:H p-layer), then about 250 nm of undoped or intrinsic a-Si:H (the i-layer), and finally about 30 nm of phosphorus-doped amorphous silicon (the n-layer). The back contact is usually formed by sputter depositing about 400 nm of aluminium. Single-junction a-Si devices can also be made using an n-i-p structure on metal substrates, but very few commercial products have been made in this configuration. The major drawback to single-junction a-Si solar cells is that they typically degrade by about 22-25% in sunlight before stabilising, and the steady-state performance of commercial modules is generally only in the 4-6% range.

Amorphous

5.6.3

223

Silicon Solar Cells

Tandem structures

In recent years, several companies have introduced multijunction a-Si PV modules since they exhibit higher stabilised conversion efficiencies than single-junction devices. The theoretical conversion efficiency of multijunction cells is higher than that of single-junction cells because they can make use of more of the solar spectrum at a higher net output voltage. The theoretical efficiency of an ideal single-junction cell is about 28%, while that of an ideal tandem device is about 36% and that for an ideal triple-junction device is about 42% (Bennett and Olsen, 1978).' In the case of aSi cells, another factor favouring multijunction devices is the better stabilised performance associated with the use of thinner /-layers (Hanak and Korsun, 1982). light

glass

--silicon dioxide ^textured tin oxide -p-layer -a-Si /-layer -tunnel junction -a-SiGe /-layer ' n-layer "zinc oxide aluminium

Figure 5.15

A schematic of a tandem device structure fabricated on a glass substrate.

Most of the commercial tandem modules have the device structure shown in Fig. 5.15 (Carlson et al., 1996). The front junction is similar to that of the singlejunction device described above, with an Mayer optical gap of-1.78 eV. However, in the case of the tandem structure, another p-layer is deposited after the n-layer of the front junction. This nip junction is often referred to as a 'tunnel' junction, but it actually functions as a recombination junction in electrically connecting the two p-i-n junctions of the tandem structure in series. The second or back junction is formed by depositing an undoped amorphous silicon-germanium alloy (a-SiGe:H) on the second p-layer and then a second a-Si:H n-layer. In the BP Solarex tandem structure, the aSiGe:H /-layer has a graded optical gap2 that ranges from -1.45 eV to -1.70 eV and contains on average -40 at.% germanium. The /-layers of the front and back junctions

1

These numerical values are sensitive to the solar spectrum used in the calculation, and to whether series connection is assumed or not. 2 The Ge content of a-SiGe:H layers may be graded in an attempt to enhance carrier collection by band-gap grading.

C. R. Wronski and D. E. Carlson

224

are both -150 nm thick, and are chosen so that the currents from the two junctions are almost the same in the degraded steady state. In fact, the best stabilised performance is usually obtained when the current from the back junction is slightly larger than that of the front junction in the degraded steady state. The back contact is formed by first depositing about 100 nm of ZnO by LPCVD or by sputtering and then coating this layer with about 300 nm of aluminium by magnetron sputtering. In general, tandem devices take longer to make than single-junction devices, but exhibit better stabilised performances. BP Solarex tandem PV cells and modules typically exhibit light-induced degradation of 13-17%.

5.6.4 Triple-junction structures As Table 5.1 shows, the highest initial and stabilised conversion efficiencies for a-Si solar cells have been obtained using triple-junction structures (Guha et al., 1999). The best triple-junction cells have been fabricated on stainless steel foil substrates coated with layers of textured silver and zinc oxide, as shown in Fig. 5.16. (In commercial product, the back reflector consists of sputtered layers of aluminium and zinc oxide.) A phosphorus-doped a-Si:H w-layer (about 20 nm thick) is deposited on the zinc oxide, and then an a-SiGe:H /-layer with a graded germanium concentration is deposited on the n-layer. The triple-junction structure contains two 'tunnel' junctions each consisting of -10 nm of boron-doped microcrystaliine Si:H and -10 nm of phosphorus-doped a-Si:H. light

'Silver grid

_T~L indium tin oxide

-p-layer a-Si /-layer - tunnel junction a-SiGe /-layer • tunnel junction a-SiGe /-layer zinc oxide textured silver

- n-layer

stainless steel

Figure 5.16

A schematic of a triple-junction device structure fabricated on a stainless steel substrate.

225

Amorphous Silicon Solar Cells

The middle junction also consists of an a-SiGe:H /-layer with a graded germanium concentration, but the average germanium content is reduced as compared to the first a-SiGe:H junction (the back junction). The front junction contains an Mayer of undoped a-Si:H deposited by glow discharge decomposition containing of silane heavily diluted in hydrogen. (As mentioned earlier, hydrogen dilution is now widely used to deposit high-quality films of both a-Si:H and a-SiGe:H.) The thicknesses of the top, middle and back junctions in a triple-junction structure are about 100, 110 and 130 nm, respectively. The p-layer of the top junction contains -10 nm of borondoped microcrystalline Si:H. The top contact is formed by depositing an antireflecting layer of indium tin oxide and then a current-collecting silver grid.

5.7 PV modules 5.7.1 The commercialisation effort A number of companies are involved in developing and commercialising a-Si PV technology. Table 5.2 lists some of the more significant commercialisation efforts, along with some information on the types of PV modules produced. There are also many other organisations such as universities and government laboratories involved in research and development on a-Si alloys and devices for PV applications. In addition, there are a large number of other organisations performing research and development and commercialising a-Si technology for other applications such as thinfilm transistors (for liquid crystal displays), sensors and other electronic devices. Table 5.2

Companies commercialising a-Si PV technology Best stabilised efficiency (module Company Module configuration area)

ASE GmbH Fuji Electric Kaneka Intersolar Iowa Thin Films Sanyo BP Solarex USSC

-5-6% (0.6 m2) 7.3% (0.32 m2) 8.1% (0.414 m2) - 5 % (0.28 m2) ~6%(~0.1m 2 ) 9.5% (0.12m2) 8.1% (0.36 m2) 10.1% (0.09 m2)

Same gap tandem on glass Same gap tandem on plastic Single-junction on glass Single-junction on glass Same gap tandem on plastic Tandem on glass Tandem on glass Triple-junction on steel

226

C. R. Wronski and D. E. Carlson

Commercial production of a-Si solar cells was initiated by Sanyo Corporation in 1980 for use in solar-powered calculators. During the 1980s and early 1990s, manufacturing plants were set up in various parts of the world by companies such as Arco Solar, Chronar, Energy Conversion Devices, Fuji Electric, Sanyo, Sharp, Solarex, Solems and Utility Power Corporation. Most of these plants were designed to produce single-junction a-Si solar cells and had annual capacities of about 1 MWP. More recently several companies have built multi-megawatt plants to manufacture a-Si PV modules, and others have announced plans to build even larger-scale manufacturing facilities (see Table 5.3). Table 5.3

Large-scale a-Si PV manufacturing plants

Company

Plant capacity (MWp/yr)

Date operational

Module configuration

BP Solarex USSC Canon Kaneka Sharp

10 5 10 20 6

1997 1997 1998 2000 1999

Tandem on glass Triple-junction on steel Triple-junction on steel Single-junction on glass Tandem on glass

5.7.2 Modules on glass substrates Most a-Si PV modules are fabricated on glass substrates. Glass is a relatively inexpensive, widely used material with excellent long-term resistance to weathering and environmental effects. While it is susceptible to breakage if not handled properly, it can be strengthened by thermal tempering or chemical ion exchange processes. In recent years, some glass companies have started commercial production of glass coated with textured tin oxide. Figure 5.17 shows the layout of the BP Solarex a-Si PV manufacturing plant in Toano, Virginia (Forrest, 1997). The substrates are 3 mm thick plates of float glass (26" x 48") coated with textured tin oxide, which are currently supplied by AFG Corporation. The edges of the plates are seamed to control the size and to also remove microcracks. The plates are washed before applying a silver frit to act as bus bars for electrical contacts. The frit is then cured by heating the plates to a temperature of about 500 C in a belt furnace.

227

Amorphous Silicon Solar Cells

Module load station Solar simulator |

|

Bed of nails I

j

Ultrasonic bath |" [ Isolation laser | Metal laser

a-Silaser

-D-O

1 |

0

Wiring, finishing

Encapsulation

EVA setup

Glass seamer

|

a

Sputtering system |

Back plate alignment

l

Washer

Cure furnace

HD n \

Frit dispenser

i O Tin oxide laser '« Washer

Vertical a-Si/ZnO deposition system

Figure 5.17

Schematic of the BP Solarex TF1 plant in Toano, Virginia. After Forrest (1997).

The tin oxide is then scribed into a series of parallel strips ~9 mm wide using a Nd-YAG laser that is frequency doubled to produce a green laser beam about 15 microns in diameter. For high-voltage modules (V^ > 200 V), the tin oxide is scribed into 143 separate strips running along the short dimension of the plate that define the individual cells. The plates are washed before being loaded into a multi-chamber deposition system that coats the plates with all the semiconductor layers shown in Fig. 5.15, including the ZnO layer. Each chamber of this system can hold four plates at a time in a vertical orientation. Seven of the twelve chambers are used to deposit amorphous and microcrystalline Si:H alloys by DC plasma-enhanced chemical vapour deposition (PECVD), one chamber is used to deposit ZnO by low-pressure CVD, and the other chambers act as buffers to minimise cross contamination. After exiting the vertical deposition system, the a-Si and ZnO layers are scribed with another Nd-YAG laser. These scribes are located several tens of microns to one side of the tin oxide scribes (see Fig. 5.18) and are performed at a lower power density so as to not scribe the tin oxide layer itself. The plates are then coated with about 300 nm of aluminium by magnetron sputtering before performing another laser scribe that removes the a-Si, the ZnO and the Al in a region to one side of the earlier scribes. This last scribe (the metal laser scribe) completes the series connection as shown in Fig. 5.18, since the front contact of each strip cell is connected in series to

228

C. R. WronskiandD.

E. Carlson

silicon dioxide tin oxide

back contact

Figure 5.18 Schematic of the BP Solarex series-connected tandem device structure. The series connection is made by the back («) contact of one cell to the tin oxide (p) contact of the adjacent cell. The indentation is a result of sequential scribing and represents the thickness of the cell that was removed to form (isolated) adjacent cells.

the back contact of the next adjacent cell by the aluminium deposited in the amorphous silicon scribe. Next, a final laser scribe is made at a relatively high power around the perimeter of the module to ensure electrical isolation. The modules are then cleaned in an ultrasonic bath to remove all debris before passing to a bed-of-nails station that applies a reverse bias to electrically cure cells that are excessively leaky due to small shorts (Nostrand and Hanak, 1979). After electrical curing, the performance of the modules is tested using a solar simulator. The modules are then completed by encapsulating a back plate of float glass to the front plate with ethyl vinyl acetate (EVA), attaching lead wires and mounting the module in a frame. In some applications, a frame is not needed or is supplied by the customer. The completed modules then undergo a final power test before being shipped to customers. This manufacturing process is capable of producing tandem modules with relatively high yields, as shown in Fig. 5.19 for a run of 4 ft2 modules made in a pilot manufacturing mode. In this pilot run of more than 160 modules, the average initial conversion efficiency was 8.9% with only a few modules falling outside the control limits of 8.1% and 9.8%. (Single-junction modules were also produced with high electrical yields (-95%) over a period of about 5 years at the Solarex facility in Newtown, PA before being phased out in 1998.)

Amorphous Silicon Solar Cells

229

I Figure 5.19

100 Run number A run chart showing the initial efficiency of 4 ft2 tandem modules vs. run number.

5.7.3 Modules on metal substrates Table 5.4 Manufacturing steps for USSC triple-junction modules on steel substrates Process step Process step description no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14

The roll of stainless steel foil is washed, rinsed and dried. Al and ZnO are sequentially deposited by magnetron sputtering. a-Si and ^c-Si alloy layers are deposited by RF PECVD. ITO is deposited by magnetron sputtering. The roll is cut into slabs. Slabs are processed to define cell size. Slabs are passivated to remove shunts. Slabs are tested to determine device quality. Conductive pads and grid wires are applied. Slabs' are cut into predetermined cell sizes. Cells are interconnected. Cell block is laminated. Modules are framed and junction boxes are added. Modules undergo highpot and performance tests.

230

C. R. Wronski and D. E. Carlson

Table 5.4 shows the various steps in the USSC manufacturing process for producing a-Si triple-junction modules on stainless steel foil (Guha et al, 1999). A roll of stainless steel foil is first washed in a wash machine, where it is also rinsed in deionised water and dried in an infrared oven. The roll is then loaded into a magnetron sputtering machine where aluminium and zinc oxide are sequentially deposited to form a highly reflective back contact. In the next step, the roll is loaded into a ninechamber, in-line PECVD system where the a-Si and pic-Si alloy layers are deposited continuously on the moving foil. The roll is then loaded into another magnetron sputtering machine where a layer of indium tin oxide (ITO) is deposited. This ITO layer acts as both a top electrical contact and an antireflection coating. The roll of foil is then moved to a semi-automated module assembly area where the foil is cut into 9.4" x 14" slabs. The slabs are processed to remove shunts and to define the specific cells before attaching electrodes and cutting the cells. The individual cells are then interconnected and laminated into a module. The module is completed by adding a frame and a junction box. The modules are subjected to a high-voltage test, and the output power is measured under simulated sunlight before shipping to customers. USSC reports that the stabilised aperture-area efficiency for these products is about 7.5% (Guha et al, 1999).

5.7.4

Modules on plastic substrates

The various steps in the manufacturing process used by Iowa Thin Film Technologies for producing a-Si tandem modules on plastic substrates are listed in Table 5.5 (Braymen et al, 1999). As a substrate material, Iowa Thin Film uses a polyimide plastic film that can tolerate temperatures of about 250 C and does not outgas significantly in a vacuum. They deposit an Al film as a back contact layer and then a same band-gap, a-Si tandem structure. A Nd-YAG laser is used to scribe through both the Al and the a-Si layers to define the individual cell segments. They then use screen-printing to place an insulating ink in the laser scribe and also a second strip of insulating ink in a region to one side and parallel to the laser scribe. Next, ZnO is sputtered onto the substrate, and then a conductive (Ag) ink is placed in the region over the laser scribe (which was previously filled with the insulating ink). A laser is then used to scribe the ZnO in the region above the second insulating ink strip. The series interconnection is completed by using a laser to weld or fuse the Ag ink to the bottom Al contact of the adjacent cell.

Amorphous Silicon Solar Cells

231

Table 5.5 Manufacturing steps for tandem modules made by Iowa Thin Film Technologies on plastic substrates Process Process step description step no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14

The roll of polyimide is washed. Al is sputtered deposited on the polyimide. a-Si layers are deposited by RF PECVD. Both the Al and the a-Si layers are scribed by a laser. Insulating ink is screen-printed in strips in the interconnect region. ZnO is deposited by sputtering. Ag ink grid lines are screen-printed in the interconnect region. The ZnO is laser scribed. A laser is used to weld the Ag ink to the Al bottom contact. Bus bars are attached to the modules. Modules are laminated. Modules are cut from the roll. Modules are framed and wired. Modules undergo performance tests.

Bus bars are attached to the modules while they are still on the roll. The modules are laminated to a Tedlar sheet using a roll-based laminator and then cut into individual modules before being framed, wired and tested. Amorphous silicon modules made on plastic substrates are lightweight and flexible and are used in a number of consumer applications.

5.8

Manufacturing costs

In past years, a number of organisations have estimated that the total cost of manufacturing a-Si PV modules should be less than $1/WP once the production volume of the plants reaches about 10 MWp per year (Carlson, 1989). More recently, Woodcock et al. (1997) have analysed the manufacturing costs for the three leading thin-film PV technologies—a-Si, copper indium diselenide (CIS), and cadmium telluride (CdTe)—at production volumes of 60 MWP /yr. As shown in Fig. 5.20, the projected manufacturing costs are significantly less than $1/Wp in each case. It is interesting to note that at a production volume of 60 MWp/yr, the materials costs are more than half of the total manufacturing costs in each case. As improvements are

232

C. R. Wronski and D. E. Carlson

Figure 5.20 Projected manufacturing costs for a-Si, CIS and CdTe thin film PV modules made in plants with 60 MWp /year capacity. Source: Woodcock el al. (1997).

made in thin-film PV manufacturing technology and as the plants become even larger, the labour costs and the fixed costs should decline even further so that materials costs will become the major factor limiting further cost reductions. For a-Si tandem modules, the major cost elements are currently the framing, the encapsulation and the tin oxide-coated glass. The framing and encapsulation alone account for about 37% of the total manufacturing cost for 8 ft2 tandem modules, while the tin oxide coated glass substrate and back glass plate constitute ~23% of the total cost. The semiconductor feedstock materials constitute much smaller percentages, with germane accounting for about 13% and silane only about 2% of the total cost. At present the utilisation of feedstock gases in commercial PECVD reactors is poor, with no more than a few percent of the silicon or germanium winding up in the a-Si alloy films. If the utilisation could be improved to about 70% or better, then the cost for the semiconductor materials would be less than $0.02/Wp.

5.9

Long-term reliability

All commercial PV modules are periodically subjected to a series of accelerated environmental tests to assure long-term reliability. BP Solarex a-Si tandem PV modules are subjected to the series of tests shown in Fig. 5.21. Some of the more critical tests are the wet hi-pot test, the thermal cycle test, the temperature humidity test and the light-soak test.

233

Amorphous Silicon Solar Cells I 8 modules [

Visual inspection

T

Performance @ STC

T

r

' GmtroL module

VVL'

11 JHU mst

K~

7^

Annealing

current

3Z

M^

Light soak

200 Thermal cycles

Damp~heat

I 2 modules Outdoor exposure

UVtest

^

3Z

Annealing

^_

50 thermal cycles

Hot spot endurance

JZ

10 Humidity freeze cycles 1 module

I 1 inodule

Robustness, of terminations

\

Mechanical load

Twist test

ir^x

Wet leakage

3Z

Wet hi-pot test

31

Light soaking

Figure 5.21

The environmental test sequence used for BP Solarex thin-film PV modules.

Since some modules will be used in high-voltage, grid-connected applications, they must be able to pass the wet hi-pot test. In some grid-connected systems, the array voltage can be greater than 1000 V so modules must exhibit a leakage current of less than 50 /lA while wet and with an applied voltage greater than twice the array voltage plus 1500 V. Because of this stringent requirement, BP Solarex tandem modules are encapsulated in EVA between two sheets of glass. This type of module passes all the accelerated environmental tests shown in Fig. 5.21. Two critical tests that check the integrity of the module against moisture penetration are the thermal cycle test and the humidity freeze test. The former involves 50 cycles between -40 C and 90 C, and the latter 10 cycles between -40 C and 85 C at 85% humidity.

234

C. R. Wronski and D. E. Carlson

600

700

800

Wavelength (nm) Figure 5.22

Spectral response of a tandem cell before and after 20 hours at 180 C.

The BP Solarex tandem structure is quite stable with respect to thermal degradation. Unlike earlier a-Si solar cells with an Al rear contact, which exhibited degradation effects at temperatures of -130C (Willing et al., 1987), there are no significant thermal degradation effects associated with the Sn0 2 or ZnO contacts used in the present structure. However, the tandem cells will start to show some degradation when heated for prolonged times at temperatures on the order of 180 C. Figure 5.22 shows the change in spectral response of a tandem cell on heating for 20 hours at 180 C in the dark. As observed in earlier work (Carlson and Rajan, 1995), the a-Si front junction exhibits a loss in quantum efficiency, mainly in the shortwavelength regime, due to hydrogen motion. The a-SiGe back junction shows a loss in quantum efficiency that is more evenly distributed over its response spectrum. After 20 hours at 180 C, the efficiency decreased by only 2.2%. The thermal degradation increases dramatically with further increases in temperature. After 150 minutes at 220 C, the conversion efficiency typically falls about 15% with about half the loss due to degradation in the fill factor. Since the activation energy for thermal degradation in the dark is -1.7 eV (Carlson and Rajan, 1995), this degradation is negligible under normal operating conditions. Light soaking of a-Si modules is another critical test since all a-Si devices exhibit light-induced degradation arising from the Staebler-Wronski effect (Staebler and Wronski, 1977). This degradation can be enhanced by contaminants in the Mayers or at interfaces, and some recent work (Carlson and Ganguly, 2000) shows that irreversible light-induced degradation can be caused by trace amounts of boron in the Mayer. BP Solarex tandem modules typically exhibit about 13-17% degradation on

235

Amorphous Silicon Solar Cells 4

3

&

2

1

0 0

5

10

15

2D

Figure 5.23 A histogram showing the degradation experienced by twelve 4 ft2 tandem modules after exposure to 600 hours of simulated sunlight.

light soaking, but this can increase to 20-25% if contaminants are present. Figure 5.23 shows the distribution of the degradation suffered by twelve BP Solarex a-Si tandem modules (4 ft2) subjected to simulated sunlight for 600 hours. On average, these modules degraded by only about 12.9%. In general, light-induced degradation of a-Si PV modules saturates or reaches a steady state after about 10 to 10 hours of exposure to sunlight, depending on the deposition conditions. Saturation occurs in ~102 hours for a-Si single-junction modules grown in discharge atmospheres that contain silane heavily diluted in hydrogen. Once the modules reach steady state, the conversion efficiency then exhibits normal seasonal variations due to changes in the average ambient temperature and seasonal changes in the solar spectrum.

S.10

Environmental issues

Photovoltaic solar energy is viewed by many as an ideal way to produce power from a virtually inexhaustible energy source without noise or pollution. However, there are environmental issues that must be addressed to assure a trouble-free future for PV. The entire process of mining and refining raw materials, manufacturing PV modules and handling of obsolete product must be designed not only for low cost, but also for the environment. If module processing requires the use of toxic materials, then systems and procedures must be established to minimise the risk to employees. In the manufacture of a-Si PV devices, BP Solarex uses toxic doping gases such as diborane and phosphine only in a diluted form (-1-20 vol.% in silane). Trimethylboron (-1-5 % in

236

C. R. Wronski and D. E. Carlson

silane) is also used as a p-type dopant source and is less toxic than diborane. Silane is pyrophoric, so if a leak develops the dopant gas will be oxidised in the flame and a silicate glass powder will be formed, thus reducing the toxicity hazard. The silane and germane feedstock gases at the BP Solarex TF1 plant in Virginia are stored in an outdoor holding area and fed into the facility through stainless steel pipes. All exhaust gases are passed through a burn box and the powder is collected in a bag house for disposal. The powder consists mainly of silicon dioxide fused with small amounts of oxides of germanium, boron and phosphorus. Since all module interconnections are made using lasers, there are no wet chemicals such as acids or solvents used in the BP Solarex manufacturing process for tandem modules. Thus, there are no harmful waste products or effluents produced in the manufacturing process. In addition, since a-Si PV modules do not contain any toxic materials, there are no environmental risks associated with fires or with longterm disposal in landfills.

5.11

Challenges for the future

While a-Si photovoltaics has the potential to become a major source of low-cost electricity worldwide in the next few decades, the fulfillment of this potential depends on continued progress in improving the stabilised performance, reducing the total manufacturing cost and establishing the infrastructure necessary to create large-scale markets. There is tremendous potential to reduce the total system cost of thin-film PV arrays significantly by integrating them into new buildings. This can be accomplished by designing the modules to function as PV roofs or windows. The PV system cost is reduced since there is no need for land and support structures, and most of the material and labour can be credited against the cost of the building, as discussed in Chapter 15. Improvements in stabilised performance will require a better understanding of the growth kinetics, the nature of the intrinsic defects and the role of hydrogen in a-Si alloys. While a-Si alloys can be grown by a number of different techniques such as DC PECVD, RF PECVD, electron cyclotron resonance remote plasmas, hot-wire CVD, photo-CVD and sputtering in argon-hydrogen atmospheres, it is not clear what precursors or conditions are necessary to assure the best-quality films. Moreover, there is still no consensus on the microscopic origin of the Staebler-Wronski effect, or on the role of hydrogen in determining the metastability or doping efficiency of aSi alloys. Thus, there is a clear need for further experimental and theoretical work on understanding the growth and the microscopic nature of the defects in a-Si alloys.

Amorphous Silicon Solar Cells

237

While today's multijunction a-Si PV cells degrade by only -10-17% under prolonged light exposure, the efficiency could be greatly improved if the lightinduced degradation could be significantly reduced or eliminated. This is because the Mayers of current a-Si cells have to be quite thin to minimise the degradation. Conversion efficiencies of 17-20% might be possible with multijunction a-Si-based structures if the a-Si alloy j'-layers could be made thicker while remaining stable. While improving the stabilised performance would lead to lower manufacturing costs on a $/Wp basis, increasing the throughput will lower the costs associated with labour and overhead. The throughput of a-Si PV manufacturing plants is limited mainly by the deposition rate of the a-Si alloys. Currently, these are deposited at rates of about 0.1 nm s"' in most commercial manufacturing processes. Since about half of the cost of an a-Si PV manufacturing plant is associated with the cost of the a-Si deposition machine, a doubling of the throughput of that machine would significantly lower the capital cost of the plant on a $ per Wp of capacity basis. A number of organisations are investigating the rapid deposition of a-Si alloys by techniques such as hot-wire CVD, high frequency PECVD etc. The major challenge is to increase the deposition rate without increasing the concentration of the intrinsic and the metastable defects that lead to a reduction in stabilised performance. Another challenge is to understand the additional defects created when alloying aSi with germanium or carbon. The defect density increases when either of these is added to a-Si, and changes are observed in the kinetics of the light-induced degradation. It appears that additional defects associated with clustering or hydrogen complexes are introduced, but there is no detailed microscopic model or theory to account for these effects. Improvements in the microstructure of a-Si alloys should lead to further improvements in stabilised module performance. Significant progress has been made in recent years in developing high-quality films of more ordered a-Si:H, protocrystalline and microcrystalline silicon, and even thin polycrystalline silicon films. These materials have some significant advantages over other thin-film materials for PV applications. Silicon is very abundant and does not pose any environmental hazards. Moreover, especially in the case of a-Si alloys, the films are very easy to deposit with good uniformity over large areas. In summary, while amorphous and microcrystalline silicon alloy materials are complex and poorly understood, there are a large number of talented scientists and engineers in laboratories around the world who are working to make the promise of low-cost photovoltaic electricity a reality. It is this large pool of experienced,and talented people that makes it highly probable that this goal will be achieved in the next few decades.

238

C. R. Wronski and D. E. Carlson

References Bennett A. and Olsen L. C. (1978), 'Analysis of multiple-cell concentrator/ photovoltaic systems', Conf. Record 13th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 868-873. Branz H. M. and Silver M. (1990), 'Potential fluctuation due to inhomogeneity in hydrogenated amorphous silicon and the resulting charged dangling bond defects', Phys. Rev. B. 42, 7420-7428. Braymen S., Grimmer D., Jeffrey F., Martens S., Noack M., Scandrett B. and Thomas M. (1999), 'Monolithic amorphous silicon modules on continuous polymer substrates', AIP Conf. Proc. 462, 73-740. Brodsky M. H., Frisch M. A., Ziegler J. F. and Lanfard W. A. (1977), 'Quantitative analysis of hydrogen in glow discharge amorphous silicon', Appl. Phys. Lett. 30, 561-563. Carlson D. E. and Wronski C. R. (1976), 'Amorphous silicon solar cells', Appl. Phys. Lett. 28, 61'1-673. Carlson D. E., Wronski C. R., Triano A. and Daniel R. E. (1976), 'Solar cells using Schottky barriers on amorphous silicon', Conf. Record 12th. IEEE Photovoltaic Specialists Conf, Baton Rouge, IEEE Press, Piscataway, 893-895. Carlson D. E. and Wronski C. R. (1979), 'Amorphous Silicon Solar Cells', in Topics in Applied Physics, Vol. 36, Brodsky M. H., ed., Springer-Verlag, Berlin, 289329. Carlson D. E. (1984), 'Solar energy conversion', in Topics in Applied Physics, Vol. 55, Joannopoulos J. D. and Lucovsky G., eds., Springer-Verlag, Berlin, 203-244. Carlson D. E. (1989), 'Low-cost power from thin film photovoltaics', in Electricity, Johansson T. B., Bodlund B. and Williams R. H., Lund University Press, Lund, Sweden, 595-626. Carlson D. E. and Catalano A. (1989), 'Improving the performance and reducing the cost of amorphous silicon solar cells', Optoelectronics 4, 185-193. Carlson D. E. and Rajan K. (1995), 'The degradation of amorphous silicon solar cells at elevated temperatures', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S Stephens & Associates, Bedford, 617-619. Carlson D. E., Arya R. R., Bennett M., Chen L. F., Jansen K., Li Y. M., Newton J., Rajan K., Romero R., Talenti D., Twesme E., Willing F. and Yang L. (1996), 'Commercialization of multijunction amorphous silicon modules', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 1023-1028.

Amorphous Silicon Solar Cells

239

Carlson D. E. and Rajan K. (1998), 'Evidence for proton motion in the field-induced recovery of light-induced degradation in amorphous silicon solar cells', J. Appl.Phys. 83, 1726-1729. Carlson D. E. and Ganguly G. (2000), to be published. Chittick R. C , Alexander J. H. and Sterling, H. E. (1969), 'The preparation and properties of amorphous silicon', J. Electrochem. Soc. 116,77-81. Collins R. W. (1994), 'Real time spectroscopic ellipsometry studies of the nucleation, growth and optical functions of thin films. Part 1: tetrahedrally bonded materials', in Physics of Thin Films, Vol. 19, Academic Press, New York, 49-125. Collins R. W. and Vedam K. (1995), 'Optical properties of solids', in Encyclopedia of Applied Physics, Vol. 12, Trigg G. L., ed., VCH Publishers, New York, 285-336. Collins R. W. and Fujiwara H. (1997), 'Growth of hydrogenated amorphous silicon and its alloys', Current Opinion in Solid State & Material Science 2, 417-424. Dawson R. M., Li Y., Gunes M., Nag S., Collins R. W., Bennett M. and Wronski C. R. (1992), 'Optical properties of the component materials in multijunction hydrogenated amorphous silicon-based solar cells', Proc. 11th. European Photovoltaic Solar Energy Conf, Montreux, Harwood Academic Publishers GmbH, Chur, Switzerland, 680-683. Deckman H., Wronski C. R. and Yablonovitch E. (1984), 'Optical enhancement of solar cells', Conf. Record 17th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 955-960. Forrest H. (1997), 'Efforts on commercialization of amorphous thin-film technology', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S Stephens & Associates, Bedford, 2018-2020. Fritzsche H. (1997), 'Search for explaining the Staebler-Wronski effect', Mat. Res. Soc. Symp. Proc. 467, 19-31. Ganguly G. and Matsuda A. (1996), 'Role of hydrogen dilution in improvement of aSiGe:H alloys', J. Non-Cryst. Solids 198-200, 559-562. Guha S., Yang J., Pawliklewicz T., Glatfelter T., Ross R. and Ovshinsky S. R. (1989), 'Band-gap profiling for improving the efficiency of amorphous silicon alloy solar cells', Appl. Phys. Lett. 54, 2330-2332. Guha S. (1996), 'Amorphous silicon alloy solar cells and modules—opportunities and challenges', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 1017-1022. Guha S., Yang J., Banerjee A., Hoffman K. and Call J. (1999), 'Manufacturing issues for large volume production of amorphous silicon alloy photovoltaic modules', AIP Conf. Proc. 462, 88-93.

240

C. R. Wronski and D. E. Carlson

Hanak J. J. and Korsun V. (1982), 'Optical stability studies of a-Si:H solar cells', Conf. Record 16th. IEEE Photovoltaic Specialists Conf., San Diego, IEEE Press, Piscataway, 1381-1383. Hirose M. (1984), 'Glow discharge', in Semiconductors and Semimetals Vol. 21A, Pankove J. I., ed., Academic Press, Orlando, 9-39. Ichikawa Y., Fujikake S., Takayama T., Saito S., Ota H., Yoshida T., Ihara T. and Sakai H. (1993), 'Large-area amorphous silicon solar cells with high stabilized efficiency and their fabrication technology', Conf. Record 23rd. IEEE Photovoltaic Specialists Conf., Louisville, IEEE Press, Piscataway, 27-33. Jackson W. B., Biegelsen D. K., Nemanich R. J. and Knights J. C. (1983), 'Optical absorption spectra of surface or interface states in hydrogenated amorphous silicon', Appl. Phys. Lett. 42,105-107. Jiao L., Liu H., Semoushikina S., Lee Y. and Wronski C. R. (1996a), 'Importance of charge defects in a-Si:H materials and solar cell structures', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 1073-1076. Jiao L., Liu H., Semoushikina S., Lee Y. and Wronski C. R. (1996b), 'Initial, rapid light-induced changes in hydrogenated amorphous silicon materials and solar cell structures: the effect of charged defects', Appl. Phys. Lett. 69, 3713-3715. Kamei T., Hata N., Matsuda A., Uchiyama T., Amano S., Tsukamoto K., Yoshioa Y. and Hirao T. (1996), 'Deposition and extensive light soaking of highly pure hydrogenated amorphous silicon', Appl. Phys. Lett. 68, 2380-2382. Koh J., Lee Y., Fujiwara H., Wronski C. R. and Collins R. W. (1998), 'Optimization of hydrogenated amorphous silicon p-i-n solar cells with two-step i layers guided by real-time spectroscopic ellipsometry', Appl. Phys. Lett. 73, 1526-1528. Koval R. J., Koh J., Lu Z., Lee Y., Jiao L., Collins R. W. and Wronski C. R. (1999), 'Performance and stability of Si:H p-i-n solar cells with i layers prepared at the thickness-dependent amorphous-to-microcrystalline phase boundary', Appl. Phys. Lett. 75, 1553-1555. Lee Y., Jiao L., Liu H., Lu Z., Collins R. W. and Wronski C. R. (1996), 'Stability of a-Si:H solar cells and corresponding intrinsic materials fabricated using hydrogen diluted silane', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 1165-1168. Lee Y., Ferlauto A. and Wronski C. R. (1997), 'Contributions of bulk, interface and build-in potential to the open-circuit voltage of a-Si:H solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 683-686.

Amorphous Silicon Solar Cells

241

Lee Y., Ferlauto A.S., Lu Z., Koh J., Fujiwara H., Collins R. W. and Wronski C. R. (1998), 'Enhancement of stable open circuit voltage in a-Si:H p-i-n solar cell by hydrogen dilution of pli interface regions', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, European Commission, 940-943. Lu Y., Kim S., Gunes M., Lee Y., Wronski C. R. and Collins R. W. (1994), 'Process property relationships for a-Sii_^C^:H deposition: excursions in parameters space guided by real time spectroellipsometry', Mat. Res. Soc. Symp. Proc. 336, 595-600. Lu Z., Jiao H., Koval R., Collins R. W. and Wronski C. R. (1999), 'Characteristics of different thickness a-Si:H/metal Schottky barrier cell structures—results and analysis', Mat. Res. Soc. Symp. Proc. 557, 780-785. Nostrand G. E. and Hanak J. (1979), 'Method of removing the effects of electrical shorts and shunts created during the fabrication process of a solar cell', U.S. Patent No. 4,166,918. Powell M. J. and Dean S. C. (1993), 'Improved defect pool model for charged defects in amorphous silicon', Phys. Rev. B 48, 10815-10827. Roxlo C , Abeles B., Wronski C. R., Cody G. D. and Tiedje T. (1983), 'Comment on optical absorption edge in a-SiiH/, Solid State Commun. 47, 985-987. Shah A., Kroll U., Keppner H., Meier J., Torres P. and Fischer D. (1996), 'Potential of VHF-plasma for low-cost production of a-Si:H solar cells', Proc. 9th. Int. Photovoltaic Sci. Eng. Conf., 267-270. Spear W. E. and LeComberP. G. (1975), 'Substitutional doping of amorphous silicon', Solid State Commun. 17, 1193-1196. Staebler D. L. and Wronski C. R. (1977), 'Reversible conductivity change in discharge produced amorphous silicon', Appl. Phys. Lett. 31, 292-294. Staebler D. L. and Wronski C. R. (1980), 'Optically induced conductivity changes in discharge-produced hydrogenated amorphous silicon,' J. Appl. Phys. 51, 32623268. Stutzmann M.(1997), 'Microscopic aspects of the Staebler-Wronski effect', Mat. Res. Soc. Symp. Proc. 467, 37-48. Tanaka K. and Matsuda A. (1987), 'Glow-discharge amorphous silicon: growth process and structure',Mat Sci. Report!, 139-184. Tanaka M., Toguchi M., Matsuyama T., Sawada T., Tsuda S., Nakano S., Hanafusa H. and Kuwano Y. (1992), 'Development of new a-Si/c-Si heterojunction solar cells: ACJ-HIT (artificially constructed junction-heterojunction with instrinsic thin-layer)', Jpn. J. Appl. Phys. 31, 3518-3522. Tauc J., Grigorovici R. and Vancu A. (1966), 'Optical properties and electronic structure of amorphous germanium', Phys. Stat. Solidi 15, 627-631.

242

C. R. Wronski and D. E. Carlson

Tiedje T. (1984), 'Information about band-tail states from time-of-flight experiments', in Semiconductors and Semimetals, Vol. 21C, Pankove J. I., ed., Academic Press, Orlando, 207-238. Tsuda S., Takahama T., Hishikawa Y., Tarui H., Nishiwaki H., Wakisaka K. and Nakano S. (1993), 'A-Si:H technologies for high efficiency solar cells', J. NonCryst. Solids 164-166, 679-684. Uchida Y. (1984), 'DC glow discharge', in Semiconductors and Semimetals, Vol. 21A, Pankove J. I., ed., Academic Press, Orlando, 41-54. Watanabe T., Azuma K., Nakatani M., Suzuki K., Sonobe T. and Shimada T. (1986), 'Chemical vapor deposition of a-Si:H films utilizing a microwave excited Ar plasma stream', Jpn. J. Appl. Phys. 25-12, 1805-1810. Willing F., Bennett M. and Newton J. (1987), 'Thermal stability of interconnected aSi:H solar modules', Conf. Record 19th. IEEE Photovoltaic Specialists Conf., New Orleans, IEEE Press, Piscataway, 1086-1089. Woodcock J. M., Schade H., Maurus H., Dimmler B., Springer J. and Ricaud A. (1997), 'A study of the upscaling of thin film solar cell manufacture towards 500 MWp per annum', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 857-860. Wronski C. R., Carlson D. E. and Daniel R. E. (1976), 'Schottky barrier characteristics of amorphous silicon diodes', Appl. Phys. Lett. 29, 602-605. Wronski C. R., Abeles B., Tiedje T. and Cody G. D. (1982), 'Recombination centers in phosphorus-doped hydrogenated amorphous silicon', Solid State Commun. 44, 1423-1426. Wronski C. R. (1984), 'The Staebler-Wronski effect', in Semiconductors and Semimetals, Vol. 21C, Pankove J. I., ed., Academic Press, Orlando, 347-373. Wronski C. R. (1996), 'Amorphous silicon technology: coming of age', Solar Energy Mat. Solar Cells 41-42, 427-439. Wronski C. R. (1997), 'The light-induced changes in a-Si:H materials and solar cells—where we are now', Mat. Res. Soc. Symp. Proc. 467, 7-17. Wronski C. R., Lu Z., Jiao L. and Lee Y. (1997), 'An approach to self-consistent analysis of a-Si:H material and p-i-n solar cell properties', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 587-590. Yablonovitch E. and Cody G. D. (1982), 'Intensity enhancement in textured optical sheets for solar cells', IEEE Trans. Electron Devices 29, 300-305. Yang J. and Guha S. (1992), 'Double-junction amorphous silicon-based solar cells with 11% stable efficiency', Appl. Phys. Lett. 61, 2917-2919

Amorphous Silicon Solar Cells

243

Yang J., Banerjee A., Glatfelter T., Hoffman T., Xu, X. and Guha S. (1994), 'Progress in triple-junction amorphous silicon-based alloy solar cells and modules using hydrogen dilution', Conf. Record 24th. IEEE Photovoltaic Specialists Conf., Waikoloa, IEEE Press, Piscataway, 380-384. Yang J., Banerjee A., Lord K. and Guha S. (1998), 'Correlation of component cells with high efficiency amorphous silicon alloy triple-junction solar cells and modules', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 387-390. Yang L. and Chen L. F. (1994), 'The effect of H2 dilution on the stability of a-Si:Hbased solar cells', Mat. Res. Soc. Symp. Proc. 336, 669-674. Zanzucchi P., Wronski C. R. and Carlson D. E. (1977), 'Optical and photoconductivity properties of discharge produced a-Si', J. Appl. Phys. 48, 5227-5236.

CHAPTER 6

CADMIUM TELLURIDE SOLAR CELLS DIETER BONNET ANTEC GmbH, D-65779 Kelkheim, Germany

Will you walk into my wavetrap? said the spiter to the shy. James Joyce, Finnegans Wake, 1939.

6.1 Introduction Until today, silicon has been used as base material for photovoltaic solar energy conversion with increasing success, albeit at relatively high cost. Only three additional semiconductors have shown real promise for replacing silicon as the primary material for PV power generation: amorphous silicon, CdTe and Cu(In,Ga)Se2. Other materials, including Se, Cu2S, Cu 2 0, InP, CdSe and Zn3P2, have been studied, but, because of disappointing results or high cost, are no longer intensively investigated. Only GaAs is being developed and used for special applications, where very high efficiency is required in spite of high cost. The discussion which follows describes the technical status and industrial prospects of CdTe thin-film solar cells. From its basic physico-chemical properties CdTe is an optimum material for use in such cells. Work to master the technology for large-scale production has been highly successful. The market introduction of commercial products is imminent. Thin-film solar cells are large area diodes tailored to enable and maximise the absorption of light within a short distance from its space-charge region. The absorbed photons create electron-hole pairs. The potential energy of an excited (minority) carrier is converted into electrical energy as it is swept through the built-in electric field of the diode. The separation from its opposite (majority) charge carrier leads to an electric voltage, which can drive a current through an external circuit, such as an electric motor. As a minority carrier device, a solar cell requires a material of good electronic base properties—mainly high minority carrier lifetime and mobility. This can be achieved only by good crystalline properties, chemical purity, suitable doping and low-resistance contacting. Although there are several varieties of solar cells, the following general description applies most directly to thin-film solar cells in which the diode is created by two materials designated as the window layer and the absorber—'heterojunction'

245

246

D. Bonnet

devices. The field region is generated at the interface between window and absorber and mainly resides in the absorber. In our case CdTe is the absorber.

6.2 Early work There have been many efforts in the past 35 years to design and realise good junctions to CdTe films for extraction and collection of light-generated charge carriers (Bube, 1988). In the case of thin-film CdTe p-n homojunctions, there has been very limited success, because of strong light absorption in CdTe, a direct-gap semiconductor, coupled with a high surface recombination rate that severely limits the minority carrier lifetime and results in low quantum efficiencies. Furthermore, it is difficult to manufacture CdTe p-n junctions in thin-film form as the interdiffusion of doping species along grain boundaries degrades and distorts the junction. CdTe tends to beptype and is difficult to manufacture in n-conducting form and therefore an n-type heterojunction partner is required in order to induce a strong space-charge region as a prerequisite for good efficiency. Heteroj unctions are therefore the most promising configuration. The first heteroj unction was the n-CdTe//?-Cu2Te junction (analogous to the CdS/Cu2S solar cell under study at that time: Cusano, 1963). Although efficiencies around 7% were achieved, stability problems arising from the diffusion of Cu stopped further development of this cell structure. A heteroj unction partner with wider bandgap than CdTe allows light to enter the CdTe material more readily, by the so-called 'window effect'. Around 1970, a new heteroj unction was identified for CdTe with CdS as the n-partner (Bonnet and Rabenhorst, 1972), and this has had much success. Over the course of time, a concentration process has taken place so that most research and commercial interest is now focussed on CdTe/CdS p-n heterojunctions. The rationale for this selection is discussed below, but it should be admitted that it is at least partly empirical {i.e. 'the CdTe/CdS structure works').

6.3 The potential of the base material 6.3.1 Energy gap CdTe has an energy gap of 1.45 eV, and is therefore very well adapted to efficient conversion of solar light into electricity. Furthermore, the energy gap is 'direct', resulting in an absorption coefficient of >105 cm"1 for visible light, so that the absorber layer needs to be only a few /xm thick to absorb >90% of photons at energies >1.45 eV.

247

Cadmium Telluride Solar Cells

Current densities of 27 mA cm 2 and open-circuit voltages of 880 mV, leading to AM 1.5 efficiencies of 18.5%, can be expected for cells made from CdTe (Sites and Liu, 1995).

6.3.2 Thermodynamic properties The phase diagram of CdTe is reproduced in Fig. 6.1 (Zanio, 1978). Above 400 C, the stoichiometric compound is the stable solid phase, because the constituting elements have a significantly higher vapour pressure than the compound. In the high-temperature phase a slight nonstoichiometry is present in the form of a slight Cd deficiency, which leads to a native p-doping of the material. This property makes it relatively easy to produce CdTe films suited for thin-film solar cells. No excessive care has to be taken in 1200 1000

" 1

1

1 — ' 1092 ± 1°'

—1

1

1

Liquidus

_ - .

^^wLiquidus Solidus

o E £

. _ ^ -

600 449 + 2°

400

200

-

324 + 2°

• 0

i

I

i

10

20

30

I 40

50

'

60

1 70

1 80

1, 90

100

Atom fraction Te

Figure 6.1

Phase diagram of CdTe (from Zanio, 1978).

preparing the CdTe films as long as the substrate temperature is sufficiently high. CdTe or Cd + Te can be used as starting materials. The only requirement is the absence of disturbing impurities, which might impair the doping. In practice, the compound can easily be prepared in sufficiently high purity, as the constituting elements—Cd and Te—can easily be purified by standard chemical procedures. Due to the material's high ionicity (72%) (Hartmann et al., 1981), fewer dangling bonds occur at grain boundaries and crystallites tend to be well passivated. The energy of all photons in the solar spectrum is lower than the bond energy (5.75 eV) of CdTe, and this strong bonding leads to extremely high chemical and thermal stability. The

248

D. Bonnet

energy of solar photons is used only for the photovoltaic effect or the generation of harmless phonons, and it cannot break chemical bonds and destabilise the material. 6.3.3 Crystal lattice The natural crystal lattice of CdTe (Fig. 6.2)—being formally cubic—is de facto hexagonal: if viewed perpendicular to the direction of the cubic 111 axis, stacked planes of hexagonally packed alternating Cd and Te layers can be identified. In most deposited CdTe films, these planes tend to lie in the plane of the substrate (the 111 axis being perpendicular to the substrate), leading to columnar growth of crystallites.

(a)

(b)

Figure 6.2 Crystal lattice of CdTe in (a) the cubic representation and (b) as seen perpendicular to the cubic (111) axis, illustrating its quasi-layer structure.

6.3.4 Growth and doping of films On heating in vacuum to about 700 C, CdTe sublimes congruently, liberating Cd and Te in equal amounts, the residue remaining stoichiometric CdTe. On arrival of Cd and Te on the substrate, even in a non 1:1 ratio, CdTe condenses stoichiometrically as long as the substrate is heated above 449 C, at which temperature excess Cd and Te are not stable (see Fig. 6.1). In many cases, films deposited at lower temperatures, and therefore not necessarily at stoichiometric ratio, can be heated to create the stoichiometric compound. This allows numerous film deposition technologies to be applied. Moreover, as the material grows natively p-doped in thin-film form, no additional doping has to be introduced. Oxygen, being isovalent with Cd, is not a critical impurity, and may even enhance p-doping (Tyan and Perez-Albuerne, 1982). In many cases, quite large crystallites (up to 10 fim in diameter) will grow. The best films have been grown at

249

Cadmium Telluride Solar Cells

substrate temperatures around 600 C and deposition rates of ~1 nm per minute (Ferekides etal., 1993).

6.4 Diodes and cells Like CdTe, CdS has a strong tendency to form stoichiometric films, but, unlike CdTe, CdS films are natively n-doped by a slight non-stoichiometry. CdS can be deposited by essentially the same techniques as CdTe, permitting compatibility of manufacturing. A potential disadvantage is that CdS has a significant lattice mismatch to CdTe. Fortunately, after the post-deposition treatments described below, the negative consequences of this are only mild. back contact

p-CdTe (3-5 prn)

l&

n-CdS(100nm)

i^sssssssssssssssssssssssssss'^a^TC0 < 200 nm >

incident light

Figure 6.3

Film sequence of the CdTe thin-film solar cell as used today.

The n-CdS/p-CdTe heterojunction solar cell must be illuminated through the CdS window, so that the light is absorbed in the CdTe close to the junction. In the preferred fabrication procedure, the n-CdS film is deposited onto a transparent conductive oxide (TCO) film, typically ln 2 0 3 or Sn0 2 . Next the CdTe is deposited onto the CdS, and finally a low-resistance contact is made to the CdTe followed by a back electrode, which can be opaque. Figure 6.3 shows the superstrate cell structure. Figure 6.4 shows the energy diagram of the heterojunction. CdS is heavily n-doped and its conductivity under cell operating conditions increases on illumination (an effect known as 'light doping'), whereas CdTe is lightly p-doped (typically to a level of p < 10IS cm"3). Therefore essentially all the electric field drops within the CdTe layer. This field extends to a depth of about 1 pm, a value comparable with the optical absorption length. Light-generated electrons in the CdTe experience a drift field and move toward the junction into the CdS. Due to the strong absorption of CdTe (a > 104

250

D. Bonnet

conduction band

metal contact

Figure 6.4

Energy band diagram of the typical CdS/CdTe thin-film solar cell.

move toward the junction into the CdS. Due to the strong absorption of CdTe (a > 104 cm"1), the majority of electrons is generated in the field region of the CdTe layer and under influence of the field drift rapidly towards the junction. Hence charge separation does not have to rely on diffusion, which would be much less effective owing to the small lifetime (< 1 ns) of minority carriers in /?-CdTe. Electronic defects are primarily located at the metallurgical junction between CdS and CdTe, and these can act as recombination centres for minority carriers. Fortunately they can be significantly reduced in number by a special 'activation/passivation' step, discussed below in Section 6.5.3. As light-generated holes in CdS—being minority carriers in this layer—have a short lifetime and experience no drift field, they do not contribute to the photocurrent. Therefore the thickness of the CdS layer should be reduced as far as possible, to allow as much light as possible to penetrate through the CdS 'window' and enter the CdTe film. Interface states also act as recombination centres for the majority carriers (electrons in CdS and holes in CdTe) that cross the junction under forward bias. This leads to increased dark currents and thereby decreased photovoltage. It is thus evident that the action of defects must be reduced for optimum performance of the solar cell. It is also possible to make a CdTe substrate cell in which CdTe is laid down on a substrate and then n-CdS and TCO are added successively by deposition. However, the TCO-superstrate configuration is more successful, probably because of the material properties of the films involved. TCO is often deposited at temperatures above 600 C and is relatively stable with respect to typical CdTe device processing. Device-quality CdS is readily deposited onto the TCO, and CdTe deposition and post-processing

Cadmium Telluride Solar Cells

251

(which often requires heat treatments above 400 C) can be performed with minimal damage to the CdS. In fact, although there is some interdiffusion between CdS and CdTe, there is a miscibility gap between the two compounds that limits the composition of the alloy to a few percent substitution of either chalcogenide. The final step in the fabrication of the superstrate cell is the deposition of a low-loss electrical contact to CdTe. Although there are many ways of doing this, contact fabrication is typically the most delicate step and no contact processing temperatures exceed 270 C. Thus use of the TCO-superstrate configuration enables use of process steps with decreasing temperatures as the device is fabricated, whereas the alternative deposition sequence would require the relatively delicate CdTe contact to be made early in the fabrication process. These basic materials aspects have led to remarkable success in research organisations and universities. Indeed, then-record efficiencies of 15.8% were obtained on 1 cm2 cells (Ferekides et al., 1993). Recently a new record efficiency of 16.0% has been announced by Ohyama et al. (1997).

6.5 Cell production If CdTe thin-film solar cells are to become a commercially successful product, their promising basic properties have to be retained while meeting the following criteria: • • • • • •

high cell efficiency (10-15%) high module production speed (100,000 m2 p.a.) robust, forgiving manufacturing processes cheap substrate (commercial glass) low materials consumption (<10g m"2) long-term stability (20 years)

The following discussion shows that this can be achieved.

6.5.7 The CdS window layer The CdS film is photoelectrically inactive, but it absorbs a part of the light to be converted in CdTe below its absorption edge at 515 ran (2.45 eV). To maximise optical transmission of active light into the CdTe, the CdS film should therefore be as thin as is feasible with remaining pinhole-free and continuous. To maximise the photovoltage, the CdS should be as highly doped as possible. Fortunately CdS grows natively n-type without additional foreign doping under most deposition conditions. Under operational

252

D. Bonnet

conditions, its conductivity is increased by photoconductive processes. CdS deposition processes are similar to those used for CdTe. At present only a few processes have been developed as production technologies. Physical vapour deposition (sublimation/condensation, e.g. close-spaced sublimation) and chemical spraying are the main options. An alternative option is chemical bath deposition of a CdS film onto the TCO glass substrate from a metastable aqueous solution of cadmium acetate, thiourea, ammonia and ammonium acetate at temperatures of -70 C. Cells of 1 cm2 area and efficiency 15.8% have been made using this process (Chu et al., 1992). In spite of the high quality of the resulting CdS films, this process may ultimately be less suited for large-scale production, because of reproducibility problems (Ferekides etal., 1997a). Recently, cells of 15% efficiency have been made using the commercially attractive process of close-spaced sublimation of CdS (see Section 6.5.2). Analysis of a large number of specifically made cells from different laboratories shows that, when the CdS layer thickness is decreased, only 4 mA cm""2 of the 8 mA cm"2 of short-circuit current due to light below 515 nm can be realised before the voltage and fill factor start to fall because of 'weak' spots or areas (Granata et al., 1996). The optimum CdS thickness seems to lie in the range 50-80 nm. In industrial production, engineers tend to keep CdS films somewhat thicker than this, sacrificing part of the photocurrent for increased process safety.

6.5.2 The CdTe absorber layer The deposition process used for CdTe should utilise the advantageous materials properties mentioned above, especially the native p-doping, good crystallinity and high minority carrier (electron) mobility. Numerous deposition technologies that have the potential to do this can be imagined, and indeed about ten different procedures have been successfully developed. The following gives a brief description of the techniques which have led to solar cells of above 10% efficiency. Sublimation-condensation Many film deposition processes are based on the fact that CdTe not only forms stoichiometric films easily but also sublimes congruently without melting on heating. If heated sufficiently—typically above 600 C—CdTe evaporates to form Cd and Te by dissociation. As can also be seen from the phase diagram (Fig. 6.1), the remaining material stays stoichiometric, so no accumulation of either Cd or Te can occur. The starting CdTe material can be used until it is completely consumed. The classical film

Cadmium Telluride Solar Cells

253

deposition process is high-vacuum evaporation: CdTe is evaporated by sublimation from a heated crucible and condenses on a substrate positioned in front of this crucible inside a vacuum vessel. The crucible and source material are kept at a temperature around 700 C, and the substrate is heated to temperatures between 200 and 400 C. The upper limit on substrate temperature is determined by the ratio of the rate of re-sublimation from the growing film to the rate of material arrival from the crucible. Typical laboratory deposition rates are -10 A s~\ but this is determined primarily by the geometry of the deposition apparatus. Two commercially viable processes, which achieve both higher deposition rate and higher substrate temperature, are modifications of mis basic process: 1. Close-spaced sublimation (CSS) The evaporation source is made in the form of an extended, flat layer (either a semiconductor plate or an open crucible filled with semiconductor granulates, and heated to a temperature at which the material sublimes—typically 700 C). The source is essentially the same size as the heated substrate and is placed a short distance ('close spaced') in front of it so that the sourcesubstrate distance is typically a few cm. Due to the confined geometry, a large mean free path is not required for the subliming species to reach the substrate, nor is a high vacuum needed. An inert gas pressure of 1 mbar or higher can easily be tolerated and good films have even been made at ambient pressure. CSS was employed to deposit the CdTe film for the first CdTe thin-film solar cell to achieve the benchmark conversion efficiency of 10% (Tyan and Peres-Albuerne, 1982). The present world record of 15.8% has also been achieved with this technology, using substrate temperatures of around 600 C at 40 mbar (Ferekides et al., 1993). This requires a glass such as borosilicate, which can resist this temperature without softening. If low-cost soda lime glass, which softens at 550 C, is used, lower CSS temperatures of around 500 C must be employed. Films of less perfect quality are expected at the lower temperatures. However, soda-lime glass substrates were successfully employed ten years later by Bonnet (1991). Using deposition rates of above 10 /xm min"1, efficiencies of ~ 12% were achieved, and 10 x 10 cm2 modules of 10.5% efficiency have been fabricated at ANTEC GmbH. 2. Modified closed-space sublimation As already mentioned, soda lime glass, the cheapest substrate, becomes soft above 550 C and has to be supported in order to avoid warping. A process in which the glass is supported by rollers during deposition was developed by Solar Cells Inc. (Meyers et al., 1993, Sandwisch, 1994). The CdTe is evaporated from semi-cylindrical troughs positioned above the horizontal substrate and the vapour is directed downward to the substrate by suitably structured shields. This has so far led to modules of 60 x 120 cm2 area, certified at NREL as up to 8.2% efficient.

254

D. Bonnet

Chemical spraying This process was used as early as 1966 for depositing the CdS films of CdS/Cu2S thinfilm solar cells (Chamberlin, 1966), and since then it has been developed for CdTe deposition. An aerosol of an aqueous solution containing heat-decomposable CdCl2 and an organic Te compound is directed onto the substrate, which is kept at a temperature of about 500 C. Cd and Te are liberated by pyrolysis on the surface and immediately react to form a CdTe film. CdS can be produced in a similar manner. For CdS films, thiourea and CdCl2 are typically used to supply S and Cd respectively. The films are somewhat porous, but have nevertheless led to cells of 12.7% efficiency (Albright etal., 1992). The liberated solvent (water) limits the deposition rate somewhat, but the process is simple and does not need a vacuum. This technology was developed into an industrial process by Golden Photon Inc., and modules of 60 x 60 cm2 area have been manufactured at efficiencies of -8%. The CdTe films appear to contain numerous voids, and perhaps for this reason, the chemical spraying method requires higher film thicknesses than do other procedures. Galvanic deposition Both CdTe and CdS can be electrodeposited from an aqueous solution at temperatures of 90 C, but fluctuations in stoichiometry can only be remedied by thermal annealing above 400 C. The first such cells to exhibit 10% efficiency were made in 1983 by two industrial efforts, at Monosolar Inc. (Basol, 1988) and AMETEK Inc. (Meyers and Liu, 1988), using p-type and n-type CdTe respectively. The early AMETEK devices had a metallic substrate and the n-CdS/i-CdTe/p-ZnTe configuration, but in later work the more popular n-CdS/p-CdTe heterojunction configuration was adopted. In the latter case deposition is onto a glass substrate coated with a transparent conductive oxide (TCO), which has a typical sheet resistance of ~10 ohm square"1. In order to keep the deposition potential constant across large-area substrates, deposition must be performed at low current densities (typically 0.5 mA cm"2), resulting in deposition rates of about 1 nm s"1. After termination of the above-mentioned industrial efforts in the USA, BP Solar acquired the rights and continued the effort (Woodcock et al., 1991), achieving 14% efficiency for small cells and >8% for 30 x 30 cm2 modules (Woodcock et al., 1995). If many substrates are coated in parallel, the long deposition time of about 1 hour can be compensated by high throughput.

Cadmium Telluride Solar Cells

255

Screen printing CdS and CdTe films can be screen printed from slurries containing CdS and CdTe (or Cd and Te powders), respectively, plus CdCl2 as flux (Yoshida, 1995; Cleminck et al., 1992). Films are then given a heat treatment for about an hour in a controlled atmosphere at about 700 C (for CdS) or 600 C (for CdTe), to produce large-grained films with thickness from 15 to 30 fim. Efficiencies above 12% have been reported. This process has appeal for manufacturing due to the simplicity of the process and equipment. On the other hand, semiconductor film thickness is 3 to 6 times that of films made by other techniques, the process involves several hours of heat treatment to produce high-quality films, and high-quality substrates (made of borosilicate glass) are required. Chemical vapour deposition Chemical vapour deposition (CVD) has some resemblance to the spraying process, insofar as CdTe is formed by chemical reaction from thermally decomposable compounds. In the case of CVD, the compounds are gaseous and are injected into the reactor by a carrier gas, e.g. H2. Typically metal-organic compounds such as dimethyl cadmium and diethyl tellurium are used as precursors for the reaction (Ghandi et al., 1987; Rohatgi, 1992). CVD has the advantage that doping species such as P or As can also be introduced (e.g. in the form of thermally decomposable AsH3 or PH3) by a suitable gas-mixing system. This process, although slower than the fast physical vapour deposition processes (/*m If1 vs. ^m min"1 ) has wide process latitude in gas composition, allowing basic studies to be made. For example, CVD has been used at Georgia Institute of Technology, where one interesting result has shown that, even under very strong deviations of the Cd:Te ratio from 1:1, device-quality stoichiometric films can be made, again giving proof of the latitude available for CdTe processing. Efficiencies achieved on experimental cells have been well above 10%. However, due to the toxicity, high cost and low materials efficiency of the metal-organic gases, this process is generally considered less suited for large-scale production of CdTe thin-film solar cells. Atomic layer epitaxy (ALE) In this process alternate monolayers of Cd and Te are deposited on the substrate by alternately directing gas streams containing Cd or Te onto it. This allows very stoichiometric and pure films to be grown. Cd and Te are evaporated into the inert gas streams in a closed system at elevated temperatures. The gas streams are of high

256

D. Bonnet

temperature and are guided inside high-temperature tubing to avoid condensation. The substrate is also heated and the deposition is driven by the chemical bonding energy between Cd and Te. Cells of 14% efficiency have been reported, and modules of 5 x 5 cm2 area at efficiencies above 10% have been made by Microchemistry Inc. (Skarp et al., 1991 and 1992). This process has some similarity to that used for the very first CdS/CdTe cells around 1970. Here the compound CdTe had been evaporated into an inert gas-stream which had been guided onto a substrate at lower temperatures—but still around 500 C (Bonnet and Rabenhorst, 1972). ALE requires very low deposition rates, but enables multiple glasses to be coated in parallel, as does electrodeposition. The technology has not been pursued further at the time of writing this chapter. Sputtering Bombardment with argon ions of a solid target of CdTe leads to emission of Cd and Te from the surface of the target. The atoms move in the ambient vacuum and condense on the substrate, forming CdTe films at suitable temperatures of up to 300 C. This technology has led to good results in first experiments at NREL (Abou-Elfotouh and Coutts, 1992) and the University of Toledo (Compaan etai, 1993). Deposition rates are typically < 100 nm min"', lower by a factor of 10 than for CSS. This process may gain industrial application for deposition of semiconductor back contacts, e.g. ZnTe, as such contacts are typically very thin (Gessert et al., 1995).

6.5.3 The CdS/CdTe interface and activation of the cell CdS and CdTe in thermal equilibrium can form mixed compounds CdS^Te,^ only for limited ranges of x (0<x
257

Cadmium Telluride Solar Cells

tion at 70 C) and impeded by high-temperature deposition (like CSS at 400-500 C). Only relatively small deviations from pure CdS and CdTe are possible, so only a small mixing of S into CdTe, and of Te into CdS, is possible. In order to achieve the ultimate performance, methods must be found to 'tailor' the junction to optimum properties by a process suited to large-scale production.

Voltage/V Figure 6.5

1-V curves for (A) as-deposited, (B) heat-treated and (C) activated cells. From Al Allak etal. (1996).

In the technology of II-VI compound semiconductors, maximum performance can often be achieved only after special treatments involving a chloride at elevated temperatures. This is also true for the CdTe/CdS heterojunction thin-film solar cell. It has become common practice to activate the cells by using the influence of CdCl2 at elevated temperatures (Basol, 1992; Meyers et al, 1998). In the 'classical' process, the CdCl2 heat treatment is a post-deposition procedure. Following deposition, the CdTe films are wetted with solutions of CdCl2 in methanol, dried and annealed at temperatures around 400 C for 10-30 minutes in air. In other cases, such as screen printing or electrodeposition, CI" ions are supplied during the growth process. After deposition of the back contact, a significantly better performance is observed for 'activated' cells (Al Allak et al, 1995). Figure 6.5 shows three typical curves for differently treated and untreated cells, showing the dramatic improvement in performance. All essential parameters (open-circuit voltage, short-circuit current and fill-factor) are increased after such treatments. Figure 6.6 shows the corresponding spectral response curves for different

258

D. Bonnet

(open-circuit voltage, short-circuit current and fill-factor) are increased after such treatments. Figure 6.6 shows the corresponding spectral response curves for different activation steps. Only for the properly activated cell does the classical 'windowresponse' expected for such a heterostructure become obvious. '

CdCI2 treated As deposited, good area

— o — Heat Treated As deposited, poor area

W a v e l e n g t h (nm)

Figure 6.6

Spectral response for as-deposited, heat-treated and activated cells. From Al Allak e/a/. (1996).

The activation process is not yet completely understood, but it has become clear that it produces beneficial morphological and electronic changes (Levi et al., 1994). Temperature-dependent current-voltage measurements (Al Allak et al., 1996) indicate that, in as-deposited and only heat-treated samples, the current transport is dominated by tunnelling recombination processes. In contrast, in CdCl2-treated cells, which show much improved electrical characteristics, current transport occurs by thermal emission across the junction, which indicates a reduction in interfacial recombination rates. This improvement goes hand-in-hand with the increase of the density of a hole trap 0.48 eV above the valence band edge (Laurengo, 1997). It is thus credible that the major effect is the improvement of crystalline quality in the depletion region of the cells. Furthermore, much more uniform photoelectric sensitivity across the cell surface results from activation (Galloway et al., 1995). In as-deposited cells, most of the junction is inactive but activated cells show high homogeneous sensitivity independent of bias. More detailed studies of the activation process have shown that the density of interface states at the physical boundary is

Cadmium Telluride Solar Cells

259

markedly reduced, decreasing the voltage-limiting reverse saturation current in the diode. Unfortunately the number of defect centres in the space-charge region is also increased, limiting the improvement (Rohatgi, 1992). Recently, direct evidence has been produced showing that the grain boundaries in the CdTe films are well passivated during the CI" activation process by the creation of /?-type accumulation regions that drive the photogenerated electrons into the grains and away from the grain boundaries (Edwards et al, 1997). With the requirement of more closely defined processes suitable for technical scaleup, different methods of activation have been studied. CdCl2 has been deposited using standard thin-film technologies, including high-vacuum evaporation and close-spaced sublimation. Heat treatments of CdTe films on which about 100 nm of CdCl2 has been deposited produced results similar to those obtained by the wet process. An alternative process is to use gaseous chlorides, such as HC1 (Zhou et al., 1994; Sasala et al., 1996) or CdCl2 (McCandless et al., 1996; Sasala et al., 1996), during the heat treatment.

6.5.4 The back contact A wide variety of contact structures have proved successful in making >10 % efficient devices. In general this topic is treated as confidential by industrial companies, and is considered too technology-oriented by universities to merit intense study. However, the back contact is fundamentally important because it bears on the essential question of long-term stability. Unfortunately, the highest efficiency cells have been made using a Cu-doped graphite back-contact layer that is not well suited for monolithic integration of cells into a large-area (60 x 120 cm2) modules. There are two general principles for making ohmic contacts to p-type semiconductors: 1. To use a metal of work function higher than the electron affinity of the semiconductor in order to align the top of the valence band with the Fermi level of the metal. The electron affinity of CdTe is 4.3 eV. 2. To create a highly doped back-surface layer in the semiconductor. The barrier created by the back-contact metal in the semiconductor will then be thin enough for holes to tunnel through efficiently. The problems for CdTe are evident for both cases: low-cost metals of work function greater than 4.5 eV are not available, and p-doping in CdTe suffers from a strong tendency for acceptors to self-compensate. Furthermore, acceptors cannot be introduced by

260

D. Bonnet

diffusion doping from the surface, as dopants generally diffuse preferentially along grain boundaries, leading to shunting of the cell before sufficient doping levels can be achieved. The latter problem can be alleviated to some extent by using CdTe films of higher-than-necessary thickness, e.g. >5 /im. In practice, some methods may combine approaches 1 and 2 by first contacting CdTe with a more easily doped semiconductor (generating a p-p* junction), and then with metals of high work function. Efforts have primarily been directed towards four semiconductors: HgTe, ZnTe:Cu, Te and Cu2Te (Chu, 1988; Meyers etal., 1990; Tyan, 1980; Tang et al., 1991; Ferekides et al., 1997b). All three are p-type semiconductors with suitable work functions. Nevertheless, modification of the surface of a polycrystalline thin film has proved to be a delicate and often material- and morphology-dependent process (Pompon, 1985; Fahrenbruch, 1987). A very thin film of Te, created by chemical etching of the CdTe surface after activation or deposition of Te, seems to be especially beneficial to the back contact (Sasala, 1997). In many cases, such as the above-mentioned graphite contact, copper (an acceptor in CdTe) is added, which on annealing can diffuse into the CdTe film. An adverse consequence of this can be reduced stability, as Cu has high diffusivity in most semiconductors (McCandless, 1995; Chou, 1995) and may propagate into the semiconductor even at cell operating temperatures (-60 C). If Cu reaches the junction, it first reduces the junction width, and then it compensates donors in the CdS layer. The photocurrent is reduced in both cases. Another alternative has been to dope the graphite with HgTe, but this also has some unsatisfactory practical aspects (Niles et al., 1996). Fortunately, recent analysis (Sites and Lui, 1995) has shown that a completely nonrectifying back contact is not required. At room temperature, a contact barrier of 200 meV will not lead to a noticeable reduction of output. The thermally activated reverse currents in such junctions generate only a low series resistance. Only if the barrier height increases with time will degradation occur. This type of contact and its time-dependent behaviour can be easily identified by observing the l-V curves in the forward direction. Observations of an increasing 'rolloff (or current saturation) can be used as early warning for degradation, even before the power output suffers. Unfortunately, publications rarely show I-V curves at / > 0 in the forward direction, and it is recommended that they should (Bonnet et al., 1992). Medium-term measurements on technical pilot modules exposed outdoors, showing minimal degradation for over 20,000 hours of continuous illumination at 0.8 Sun intensity, have been performed for BP Solarex and Solar Cells Inc. (which has since changed its name to First Solar LLC) by Sasala et al. (1996). The success of these tests is a strong indication that long-term stability can be achieved by careful design of the back contact.

Cadmium Telluride Solar Cells

261

6.5.5 Substrates The substrate onto which thin-film solar cells are deposited will to a large extent determine the cost of the final module. Many experiments have been made on borosilicate glass, which is stable at temperatures up to 600 C. The cost of this type of glass is about twice that of standard soda-lime window glass, which costs about $5 per m2. Whereas the world-record cell was deposited at low speed and elevated temperature onto borosilicate glass, the use of soda-lime glass under suitably reduced substrate temperatures has also resulted in efficiencies above 13% (Ferekides, 1994). There exists strong evidence that a certain amount of diffusion of Na from the soda-lime glass into the growing film promotes improved structure and properties. Comparison of cells made on glass + TCO and glass + Na-diffusion barrier (Si02) + TCO shows improved results for the first option (Bonnet et al., 1994). All industrial efforts today use soda-lime standard window glass as substrate.

6.5.6 The TCO film The transparent conducting oxide (TCO) base contact of the diode is an essential part of the cell and its optimisation can lead to a significant improvement in cell and module efficiency. Generally, metal oxides based on indium and tin are used, which can be made highly conductive by either native or additive doping while still keeping the high optical transmission that results from their large energy gap. Free carrier absorption and impurities lead to reduced transmission. Generally a TCO film is considered good if its conductivity is <10 ohms square"1 and its optical transmission is >80% in the region of solar sensitivity of the diode. Good films usually are made by cathode sputtering of targets of ln 2 0 3 or Sn0 2 . Detrimental in-diffusion of In into the CdS/CdTe film stack can be reduced by a pure Sn0 2 film between ln 2 0 3 and CdS. Recently, CdSn0 3 has been studied as TCO material, and has shown greatly improved performance, especially extremely high optical transmission even at low sheet resistances. Cells showing 14% efficiency have been made using this type of TCO film (Wu et al., 1997). However, high-temperature processing makes borosilicate glass mandatory with CdSn0 3 films. Generally the influence of TCO film morphology on cell performance has been neglected. It has been shown, however, that the roughness of Sn0 2 can improve cell performance (Ferekides et al., 1997c).

262

D. Bonnet

6.6 Module production As was demonstrated in the field of amorphous silicon more than 10 years ago, thinfilm solar cell technology easily lends itself to making integrated modules. A typical module has around 100 individual cells per metre of module length, connected in series

glass substrate

Figure 6.7

10 cells

interconnection region

Arrangement of the individual cells in a thin-film module.

by applying three sets of separation cuts to the growing film stack during production by means of laser ablation or mechanical machining. Figures 6.7 and 6.8 illustrate the principle. Narrow individual cells running parallel to one edge of the substrate are defined and sequentially connected: the top electrode of cell 1 contacts the bottom

CdS/CdTe fit TCO film,

back conta cell!

first cut: TCO .second cut: CdS/CdTe .third cut: back contact / cell2

substrates

Figure 6.8

Principle of interconnection by triple scribing of individual layers.

electrode of cell 2 while being separated from the top electrode of cell 2. The top electrode of cell 2 connects to the bottom electrode of cell 3 and so on. The cuts into and through the individual films are often made by lasers, such as Nd:YAG lasers. In CdTe

Cadmium Tellaride Solar Cells

263

thin-film solar cells another technique can be used: the CdS/CdTe double film adheres quite well to the TCO film, which is very flat and rather hard. CdTe, being very brittle, allows the second and third cut to be made by mechanical tools, which cut the brittle CdTe but slide on the TCO film without cutting it. The third cut can also be made with a mechanical scribe and there is no disadvantage in cutting down to the TCO film. Contacting and encapsulation technologies for CdTe have been adapted from those for Si and readily lead to modules of 0.5-1 m2 in size. Figure 6.9 shows a typical module of 60 x 120 cm2 size. The following are the individual steps for module production from glass to module: cleaning of substrate glass; TCO-deposition; scribing step; CdS deposition; CdTe deposition; activation; scribing step 2; back contact deposition; scribing step 3; edge insulation and contact attachment; lamination; safety check; power test and inspection. For mass production, these processes must be integrated into a production line. A processing speed of >1 m min~'. is needed if 100,000 m2 per annum is to be produced, corresponding to 10 MWp at 10% efficiency. The glass sheets to be coated must move fully automatically under defined processing parameters and environment. As many

Figure 6.9

60 x 120 cm2 module made by First Solar LLC (formerly Solar Cells Inc.)

264

D. Bonnet

steps are executed at elevated temperatures of 200-500 C, heat-up and cool-down chambers must be included, increasing the total length of a typical processing line to as much as 150 m. This line will be folded in the production hall, connecting the input storage area with the output storage area, with added shipping facilities for 3000 tonnes of (essentially) glass per annum. The investment cost for the plant will be in the range 10-20 million ECU. For optimum utilisation of this capital, such a plant must operate in three shifts, seven days a week, with the sole interruptions being breaks for refilling and service.

6.7 Industrial status—achievements and projections Recently 14.3% efficiency has been achieved on small cells using processes and materials suited for large-scale production (close-spaced sublimation and float glass, respectively) (Ferekides etal., 1997a). Table 6.1

Status of pre-production efforts in CdTe thin-film solar cells

Company

Process

Matsushita Corp. Golden Photon Solar Cells Inc.

Screen printing Spraying Sublimation

It

ANTEC GmbH BP Solarex 1)

II

Close-spaced sublimation Electrodeposition tl

Module area (cm2) 1200 3528 6728 64 86 706 4540

Efficiency (%) 8.7 7.7 9.1 10.5 10.5 10.1 8.4

Source: Ullalef al. (1997).

It is the central aim of all public and private investment into solar cell development work to facilitate production under conditions of competitive industrial operation. Under the present situation of cheaply extracted non-regenerative fossil fuels, this is a difficult task. Governments and international agencies are therefore helping industrial R&D entities with limited subsidies; this is especially difficult to justify under present conditions of decreasing budgets. A few companies are actively pursuing the aim of ensuring the financial and technical assets to enter production. Table 6.1 gives industrial pre-production results (Zweibel et al., 1996).

Cadmium Telluride Solar Cells

265

6.7.1 First Solar LLC {formerly Solar Cells Inc.) Solar Cells Inc. (SCI) of Toledo, Ohio started work on CdTe thin-film solar cells around 1990, using a sublimation process at reduced inert atmosphere of 5 torr, directly aiming at large-area modules. They demonstrated 3 years of stability under real outdoor exposure, monitored by NREL and 11 years under accelerated testing conditions. SCI produced a 60.3 Wp 8.4% efficient module—the most powerful thin-film polycrystalline module in the world (Sandwisch, 1996). The process for making these modules has been proved on a 100 kWp pilot line, and no technology barriers have been identified. SCI will now concentrate on the expansion of the current pilot line to 200 kWp while installing a 100 MW/yr facility. In 1999, SCI changed its name to First Solar LLC.

6.7.2 BP Solarex The technology of electrodeposition of CdTe goes back to studies at AMETEK Corp. and MONOSOLAR Corp. around 1980. The UK-based BP subsidiary, BP Solar, acquired Monosolar's CdTe technology in 1984 and initiated its CdTe development efforts around 1985 in Sunbury-on-Thames, near London. 30 x 30 cm2 integrated modules of 8% efficiency were made and studied outdoors and under simulated sunlight. In 1999, BP Solar merged with the US PV company Solarex to become BP Solarex, headquartered in Baltimore, Maryland. BP Solarex is aiming at commercial production of CdTe modules, and has acquired a factory for thin-film module production in Fairfield, California, where it intends to employ 100 people and invest US $20 million to install a production capacity of 1000 'Apollo' CdTe panels a day, each about 150 x 60 cm2 (Marshall, 1997; www.solarex.com). This would correspond to annual production of >18 MWp per annum at 10% efficiency. Recently BP Solarex has achieved a new world record efficiency for large-area CdTe: a 40 x 150 cm2 module has been certified at 10.8% aperture efficiency at NREL (Bolko von Roedern, personal communication).

6.7.3

ANTECGmbH

ANTEC has continued development of the 'CTS' technology, which had been started anew at the Battelle Institute in Frankfurt, Germany in 1990 after Battelle's historic initial work in 1970. ANTEC uses its proprietary CSS process and has manufactured 10 x 10 cm2 modules of 10.5% efficiency. The basic technology steps have now been

266

D. Bonnet

developed for transfer into production, and ANTEC will build a highly automated 10 MWp production plant in Rudisleben, Germany, for which it has secured the financial capital. Full-scale production of 60 x 120 cm2 CTS modules is envisaged to start early in the year 2000. Module costs of 2.20 DM/WP have been calculated at 8% module efficiency and 80% factory yield.

6.7.4 Golden Photon Inc. Under the name Photon Power Inc, a small group of engineers based in El Paso, Texas studied the technology of chemical spraying for about 20 years, starting with the CdS/Cu2S thin-film solar cell and later adapting the technology to CdTe/CdS devices. After achieving technical success, they were taken over in 1992 by COORS Inc. of Golden, Colorado and renamed Golden Photon Inc. Work continued and was intensified, resulting in a pilot-production line generating integrated modules. The highest conversion efficiency reported on a 60 x 60 cm2 module was 7.7%. Small cells have shown efficiencies exceeding 14%. However, in 1996 Golden Photon encountered stability problems in their modules and in May 1997 the CdTe effort was terminated and will be divested (Anon., 1997).

6.7.5 Matsushita Corp. Based on their early work on the CdS/Cu2S solar cell in the sixties, Matsushita Corp., Japan started earlier than 1977 to apply the process of screen printing to the fabrication of CdTe thin-film solar cells. Efficiencies of around 10% were reported for individual cells, and integrated modules of up to 1200 cm2 of 8% efficiency were made. Small modules made by this technology are deployed in commercial pocket calculators, the total annual production being ~1 MWp. In 1996, Matsushita stated their intention to change their deposition technology to CSS. They have now made 1 cm2 cells at 15 % efficiency and intend to develop the fabrication technology for large-area cells (Nishio, 1996).

6.7.6 Deployment of modules Apart from Matsushita's ongoing production of small screen-printed modules at a rate of 1 MWP per annum, a number of modules and arrays have been installed from other

Cadmium Telluride Solar Cells

267

industrial pilot-line production efforts (Zweibel et al., 1996). Golden Photon Inc. has installed CdTe systems of total capacity -100 kWp, mainly for water pumping. Two 25 kWp CdTe installations, supplied by Golden Photon Inc. and Solar Cells Inc., have been deployed at China Lake, California. SCI has also deployed two 10 kWp gridconnected arrays at Davies, California and in Toledo, Michigan, and supplied four nominal 1 kWp arrays to NREL, Toledo Edison (2 arrays) and Tunisia.

6.8 Economic aspects The 1999 world market for photovoltaic modules was -200 MWp, of which the USA delivered about 40% and Europe about 22.5%, which is too low for the region's economic potential. A study from the Canada Centre for Mineralogy and Energy

1995

1996 ]

1997

New thin film

1998

1999 ]

2000

2001

Current thin film

2002 |

2003

2004

2005

Crystalline silicon

Figure 6.10 Expected market share of advanced thin-film systems (polycrystalline thin-film solar cells). Adapted from Woodcock et al. (1997).

Technology (Leng et al., 1996) predicts an annual increase of 15-30% for PV over the next 15 years, with sales volumes of 830 MWp under a business-as-usual scenario, and 4000 MWp under an accelerated scenario, for the year 2010 at average module prices of $2 and $1.50 per Wp respectively. Such a market would enable PV manufacturers to construct several 100-200 MWp/year plants. It is obvious that market demand depends on the price at which solar cells can be offered. This means that increase in efficiency and reduction of production cost are re-

268

D. Bonnet

quired. According to experts (Ostwald, 1996), the world-wide market can increase from today's figure of -120 MWp per annum to the mentioned size only by exploitation of low-cost thin-film technology. The low cost expected for polycrystalline thin-film solar cells is predicted lead to a gradual take-over of market share under a strongly increasing market, as illustrated in Fig. 6.10. The expected low production cost of about 1 ECU per Wp can be achieved only by using 'economies of scale', i.e. by large-scale production in units having capacities of 10-30 MWp/year. Investment costs for a 10 MWp/year CdTe plant will amount to approx. 15 million ECU, independent of technology, and the construction of the first plant will take between 1.5 and 2.5 years. This will require significant commitment from the industrial entrepreneurs. Detailed cost analyses were made by a representative group of industrial players in the study, MUSIC-FM, financed by the EU (Woodcock, 1997). For CdTe modules, this study predicted production cost of 0.66 ECU per Wp at an efficiency of 10% and a plant capacity of 60 MWp/year. Generally 50% of this value is attributable to consumables—mainly substrate and cover glass and semiconductor materials, 25% to equipment depreciation, and the remaining 25% to staff and rent of site.

6.9 Health and environmental aspects As with many advanced complex products, CdTe thin-film solar cells contain materials with potential impact on the environment and human health. Cadmium is not uncommon in objects used by humans. For example, classic art has used CdS and mixtures with CdSe as yellow and red pigments, without any undue consequences. The crucial question to be answered is whether this metal can enter the human food chain and body. This question has been under intense study in recent years by independent organisations, as well as the development groups involved in thin-film solar cell research (Alema and van Engelnburg, 1992; Patterson et al, 1994; Steinberger, 1998; Moskowitz et al., 1994). Two essential parts of the life cycle of CdTe thin-film solar modules can be identified: production and use. Production of future polycrystalline thin-film solar cells basically employs techniques common in chemical industry, and the substances involved are easily manageable by standard processes. Workers in development laboratories have shown no unusual uptake of Cd under periodic medical scrutiny. Production is possible under existing safety laws without putting the health of staff at risk. It appears to be technically and economically possible to design and operate a factory with zero Cd emission.

Cadmium Telluride Solar Cells

269

The studies mentioned furthermore show the use of CdTe solar modules is coupled with negligible risk for the environment and humans even under severely irregular conditions. Cd and Te are strongly bound to form the inert compound CdTe, which will yield Cd only at temperatures above 1000 C in air. In case of exposure to fire, the substrate- and cover-glasses will melt long before the CdTe decomposes, thereby including the semiconductor into the re-solidifying glass. There are early indications that the CdTe-material is actually dissolved in the glass flux. Incineration experiments conducted by BP Solarex in cooperation with a fire research institution have not led to detectable emissions. During use, a CdTe module bears comparison with laminated glass similar to that used in cars. Thus modules will not easily break and release their contents. At their end of life, modules can be recycled by crushing the whole and returning the debris to the smelters, who can inject the material into their processes without significant additional cost. The glass used as encapsulant in the modules under this procedure will substitute the silicate additive (usually sand) the melts will need. Other compounds, such as CdTe, will re-enter the product line. As a proactive measure, manufacturers will not recommend the disposal of modules in regular land-fill sites.

6.10 Conclusions Thirty-five years after the first demonstration of the potential of the CdTe thin-film solar cell by Cusano at General Electric and Lebrun from Philips, and thirty years after the first manufacture of the CdS/CdTe heteroj unction solar cell in its present configuration by Bonnet and Rabenhorst at Battelle Institute, the CdTe thin-film solar cell is approaching production status. More than twenty university groups and ten industrial companies have expended efforts, for more or less extended periods of time, to develop cells and modules, using about ten deposition technologies for CdTe and CdS. In recent work, no technology barriers for large-scale production have been identified, although some processes, such as screen printing, have shown limitations. At present, four industrial enterprises in Europe, USA and Japan are on track to install low-cost production processes. This review has tried to illustrate that the CdTe thin-film solar cell shows high promise of achieving cost levels of -0.5 Euro/Wp in mass production, and has the potential of 15% efficiency for a mature technology. Further aid by basic R&D will assist the technology to obtain increased performance and reduce the cost of the product. A bright future is expected for the CdTe photovoltaic industry.

270

D. Bonnet

References Abou-Elfotouh F. A. and Coutts T. J. (1992), 'RF planar magnetron sputtering of polycrystalline CdTe thin-film solar cells', Int. J. Solar Energy 12, 223-231. Al Allak H. M., Galloway S. A., Brinkman A. W. and Durose K. (1995), 'The effect of processing conditions on the electrical and structural properties of CdTe/CdS solar cells, Proc. 13th. European Photovoltaic Solar Energy Conf., Nice, H. S. Stephens & Associates, Bedford, 2135-2138. Al Allak H. M., Brinkman A. W., Richter H. and Bonnet D. (1996), 'Dependence of CdS/CdTe thin-film solar cell characteristics on the processing conditions', J. Crystal Growth 159, 910-915. Albright S. P., Chamberlin R. C , Ackerman B. and Jordan F. J. (1992), 'Performance measurement irregularities on CdS/CdTe devices and modules', Int. J. Solar Energy 12, 109-120. Alema E. A. and van Engelenburg B. C. W. (1992), 'Environmental risk of CdTe and CIS solar cell modules', Proc. 11th. European Photovoltaic Solar Energy Conf., Montreux, Harwood Academic Publishers, Chur, 995-998. Anon. (1997), 'Golden Photon shuts down its CdTe thin-film R&D after halting production', Photovoltaic Insider Report XVI, No. 5, p. 1. Basol B. M. (1988), 'Electrodeposited CdTe and HgCdTe solar cells', Solar Cells 23, 69-88. Basol B. M. (1992), 'Processing high efficiency CdTe solar cells', Int. J. Solar Energy 12, 25-36. Bonnet D. and Rabenhorst H. (1972), 'New results on the development of a thin film p-CdTe/n-CdS heterojunction solar cell', Conf. Record 9th. IEEE Photovoltaic Specialists Conf, Silversprings, IEEE Press, Piscataway, 129-131. Bonnet D., Henrichs B. and Richter H. (1991), 'High-rate deposition of high-quality CdTe films for high-efficiency solar cells', Conf Record 22nd. Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 1165-1168. Bonnet D., Henrichs B. and Richter H. (1992), 'Some phenomena in CdTe/CdS thin film solar cells made by close-spaced sublimation', Int. J. Solar Energy 12, 133-136. Bonnet D., Henrichs B., Jager K.-H. and Richter H. (1994), 'Methods of making p-n CdTe/CdS thin film solar cells', U.S. Patent No. 5,304,499. Britt J. and Ferekides C. S. (1993), 'Thin-film CdS/CdTe solar cell with 15.8% efficiency', Appl. Phys. Lett. 62,2851-2852. Bube R. H. (1988), 'CdTe junction phenomena', Solar Cells 23, 1-17.

Cadmium Telluride Solar Cells

271

Chamberlin R. R. and Skarman J. S. (1966), J. Electrochem. Soc. 113, 86-89. Chou H. C , Rohatgi A., Thomas E. W., Kamra S. and Bhat A. K. (1995), 'Effects of Cu on CdTe/CdS heterojunction solar cells with Au/Cu contacts', J. Electrochem. Soc. 142, 254-259. Chu T. L., Chu S. S., Han K. D. and Mantravadi M. (1988), 'Mercury telluride as an ohmic contact to efficient thin film cadmium telluride solar cells', Conf. Record 20th. Photovoltaic Specialists Conf., Las Vegas, IEEE Press, Piscataway, 14221425. Chu T. L., Chu S. S., Britt J., Chen G., Ferekides C. and Ullal H. S. (1992), 'High efficiency thin-film cadmium telluride solar cells', Proc. 11th. European Photovoltaic Solar Energy Conf., Montreux, Harwood Academic Publishers, Chur, 988-990. Clemminck I., Burgelman M., Casteleyn M. and Depuydt B. (1992), 'Screen-printed and sintered CdTe-CdS solar cells', Int. J. Solar Energy 12, 67-78. Compaan A. D., Tabory Ch. N., Li Y., Feng Z. and Fischer A. (1993), 'CdS/CdTe solar cells by RF sputtering and by laser physical vapor deposition, Conf. Record 23rd. Photovoltaic Solar Energy Conf, Louisville, IEEE Press, Piscataway, 394-399. Cusano D. A. (1963), 'CdTe solar cells and photovoltaic heteroj unctions in II-VI compounds', Solid State Electron. 6, 217-232. Edwards P. R., Halliday D. P., Durose K., Richter H. and Bonnet D. (1997), 'The influence of CdCl2 treatment and interdiffusion on grain boundary passivation in CdTe/CdS solar cells', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 2083-2086. Fahrenbruch A. F. (1987), 'Ohmic contacts and doping of CdTe', Solar Cells 21, 399-412. Ferekides C , Britt J., Ma Y. and Killian L. (1993), 'High efficiency CdTe solar cells by close spaced sublimation', Conf. Record 23rd. IEEE Photovoltaic Specialists Conf, Louisville, IEEE Press, Piscataway, 389-393. Ferekides C. S., Dugan K., Ceekala V., Killian J., Oman D., Swaminathan R. and Morel D. L. (1994), 'The effects of CdS processing and glass substrates on the performance of CdTe solar cells', Proc. 1st. World Conf. Photovoltaic Solar Energy Conversion, Waikoloa, IEEE Press, Piscataway, 99-103. Ferekides C. S., Marinsky D., Marinskaya S., Tetali B., Oman D. and Morel D. L. (1996), 'CdS films prepared by close-spaced sublimation and their influence on CdTe/CdS solar cell performance', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 751-756.

272

D. Bonnet

Ferekides C , Marinski D. and Morel D. L. (1997a), 'CdS: characterization and recent advances in CdTe solar cell performance', Conf. Record 26th. Photovoltaic Specialists Conf., Anaheim, IEEE, Press, Piscataway, 339-342. Ferekides C. S., Viswanathan V. and Morel D. L. (1997b), 'RF sputtered back contacts for CdTe/CdS thin film solar cells', Conf. Record 26th. Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 423^126. Ferekides C , Poorsola U. and Morel D. L. (1997c), 'The effect of Sn0 2 roughness on the properties of CdTe/CdS solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 427-430. Galloway S. A., Holland A. J., Wilshaw P. R., Brinkman A. W. and Durose K. (1995), 'Microscopic characterisation of CdS/CdTe thin film photovoltaic devices using scanning optical and electronic beam injection techniques', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S. Stephens & Associates, Bedford, 2073-2075. Gessert T. A., Mason A. R., Reedy R. C , Matson R., Coutts T. J. and Sheldon P. (1995), 'Development of RF sputtered Cu-doped ZnTe for use as a contact interface layer top-CdTe, /. Electron Mat. 24, 1443-1449. Ghandi S. K., Tsakar N. R., Bhat I. B. (1987), 'Arsenic-doped /?-CdTe layer grown by organometallic vapor phase epitaxy', Appl. Phys. Lett. 50, 900-902. Granata J. E., Sites J. R., Contreras-Puente G. and Compaan A. D. (1996), 'Effect of CdS thickness on CdS/CdTe quantum efficiency', Conf. Record 25th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 853-856. Hartmann H., Mach R. and Selle B. (1981), 'Wide gap II-VI compounds as electronic materials', in Kaldis E., ed., Current Topics in Materials Science, Amsterdam, pp. 1-414. Jensen D. G., McCandless B. E. and Birkmire R. W. (1996), 'Thin film cadmium telluride cadmium sulfide alloys and devices', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 773-776. Laurengo M. A., Yew Y. K., Homewood P. K., Durose, K., Richter, H. and Bonnet D. (1997), 'Deep level transient spectroscopy of CdS/CdTe thin film solar cells', J. Appl. Phys. 82, 1423-1426. Lebrun J. (1965), 'Realisation et propriete des photopiles solaires en couches minces de tellure de cuivre et tellure de cadmium', Rev. Phys. Appliquee 1, 204-208. Leng G., Dignard-Bailly L., Tamizhmani G., Usher E. and Bargagnolo J. (1997), Overview of the Worldwide Photovoltaic Industry, Report No. EDRL 96-41-A1 (TR), Canada Centre for Mineral and Energy Technology.

Cadmium Telluride Solar Cells

273

Levi D. H., Moutinho H. R., Hasoon F. A., Keyes B. M., Ahrenkiel R. K. and Kazmerski L. L. (1994), 'Micro through nanostructure investigation of polycrystalline CdTe: correlations with processing and electronic structures', Proc. 1st. World Conf. Photovoltaic Solar Energy Conversion, Waikoloa, IEEE Press, Piscataway, 127-131. Marshall J. (1997), 'BP Investing $20 million in Fairfield operation', San Francisco Chronicle, 20 May 1997. McCandless B. E., Qu Y. and Birkmire R. W. (1995), 'A treatment to allow contacting CdTe with different conductors, Proc. 1st. World Conf. Photovoltaic Solar Energy Conversion, Waikoloa, IEEE Press, Piscataway, 107-110. McCandless B. E., Hichri H., Hanket G. and Birkmire R. W. (1996), 'Vapor phase treatment of CdTe/CdS thin films with CdCl 2 :0 2 ', Conf. Record 25th. Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 781-784. McCandless B. E. and Birkmire R. W. (1997), 'CdTe,.A absorber layers for thin-film CdTe/CdS solar cells', Conf. Record 26th. Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 307-312. Meyers P. V. and Liu C. H. (1988), 'Progress toward development of a production process for thin film CdTe solar modules', Proc. 8th. European Photovoltaic Solar Energy Conf, Florence, Kluwer Academic Publishers, Dordrecht, 1588-1593, and Solar Cells 23,59-67. Meyers P. V., Liu C. H. and Frey T. J. (1998), 'Method of making photovoltaic cell with chloride dip', U.S. Patent No. 4,873,198. Meyers P. V., Liu C. H. and Doty M. (1990), 'Method of making heterojunctionp-j'-n cell, U.S. Patent No. 4,977,097. Meyers P. V., Zhou T., Powell R. C. and Reiter N. (1993), 'Elemental vapor-deposited polycrystalline CdTe thin film photovoltaic modules', Conf. Record 23rd. IEEE Photovoltaic Specialists Conf, Louisville, IEEE Press, Piscataway, 400-404. Moskowitz P. D., Steinberger H. and Thumm W. (1994), 'Health and environmental hazards of CdTe photovoltaic module production, use and decommissioning', Proc. 1st. World Conf. Photovoltaic Solar Energy Conversion, Waikoloa, IEEE Press, Piscataway, 115-118. Niles D. W., Li X. N., Albin D., Rose D., Gessert T. and Sheldon P. (1996), 'Evaporated Te on CdTe: a vacuum-compatible approach to making back contact to CdTe solar cell devices, Progr. in Photovoltaics 4, 225-229. Nishio T. (1996), 'Thin film CdS/CdTe solar cell with 15.05% efficiency', Conf. Record 25th. IEEE Photovoltaic Solar Energy Conf, Washington D.C., IEEE Press, Piscataway, 953-956.

274

D. Bonnet

Nunoue Sh., Hemmi T. and Kato E. (1990), 'Mass spectrometric study of the phase boundaries of the CdS-CdTe system', J. Electrochem. Soc. 137, 1248-1251. Ohyama H., Aramoto T., Kumazawa S., Higuchi H., Arita T., Shibutani S., Nishio T., Nakayama J., Tsuji M., Hanafusa A., Hibino T., Omura K. and Murozono M. (1997), '16.0% efficient thin-film CdS/CdTe solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 343-346. Patterson M. H., Turner A. K., Sadeghi M. and Marshall R. J. (1994), 'Health, safety and environmental aspects of the production and use of CdTe thin-film photovoltaic modules', Proc. 12th. European Photovoltaic Solar Energy Conf., Amsterdam, H. S. Stephens & Associates, Bedford, 951-953. Ponpon J. P. (1985), 'A review of ohmic and rectifying contacts on cadmium telluride, Solid State Electronics 28, 689-706. Reetz T. (1993), 'ProzeBkettenanalyse zum CdTe Solarmodul', Forschungszentrum Jiilich, Interner Bericht KFA-STE-IB-2/93. Rohatgi A. (1992), 'A study of efficiency-limiting defects in polycrystalline CdTe/CdS solar cells', Int. J. Solar Energy 12, 3 7 ^ 9 . Sandwisch D. W. (1994), 'Development of CdTe module manufacturing', Proc. 1st. World Conf. Photovoltaic Solar Energy Conversion, Waikoloa, IEEE Press, Piscataway, 836-839. Sandwisch D. W. (1996), 'Progress in high-throughput manufacturing of thin-film CdTe modules', NREL/SNL Program Review, AIP Conf. Proc. 394,425^132. Sasala R. A., Powell R. C , Dorer G. L. and Reiter N. (1996), 'Recent progress in CdTe solar cell research at SCI', NREL/SNL Photovoltaics Program Review Meeting, Nov. 18-22, Lakewood CO, USA, APS Conf. Proc. 394, 171-188. Sites J. R. and Lui X. (1995), 'Performance comparison between polycrystalline thinfilm and single-crystal solar cells', Progr. in Photovoltaics 3, 307-314. Skarp J., Koskinen Y., Lindfors S., Rautiainen A. and Suntola T. (1991), 'Development and evaluation of CdS/CdTe thin filmPV cells', Proc. 10th. European Photovoltaic Solar Energy Conf, Lisbon, Kluwer Academic Publishers, Dordrecht, 567-569. Skarp J., Anttila E., Rautiainen A. and Suntola T. (1992), 'ALE CdS/CdTe PV cells', Int. J. Solar Energy 12, 137-142. Steinberger H. (1998), 'Environmental and health risks from the operation of CdTeand CIS-thin-film modules', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 2276-2278. Stollwerk G. and Sites J. (1995), 'Analysis of CdTe back-contact barriers', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S. Stephens & Associates, Bedford, 2020-2023.

Cadmium Telluride Solar Cells

275

Tang J., Mao D., Ohno T. R., Kaydanov V. and Trefny J. U. (1997), 'Properties of ZnTe:Cu thin films and CdS/CdTe/ZnTe solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 439-442. Tyan Y.-S. (1980), 'Semiconductor devices having improved low-resistance contacts to p-type CdTe, and method of preparation', U.S. Patent No. 4,319,069. Tyan Y.-S. and Perez-Albuerne E. A. (1982), 'Efficient thin film CdS/CdTe solar cells', Conf. Record 16th. IEEE Photovoltaic Specialists Conf, San Diego, IEEE Press, Piscataway, 794-800. Ullal H. S., Zweibel K. and Roedern B. V. (1977), 'Current status of polycrystalline thin-film PV technologies', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 301-305. Woodcock J. M., Turner A. K., Ozsan M. E. and Summers J. G. (1991), 'Thin film solar cells based on electrodeposited CdTe', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 842-847. Woodcock J. M., Ozsan M. E., Turner A. K., Cunningham D. W., Johnson D. R., Marshall R. J., Mason N. B., Otik S., Patterson M. H., Ransome S. J., Roberts S., Sadeghi M., Sherborne J. M., Sivapathasundaram D. and Walls I. A. (1994), 'Thin film CdTe photovoltaic cells', Proc. 12th. European Photovoltaic Solar Energy Conf, Amsterdam, H. S. Stephens & Associates, Bedford, 948-950. Woodcock J. M., Schage H., Maurus H., Dimmler B., Springer J. and Ricaud H. (1997), 'A study of the upscaling of thin film solar cell manufacture towards 500 MWp per annum', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 857-860. Wu X., Sheldon T. J., Coutts T. J., Rose D. H. and Moutinho H. R. (1997), 'Application of Cd 2 Sn0 4 transparent conducting oxide in CdS/CdTe thin-film devices', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 347-350. Yoshida T. (1995), 'Photovoltaic properties of screen-printed CdTe/CdS solar cells on indium-tin-oxide coated glass substrates', J. Electrochem. Soc. 142, 3232-3237. Zanio K. (1978), 'Cadmium telluride: materials preparation; physics: defects; applications', in Semiconductors and Semimetals, Vol. 13, Willardson R. K. and Beer A. C , eds., Academic Press, New York. Zhou T. X., Reiter N., Powell R. C , Sasala R. and Meyers P. V. (1994), 'Vapour chloride treatment of polycrystalline CdTe/CdS films', Proc. 1st. World Conf. Photovoltaic Solar Energy Conversion, Waikoloa, IEEE Press, Piscataway,. 103-106. Zweibel K., Ullal H. S. and von Roedern B. (1996), 'Progress and issues in polycrystalline thin film PV technologies', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 745-750.

CHAPTER 7

Cu(In,Ga)Se2 SOLAR CELLS UWE RAU and HANS W. SCHOCK Institutfiir Physikalische Elektronik (ipe), Universitdt Stuttgart, Pfaffenwaldring 47, D-70569, Stuttgart, Germany. [email protected], [email protected]

Mephistopheles:

Wer will was Lebendigs erkennen und beschreiben, Sucht erst den Geist heraus zu treiben, Damn hat er die Telle In der Hand, Fehlt leider! nur das geistige Band. Encheiresin naturae nennt's die Chemie, Spottet Ihrer selbst und weifi nlcht wle. Johann Wolfgang von Goethe, Faust, 1808.

7.1 Introduction From the early days of photovoltaics until today, thin-film solar cells have always competed with technologies based on single-crystal materials such as Si and GaAs. Owing to their amorphous or polycrystalline nature, thin-film solar cells always suffered from power conversion efficiencies lower than those of the bulk technologies. This drawback was and still is counterbalanced by several inherent advantages of thinfilm technologies. As in the early years of photovoltaics space applications were the driving force for the development of solar cells, the argument in favour of thin films was their potential lighter weight as compared to bulk materials. An extended interest in solar cells as a source of renewable energy emerged in the mid-seventies as the limitations of fossil energy resources were widely recognised. For terrestrial power applications the cost arguments and the superior energy balance strongly favoured thin films. However, from the various materials under consideration in the fifties and sixties, only four thin-film technologies, namely amorphous (a-)Si and the polycrystalline heteroj unction systems CdS/Cu^S, CdS/CdTe, and CdS/CuInSe2, entered pilot production. As yet, only a-Si contributes significantly to the world photovoltaic market, with a -12% share. Activities in the CdS/Cu^S system 277

278

U.RauandH. W. Schock

stopped at the beginning of the eighties when amorphous silicon became the front runner in thin-film technologies. The highest potential with regard to cost reduction and high efficiencies is provided by the heteroj unction solar cells based on CdTe and CuInSe2 absorbers. Both materials look back to around 30 years of research and development with the favourite attribute 'promising materials' at every stage. Nevertheless, only now, at the beginning of the new millennium, are these materials switching from the position of eternal talents towards that of mature technologies that have to prove their ability to keep their promises. CuInSe2 was synthesised for the first time by Hahn in 1953 (Hahn et al, 1953). In 1974, this material was proposed as a photovoltaic material (Wagner et al., 1974) with a power conversion efficiency of 12% for a single-crystal cell. Thin-film development achieved a historical milestone in the years 1980 to 1982, when Boeing Corp. boosted the efficiencies of thin film solar cells obtained from a three-source co-evaporation process from 5.7% (Mickelsen and Chen, 1980) to over 10% (Mickelsen and Chen, 1982). We will discuss the co-evaporation process in Section 7.3.1. The Boeing result was surpassed in 1987 by Arco Solar with a long-standing record efficiency for a thinfilm cell of 14.1% (Mitchell et at, 1988). They used a different approach for absorber preparation, namely the selenisation of stacked metal layers by H2Se (see Section 7.3.1). The lack of reproducibility and resulting low production yield considerably delayed the pilot production envisaged at that time. It took a further ten years, before Arco Solar, now Siemens Solar Industries in the USA, entered the stage of production. In 1998 they produced the first commercially available Cu(In,Ga)Se2 solar modules. In parallel, Siemens Solar in Germany is developing a process that avoids the use of H2Se. Other players in the USA are Global Solar and ISET, who plan to commercialise modules prepared on other than glass substrates. In Europe, the longterm development efforts of the EUROCIS consortium on the co-evaporation process resulted in the activity by Wiirth Solar with pilot production envisaged in 2000. In Japan, two lines for film preparation are planned by Showa (selenisation by H2Se) and Matshushita (co-evaporation). With an NREL-demonstrated cell efficiency of 18.8% on a 0.5 cm2 laboratory cell (Contreras et a/., 1999) and 14.7% for mini-modules with an area of around 20 cm2 demonstrated by Siemens Solar (Karg, 1999) and Angstrom Solar Centre, Sweden (Kessler et al, 1999), Cu(In,Ga)Se2 is now by far the most efficient thin-film solar cell technology. The start of production at several places provides a new challenge also for research on this material. In contrast to all other solar cell technologies, research on Cu(In,Ga)Se2 has no backup from other applications of the same material. As a consequence, the body of knowledge on Cu(In,Ga)Se2 is still slight compared with what is known about crystalline Si, which can draw on ample reserves of

Cu(In, Ga)Se2 Solar Cells

279

knowledge from microelectronics research. Hence, most of the achievements of Cu(In,Ga)Se2-based solar cells accumulated over the years have been initiated by intuition rather than knowledge-based technological design. This chapter aims to summarise our present knowledge of Cu(In,Ga)Se2-based heterojunction solar cells. We focus on four main areas: (i) The description of basic material properties such as crystal properties, phase diagram, and defect physics in Section 7.2; (ii) cell technology starting from the growth of the polycrystalline Cu(In,Ga)Se2 absorber up to device finishing by heterojunction formation and window layer deposition (Section 7.3). This section also discusses options which can be used to design the electronic properties of the absorber material as well as basic technologies for module production; (iii) the electronic properties of the finished heterostructure and some methods of analysing them (Section 7.4); (iv) finally, Section 7.5 discusses the photovoltaic potential of wide-gap chalcopyrites, namely CuGaSe2 and CuInS2, as well as that of the pentenary alloy system Cu(In,Ga)(S,Se)2 and the possibility of building graded-gap structures with these alloys. This chapter can only very briefly cover those scientific issues that are relevant for photovoltaic applications. For other important points and for more detailed information, we refer the reader to the literature. More about the structural properties of Cu(In,Ga)Se2 can be found, e.g. in Shay and Wernick (1975), Kazmerski and Wagner (1985), Coutts et al. (1986), Rockett and Birkmire (1991), Schock (1996), Bube (1998) and Rau and Schock (1999). Interface properties of Cu(In,Ga)Se2 and related compounds were recently reviewed by Scheer (1997). For up-scaling and module technologies see, for example, Dimmler and Schock (1996), and for economic aspects, see Zweibel (1995).

7.2 Material properties 7.2.1 Basics CuInSe2 and CuGaSe2, the materials that form the alloy Cu(In,Ga)Se2, belong to the semiconducting I—III—VI2 materials family that crystallise in the tetragonal chalcopyrite structure. The chalcopyrite structure of, for example, CuInSe2 is obtained from the cubic zinc blende structure of II-VI materials like ZnSe by occupying the Zn sites alternately with Cu and In atoms. Figure 7.1 compares the two unit cells of the cubic zinc blende structure with the chalcopyrite unit cell. Each I (Cu) or III (In) atom has four bonds to the VI atom (Se). In turn each Se atom has two bonds to Cu and two to In. Because the strengths of the I-VI and III-VI bonds are in general different, the

280

U.RauandH. W. Schock

(a)

(b)

Se )Se

In

Zn

Cu

Figure 7.1 Unit cells of chalcogenide compounds, (a) Sphalerite or zinc blende structure of ZnSe (two unit cells); (b) chalcopyrite structure of CuInSe2. The metal sites in the two unit cells of the sphalerite structure of ZnSe are alternately occupied by Cu and In in the chalcopyrite structure.

ratio of the lattice constants cla is not exactly 2. The quantity 2 - da (which is -0.01 in CuInSe2, +0.04 in CuGaSe2) is a measure of the tetragonal distortion in chalcopyrites. The band-gap energies of I-UI-VI 2 chalcopyrites are considerably smaller than those of their binary analogues (this is the binary material where the I/III elements are replaced by their average II element; thus ZnSe is the binary analogue of CuGaSe2 and Zno.5Cdo.5Se that of CuInSea). This difference is because the Cu 3d band, together with n

2.4

-

1

2.2

CuGaS2

-

2.0 1.8

\ _

UJeV 1.6

CuGaSe2

-

CulnS2 1

1.4 CulnSe 2 .

1.2

"^

1.0 5.4

Figure 7.2

5.5

5.6 a/Angstroms

5.7

5.8

Band-gap energies Us vs. lattice constants a of the Cu(In,Ga)(S,Se)2 system.

Cu(In,Ga)Se2 Solar Cells

281

the Se Ap band, forms the uppermost valence band in the Cu-chalcopyrites, which is not so in II-VI compounds. However, the system of copper chalcopyrites covers a wide range of band-gap energies Ug from 1.04 eV in CuInSe2 up to 2.4 eV in CuGaS2, covering most of the visible spectrum. Figure 7.2 summarises lattice constants a and band-gap energies Ug of this system. Any desired alloys between these compounds can be produced, as there is no miscibility gap in the entire system. We will discuss the status and prospects of this system in more detail in Section 7.5.

7.2.2

Phase diagram

Compared with all other materials used for thin-film photovoltaics, Cu(In,Ga)Se2 has by far the most complicated phase diagram. Figure 7.3 displays the ternary phase diagram, which comprises all ternary Cu-In-Ga compounds. This complex ternary phase diagram can be reduced to a simpler pseudo-binary phase diagram along the tie line between Cu2Se and In2Se3 (solid line in Fig. 7.22). Figure 7.4 shows the phase diagram of CuInSe2 given by Haalboom et al. (1997). This investigation had a special focus on temperatures and compositions relevant for the preparation of thin-films. The phase diagram in Fig. 7.4 shows the four different phases which have been found to be

90

80

70

60

50

40

30

20

10

Cu„ln 9

Figure 7.3 Ternary phase diagram of the Cu-In-Se system.

U. Rau andH. W. Schock

282

relevant in this range: the a-phase (CuInSe2), the /3-phase (CuIn3Se5), the 8-phase (the high-temperature sphalerite phase) and Cu2_j.Se. An interesting point is that all neighbouring phases to the a-phase have a similar structure. The /3-phase is actually a defect chalcopyrite phase built by ordered arrays of defect pairs (Cu vacancies Vc„ and In-Cu antisites InCu). Similarly, Cu2_j.Se can be viewed as constructed from the chalcopyrite by using Cu-In antisites CuT„ and Cu interstitials Cui. The transition to the sphalerite phase arises from disordering the cation (Cu, In) sub-lattice, and leads back to the zinc blende structure (cf. Fig. 7.1a).

ft? . L+8

20

25

30

Cu content (at%) o

single-phase region

•

DTA cooling

a>

two-phase region

A

DTA heating

Figure 7.4 Quasi-binary phase diagram of CuInSe2 established by Differential Thermal Analysis (DTA) and microscopic phase analysis. Note that at 25% Cu no single phase exists. After Haalboom el at. (1997).

The existence range of the a-phase in pure CuInSe2 on the quasi-binary tie line Cu2Se-In2Se3 extends from a Cu content of 24% to 24.5%. Thus, the existence range of single-phase CuInSe2 is astonishingly small and does even not include the stoichiometric composition of 25% Cu. The Cu content of absorbers for thin-film solar cells typically varies between 22 and 24 at. % Cu. At the growth temperature this region lies within the single-phase region of the a-phase. However, at room temperature it lies in the two-phase a + /? region of the equilibrium phase diagram in

Cu(ln,Ga)Se2 Solar Cells

283

Haalboom et al. (1997). Hence one would expect a tendency for phase separation in photovoltaic-grade CuInSe2 after deposition. Fortunately, it turns out that partial replacement of In with Ga, as well as the use of Na-containing substrates, considerably widens the single-phase region in terms of (In + Ga)/(In + Ga + Cu) ratios (Herberholz et al, 1999). Thus, the phase diagram hints at the substantial improvements actually achieved in recent years by the use of Na-containing substrates, as well as by the use of Cu(In,Ga)Se2 alloys.

7.2.3

Defect physics of Cu(In, Ga)Se2

Basics The role of defects in the ternary compound CuInSe2, and even more in Cu(In,Ga)Se2, is of special importance because of the large number of possible intrinsic defects and the role of deep recombination centres in the performance of the solar cells. For insight into the defect physics of Cu(In,Ga)Se2, see Cahen (1987), and for a recent discussion see Burgelman et al. (1997). The challenge of defect physics in Cu(In,Ga)Se2, according to Zhang et al. (1998), is to explain three unusual effects in this material: (i) the ability to dope Cu(In,Ga)Se2 with native defects; (ii) the structural tolerance to large off-stoichiometries; and (iii) the electrically neutral nature of the structural defects. It is obvious that the explanation of these effects significantly contributes to the explanation of the photovoltaic performance of this material. It is known that the doping of CuInSe2 is controlled by intrinsic defects. Samples with ptype conductivity are grown if the material is Cu-poor and annealed under high Se vapour pressure, whereas Cu-rich material with Se deficiency tends to be n-type (Migliorato et al., 1975; Noufi et al, 1984). Thus, the Se vacancy Vse is considered to be the dominant donor in n-type material (and also the compensating donor in p-type material), and the Cu vacancy VCu the dominant acceptor in Cu-poor p-type material. Theoretical considerations By calculating the metal-related defects in CuInSe2 and CuGaSe2, Zhang et al. (1998) found that the defect formation energies for some intrinsic defects are so low that they can be heavily influenced by the chemical potential of the components (i.e., by the composition of the material) as well as by the electrochemical potential of the electrons. For VCu in Cu-poor and stoichiometric material, a negative formation energy is even calculated. This would imply the spontaneous formation of large

284

U.RauandH. W. Schock

numbers of these defects under equilibrium conditions. Low (but positive) formation energies are also found for the Cu-on-In antisite Cuto in Cu-rich material (this defect is a shallow acceptor which could be responsible for the />type conductivity of Cu-rich, non-Se-deficient CuInSe2). The dependence of the defect formation energies on the electron Fermi level could explain the strong tendency of CuInSe2 to selfcompensation and the difficulties of achieving extrinsic doping. The work of Zhang et al. (1998) provides a good theoretical basis for the calculation of defect formation energies and defect transition energies, which exhibit good agreement with experimentally obtained data. Further important results in Zhang et al. (1997) are the formation energies of defect complexes such as (2Vcu,InCu), (Cu^Inc) and (ZCu^Cu^), where CUJ is an interstititial Cu atom. These formation energies are even lower than those of the corresponding isolated defects. Interestingly, (2Vcu,InCu) does not exhibit an electronic transition within the forbidden gap, in contrast to the isolated InCu-anti-site, which is a deep recombination centre. As the (2Vcu,InCu) complex is most likely to occur in Inrich material, it can accommodate a large amount of excess In (or likewise deficient Cu) and, at same time, maintain the electrical performance of the material. Furthermore, ordered arrays of this complex can be thought as the building blocks of a series of Cu-In-Se compounds such as CuIn3Se5 and CuIn5Se8 (Zhang et al., 1997). Table 7.1 Electronic transition energies and formation energies of the twelve intrinsic defects in CuInSe2 Defect transition energies" and formation energies''

Trans -ition

Vcu

Vi„

HO)

0.03

0.17

0.03

0.04

H2-)

0.41

(2-/3-)

0.67

V«

Cm In,

Se,

Cum

Sec

Cus«

Se.„

0.75

0.08

Inse

0.29 0.04c

0.07

0.05 0.58

0.2

(0/+)

0.25

0.11'

0.08

0.07

2.6

2.88

9.1

0.06

0.04

0.09

0.44

(+/2+) Atf/eV

Incu

0.60

3.04

2.9

2.8

4.4

22.4

3.34

1.54

1.4

7.5

7.5

7.5

5.5

5.0

"Difference between the valence/conduction band energy for acceptor/donor states; formation energy At/ of the neutral defect in the stoichiometric material; ccovalent; rfionic. All energies in eV. Source: The ionisation energies in italics are derived from Abou-Elfotouh et al. (1991), and the formation energies in brackets from Neumann (1983). All the bold numbers are from Zhang et al. (1998).

Cu(ln,Ga)Se2

285

Solar Cells

Table 7.1 summarises the ionisation energies and the defect formation energies of the twelve intrinsic defects in CulnSe2. The energies given in bold type for VCu, V,„, Cm, Cm„ and Inc„ were obtained from a first-principles calculation, whereas the formation energies in italics were calculated from the macroscopic cavity model. The numbers in the first column represent the ionisation states of the defects. Device-relevant

defects

Let us now concentrate on the defects experimentally detected in photovoltaic grade (and thus In-rich) polycrystalline films. In-rich material is in general highly compensated, with a net acceptor concentration of the order of 1016 cm" . The shallow acceptor level Vc„ (which lies about 30 meV above the valence band) is assumed to be the main dopant in this material. As compensating donors, the Se-vacancy Vsc as well as the double donor InCu are considered. The most prominent defect is an acceptor level about 270-300 meV above the valence band, which is reported by several groups from deep-level transient spectroscopy (Igalson and Schock, 1996) and admittance spectroscopy (Schmitt et al., 1995; Walter et al., 1996b). This defect is also present in single crystals (Igalson et al., 1995).

high-energy tail D, ~exp(-l//l/*)

£ KT1

0.1

0.2 0.3 0.4 0.5 Activation energy C/eV

0.6

Figure 7.5 Defect density spectrum obtained from admittance spectroscopy of a ZnO-CdS-CuInGaSe2 heterojunction. The peaks Ni and N2 can be related to interface and bulk defects (see inset).

As an example. Fig. 7.5 displays a defect density spectrum obtained from admittance spectroscopy by the method of Walter et al. (1996a). The transition at -300 meV exhibits a broadened energy distribution with a tail in the defect density

286

U. Rau and H. W. Schock

towards larger energies. This tail-like distribution is best described by a characteristic energy U*, as shown in Fig. 7.5. This defect is detected, not only in In-rich, but in equal amounts also in Cu-rich polycrystalline materials (Herberholz, 1998). An assignment of this defect to the Cu^ anti-site is in agreement with the theoretical calculations of Zhang et al. (1998) as well as with the proposition of several experimentalists. The importance of this transition derives from the fact that its concentration is related to the open-circuit voltage of the device (Herberholz et al, 1997a) and that the defect seems to be involved in the defect metastability (Igalson and Schock, 1996) (cf. Section 7.4.6). The lower-energy transition in Fig. 7.5 is attributed to interfacial defects, rather than to a bulk defect (Herberholz et al., 1998) because its activation energy can vary between 50 meV and 250 meV depending on air-annealing prior to the measurement (Rau et al, 1999a). Thus, the activation energy of this transition measures the depth At/ Fn from the vacuum level of the (electron) Fermi level and the conduction-band energy at the Cu(In,Ga)Se2 surface (Herberholz et al, 1998), as shown in the inset of Fig. 7.5.

7.3 7.3.1

Cell and module technology Structure of the heterojunction solar cell

The complete layer sequence of a ZnO/CdS/Cu(In,Ga)Se2 heterojunction device is shown in Fig. 7.6. It consists of a typically 1 fjm thick Mo layer deposited on a sodalime glass substrate and serving as the back contact for the solar cell. The Cu(In,Ga)Se2 is deposited on top of the Mo back electrode as the photovoltaic absorber material. This layer has a thickness of 1-2 /an. The heterojunction is then completed by chemical bath deposition (CBD) of CdS (typically 50 nm) and by the sputter deposition of a nominally undoped (intrinsic) /-ZnO layer (usually of thickness 50-70 nm) and then a heavily doped ZnO layer. As ZnO has a band-gap energy of 3.2 eV it is transparent for the main part of the solar spectrum and therefore is denoted as the window layer of the solar cell. We will first mention four important technological innovations which, during the last decade, have led to a considerable improvement of the efficiencies and finally to the record efficiency of 18.8% (Contreras et al, 1999). These steps are the key elements of the present Cu(In,Ga)Se2 technology.

Cu(ln,Ga)Se2

287

Solar Cells

,ZnO:AI - /-ZnO .CdS - Cu(ln,Ga)Se2 .Mo

. glass

^ Figure 7.6

Schematic layer sequence of a standard ZnO-CdS-Cu(In,Ga)Se2 thin-film solar cell.

1. The film quality has been substantially improved by the crystallisation mechanism induced by the presence of CuvSe (y < 2). This process is further supported by a substrate temperature close to the softening point of the glass substrate (Stolt et al., 1993). 2. The glass substrate has been changed from Na-free glass to Na-containing sodalime glass.(Hedstrom et al., 1993; Stolt et al., 1993). This has led to an enormous improvement of the efficiency and reliability of the solar cells, as well as to a larger process tolerance. It was first assumed that this improvement was due to better match of thermal expansion coefficients, but the beneficial impact of Na—diffusing from the substrate through the Mo back contact—on the growth of the absorber layer and its structural and electrical properties was soon recognised. 3. Initially, the absorbers consisted of pure CuInSe2- The partial replacement of In with Ga (Devaney et al., 1990) is a further noticeable improvement, which has increased the band gap of the absorber from 1.04 eV to 1.1-1.2 eV for the highefficiency devices. The benefit of 20-30% Ga incorporation stems not only from the better band-gap match to the solar spectrum but also from the improved electronic quality of Cu(In,Ga)Se2 with respect to pure CuInSe2 (Hanna et al., 2000; Herberholz etal., 1999). 4. The counter electrode for the CuInSe2 absorber of the earlier cells was a 2 ,um thick CdS layer laid down by Physical Vapour Deposition (PVD). This has been replaced by a combination of a 50 nm thin CdS buffer layer laid down by chemical bath deposition (Potter et al., 1985; Birkmire et al., 1989; Mauch et al, 1991) and a highly conductive ZnO window layer.

288

U. Rau andH. W. Schock

The effect of items 1-4 on the electronic properties and performance of Cu(In,Ga)Se2 solar cells will be considered in detail below, as we discuss the preparation of a Cu(In,Ga)Se2 solar cell step by step.

7.3.2

Absorber preparation techniques

Basics The preparation of Cu(In,Ga)Se2-based solar cells starts with the deposition of the absorber material on a Mo-coated glass substrate (preferably soda-lime glass). The properties of the Mo film and the choice of the glass substrate are of primary importance for the final device quality, because of the importance of Na, which diffuses from the glass through the Mo film into the growing absorber material. Some processes use blocking layers such as SiN^, Si0 2 or Cr between the glass substrate and the Mo film to prevent the out-diffusion of Na. Instead, Na-containing precursors like NaF, Na2Se or NaS are now deposited prior to absorber growth to provide a controlled, more homogeneous, incorporation of Na into the film. The control of Na incorporation in the film from precursor layers allows the use of other substrates like metal or polymer foils. There seems to be no fundamental efficiency limitation due to the substrate provided a proper supply of sodium is provided. During absorber deposition, a MoSe2 film forms at the Mo surface (Wada et ah, 1996; Takei etal., 1996). MoSe2 is a layered semiconductor withp-type conduction, a band gap of 1.3 eV and weak van der Waals bonding along the c-axis. If the layer were oriented parallel to the plane of contact, the MoSe2 would inhibit adhesion of the absorber as well as leading to unfavourable electronic transport. Fortunately, the caxis is found to be in parallel with, and the van der Waals planes thus perpendicular to the interface (Wada et al., 1996). Because of the larger band gap of the MoSe2 compared with that of standard Cu(In,Ga)Se2 films, the MoSe2 layer provides an electronic mirror for the photogenerated electrons and at the same time provides a low-resistance contact for the holes (see Section 7.4.1). Photovoltaic-grade Cu(In,Ga)Se2 films have a slightly In-rich overall composition. The allowed stoichiometry deviations are astonishingly large, yielding a wide process window with respect to composition. Devices with efficiencies above 14% are obtained from absorbers with (In + Ga)/(In + Ga + Cu) ratios between 52 and 64% if the sample contains Na (Ruckh et al., 1994a). Cu-rich Cu(In,Ga)Se2 shows the segregation of a secondary Cu2.ySe phase preferentially at the surface of the absorber film. The metallic nature of this phase does not allow the formation of efficient

Cu(In,Ga)Se2

Solar Cells

289

Vapour

Fte-es/aporation Conderealion ,

Liquid

QoAth /

Solid

/ QjInSea

Substrate

Figure 7.7 Schematic illustration of the growth of a Cu(In,Ga)Se2 film under Cu-rich conditions. A quasi-liquid Cu-Se phase acts as a flux in a vapour-liquid solid growth mechanism.

heterojunctions. Even after removal of the secondary phase from the surface by etching the absorber in KCN, the utility of Cu-rich material for photovoltaic applications is limited, probably due to the high doping density of 1018 cm -3 in the bulk and the surface defects. However, the importance of the Cu-rich composition is given by its role during film growth. Cu-rich films have grain sizes in excess of 1 /an whereas In-rich films have much smaller grains. A model for the film growth under Cu-rich compositions comprises the role of Cu2o,Se as a flux agent during the growth process of co-evaporated films (Klenk et al., 1993). This model for the growth of Cu(In,Ga)Se2 in the presence of a quasi-liquid surface film of CuySe is highlighted in Fig. 7.7. For Cu(In,Ga)Se2 prepared by selenisation, the role of C^-^Se is similar (Probst et al., 1996), therefore growth processes for high quality have to go through a copper-rich stage and end with an indium-rich composition. Co-evaporation processes The absorber material yielding the highest efficiencies is Cu(In,Ga)Se2 with a Ga/(Ga + In) ratio of -20%, prepared by co-evaporation from elemental sources. Figure 7.8 sketches a co-evaporation set-up as used for the preparation of laboratoryscale solar cells and mini-modules. The process requires a maximum substrate temperature of -550 C for a certain time during film growth, preferably towards the end of growth. One advantage of the evaporation route is that material deposition and film formation are performed during the same processing step. A feed-back loop

290

U.RauandH. W. Schock substrate heater thermo element

chalcogen sources Figure 7.8 Arrangement for the deposition of Cu(In,Ga)Se2 films by co-evaporation on a heated substrate. The rates of the sources are controlled by mass spectrometry.

based on a quadrupole mass spectrometer or an atomic absorption spectrometer controls the rate of each source. The composition of the deposited material with regard to the metals corresponds to their evaporation rates, whereas Se is always evaporated in excess. This precise control over the deposition rates allows for a wide range of variations and optimisations with different sub-steps or stages for film deposition and growth. These sequences are defined by the evaporation rates of the different sources and the substrate temperature during the course of deposition. Figure 7.9 illustrates some of the possibilities, starting with a simple single-step process where all rates as well as the substrate temperature are kept constant during the whole process (Fig. 7.9a). Advanced preparation sequences always include a Cu-rich stage during the growth process and end up with an In-rich overall composition in order to combine the large grains of the Cu-rich stage with the otherwise more favourable electronic properties of the In-rich composition. The first example of this kind of procedure is the so-called Boeing or bilayer process (Mickelsen and Chen, 1980), which starts with the deposition of Cu-rich Cu(In,Ga)Se2 and ends with an excess In rate, as illustrated in Fig. 7.9b. Another possibility is the inverted process where first (In,Ga)2Se3 (likewise In, Ga, and Se from elemental sources to form that compound) is deposited at a lower temperatures (typically around 300 C). Then Cu and Se are evaporated at an

Cu(ln,Ga)Se2 Solar Cells

291

elevated temperature until an overall composition close to stoichiometry is reached (Kessler et al, 1992). This process leads to a smoother film morphology than the bilayer process. The most successful version of the inverted process is the so-called three-stage process (Gabor et al., 1994) shown in Fig. 7.9c. This process puts the deposition of In, Ga, and Se at the end of an inverted process to ensure the overall Inrich composition of the film even if the material is Cu-rich during the second stage. The three-stage process currently leads to the best solar cells. Variations of the Ga/Inratio during deposition, as shown in Fig. 7.9d, allow the design of graded band-gap structures. (Gabor et al., 1996).

a) single layer

h.Go

F.

Cu

6C0 3C0 Time b) bi-layer

Time a) g r a d e d b a n d g a p

Time

Time

Figure 7.9 Schematic rate and substrate temperature profiles for co-evaporation processes. All processes lead to single-phase films, (a) Single-layer process without Cu-rich growth step; (b) bilayer process ('Boeing recipe') with Cu-rich growth at the start; (c) three-stage inverted process with intermediate Cu-rich growth; (d) growth of graded-gap films under Cu-poor conditions

Selenisation

processes

The second class of absorber preparation routes is based on the separation of deposition and compound formation into two different processing steps. High efficiencies are obtained from absorber prepared by selenisation of metal precursors in H2Se (Binsma and Van der Linden, 1982; Chu et al, 1984; Kapur et al, 1987) and by rapid thermal processing of stacked elemental layers in a Se atmosphere (Probst et al.,

U. Rem and H. W. Schock

292

1996). These sequential processes have the advantage that approved large-area deposition techniques such as sputtering can be used for the deposition of the materials. The Cu(In,Ga)Se2 film formation then requires a second step, the selenisation. The very first large-area modules were prepared by the selenisation of metal precursors in the presence of H2Se more than ten years ago (Mitchell et al., 1988). Today, a modification of this process is providing the first commercially available Cu(In,Ga)Se2 solar cells, manufactured by Siemens Solar Industries. This process is

DEPOSITION

heating

SELENISATION

H2S inlet

Figure 7.10 Illustration of the sequential process. Stacked metal layers are selenised and converted into CulnSe2 in FhSe atmosphere.

schematically drawn in Fig. 7.10. First, a stacked layer of Cu, In and Ga is sputterdeposited on the Mo-coated glass substrate. Then selenisation takes place under H2Se. To improve device performance, a second thermal process under H2S is added, resulting in an absorber that is Cu(In,Ga)(S,Se)2 rather than Cu(In,Ga)Se2. A variation of this method that avoids the use of the toxic H2Se during selenisation is the rapid thermal processing of stacked elemental layers (Probst et al., 1996). Here the precursor includes a layer of evaporated elemental Se. The stack is then selenised

Cu(In,Ga)Se2 Solar Cells

293

by a rapid thermal process (RTP) in either an inert or a Se atmosphere. The highest efficiencies are obtained if the RTP is performed in an S-containing atmosphere (either pure S or H2S). On the laboratory scale, the efficiencies of cells made by these preparation routes are smaller by about 3% (absolute) as compared with the record values. However, on the module level, co-evaporated and sequentially prepared absorbers have about the same efficiency. Sequential processes need two or even three stages for absorber completion. These additional processing steps may counterbalance the advantage of easier element deposition by sputtering. Also the detailed and sophisticated control over composition and growth achieved during co-evaporation is not possible for the selenisation process. Fortunately, the distribution of the elements within the film grown during the selenisation process turns out to be close to what one could think to be an optimum, especially if the process includes the sulphurisation stage. Since the formation of CuInSe2 is much faster than that of CuGaSe2, and because film growth starts from the top, Ga is concentrated towards the back surface of the film. An increasing Ga content implies an increase in band-gap energy. This introduces a socalled back-surface field, improving carrier collection at the same time as minimising back-surface recombination. In turn, S from the sulphurisation step is found preferentially towards the front surface of the film, where it reduces recombination losses and also increases the absorber band gap in the space-charge region of the heteroj unction. Other absorber deposition processes Besides selenisation and co-evaporation, other deposition methods have been studied, either to obtain films with very high quality or to reduce cost of film deposition on large areas. Methods that are used to form epitaxial III-V compound films, such as molecular beam epitaxy (MBE) (Niki et al., 1994) or metal organic chemical vapour deposition (MOCVD) (Gallon et al., 1998) have revealed interesting features for fundamental studies, such as phase segregation and defect formation, but cannot be used to form the base material for high-efficiency solar cells. Attempts to develop so-called low cost processes include electrodeposition, (Abken et al., 1998; Lincot et al., 1998) screen printing and particle deposition (Eberspacher et al., 1998). Electrodeposition can be carried out in either one or two steps. The crucial step is final film formation in a high-temperature annealing process. The recrystallisation process competes with the decomposition of the material, so process optimisation is quite difficult. Cells with good efficiencies were obtained by electrodeposition of a Cu-rich CuInSe2 film and subsequent conditioning by a vacuum

294

U.RauandH. W. Schock

evaporation step of In(Se) (Ramanathan et al, 1998a). Films prepared by spray pyrolysis did not lead to high-performance devices. Influence of sodium The outstanding role played by Na in the growth of Cu(In,Ga)Se2 films was realised some years ago (Hedstrom et al, 1993; Stolt et al, 1993; Ruckh et al, 1994a). In most cases, the Na comes from the glass substrate and diffuses into the absorber. But there are also approaches where Na is incorporated by the use of Na-containing precursors such as NaSe (Holz et al., 1994; Nakada et al, 1997), Na 2 0 2 (Ruckh et al, 1994a), NaF (Contreras et al, 1997a) or Na2S (Nakada et al, 1998). Other alkali precursors have been investigated by Contreras et al. (1997), who found that Nacontaining precursors yielded the best cell efficiencies. The most obvious effects of Na incorporation are better film morphology and higher conductivity of the films (Ruckh et al, 1994a). Furthermore, the incorporation of Na induces beneficial changes in the defect distribution of the absorber films (Keyes et al, 1997; Rau et al, 1998b). The explanations for the beneficial impact of Na are manifold, and it is most likely that the incorporation of Na in fact results in a variety of consequences. During film growth, the incorporation of Na leads to the formation of NaSe* compounds. This slows down the growth of CuInSe2 and could at same time facilitate the incorporation of Se into the film (Braunger et al, 1998b). Also the widening of the existence range of the oc-(CuInSe2) phase in the phase diagram, discussed above, as well as the reported larger tolerance to the Cu/(In + Ga) ratio of Na-containing thin films, could be explained in this picture. Furthermore, the higher conductivity of Na-containing films could result from the diminished number of compensating Vse donors. Wolf et al. (1998) investigated the influence of Na incorporation on the formation of CuInSe2 films from stacked elemental layers by means of thin-film calorimetry. The addition of Na inhibits the growth of CuInSe2 at temperatures below 380 C. The retarded phase formation is responsible for the better morphology in the case of Na-containing samples. Another explanation put forward by Kronik et al. (1998) is that Na promotes oxygenation and passivation of grain boundaries. This could account for the observed enhancement of the net film doping by Na incorporation, through the diminished positive charge at the grain boundaries. It has in fact been observed that the surfaces of Na-containing films are more prone to oxygenation than are Na-free films (Braunger et al, 1998a).

Cu(In,Ga)Se2 Solar Cells

295

The above explanations deal with the role of Na during growth. However, the amount of Na in device-quality Cu(In,Ga)Se2 films is of the order 0.1 at%, which is a concentration of 1020 cm"3 (Niles et al., 1997), and one may ask the question: where are these tremendous quantities of Na in the finished absorber? The electronic effect, i.e. the change of effective doping resulting from Na incorporation, is achieved at concentrations of ~1016 cm"3, four orders of magnitude below the absolute Na content. It has long been believed that the main part of the Na is situated at the film surface and the grain boundaries. Final evidence for this hypothesis was recently found by Niles et al. (1999) with the help of high spatial resolution Auger electron spectroscopy. Heske et al. (1996, 1997) investigated the behaviour of Na on the surface of polycrystalline Cu(In,Ga)Se2 films by X-ray photoelectron spectroscopy (XPS). They found two different species of Na: (i) The first, denoted 'reacted', was observed on the air-exposed sample or after storing the sputter-cleaned sample for three days in an ultra-high vacuum (UHV). The second, denoted 'metallic', was found on clean samples either after annealing at 410 K in UHV or after deliberate Na deposition from a metallic source. The latter species is considered to be the active one during crystal growth. In addition, Heske et al. found an increase of band bending of -150 meV induced by the deposition of Na. This finding, as well as the occurrence of two different Na species, is consistent with results obtained from vacuum-cleaved single crystals (Klein and Jaegermann, 1996). Another interpretation of the beneficial effect of Na is based on the incorporation of Na into the Cu(In,Ga)Se2 lattice (Niles et al., 1997). Niles and co-workers identified Na-Se bonds by means of XPS and concluded that the Na is built into the lattice, replacing In or Ga. The extrinsic defect Nain/Ga should then act as an acceptor and improve the p-type conductivity. The incorporation of Na into the Cu(In,Ga)Se2 lattice is supported by X-ray diffraction measurements that indicate an increased volume of the unit cell (Contreras et al, 1997a). Here, the authors assume that Na in a Cu site prevents the formation of the deep double donor InCu- Schroeder and Rockett (1997) found that Na driven into epitaxial Cu(In,Ga)Se2 films at a temperature of 550 C decreases the degree of compensation by up to a factor 104. Schroeder and Rockett attributed their findings to an Na-enhanced reorganisation of the defects, which allows them to build electrically passive clusters. We see from these numerous approaches that, despite the significance of Na incorporation, the benefit is far from being explained in terms of simple models. However, we feel that in view of the amounts of Na (~0.1 at%) necessary for optimum film preparation, arguments based on its effect on film growth are slightly favoured over those based on the incorporation of Na into the completed film.

296

U. Rau andH. W. Schock

Influence of oxygen Air annealing has been an important process step, crucial for the efficiency especially of the early solar cells based on CuInSe2. Also, though often not mentioned explicitly, an oxygenation step is still used for most of the present-day high-efficiency devices. The beneficial effect of oxygen was explained within the defect chemical model of Cahen and Noufi (1989). In this model, the surface defects at grain boundaries are positively charged Se vacancies V&. (Fig. 7.11a). During air annealing, these sites are passivated by O atoms (Fig. 7.11b). Because of the decreased charge at the grain boundary, the band bending and the recombination probability for photogenerated electrons are reduced. The surface donors and their neutralisation by oxygen are important for the free Cu(In,Ga)Se2 surface as well as for the formation of the surface states

Uc

\

dangling bonds

/ grain boundaries

/

c

\

•

ln;

''

)qVb

^V

Figure 7.11 Band diagram of the conduction and valence band energies across a single grain of Cu(In,Ga)Se2. (a) The electronic states at the grain boundaries are positively charged. This surface charge is compensated by the negative charges in the depleted grain. This induces the band-bending electronic states at the grain boundaries shown at the left-hand grain boundary, and the defect chemical equivalent, dangling bonds, shown at the right-hand boundary; (b) oxygen passivates these dangling bonds and reduces the band bending.

CdS/Cu(In,Ga)Se2 interface (Kronik et al, 2000). Electrical analysis of oxidised and unoxidised samples revealed the validity of the Cahen-Noufi model for the earlier CdS/CuInSe2 devices (Sasala and Sites, 1993), as well as for the more recent ZnO/CdS/Cu(In,Ga)Se2 heterostructures (Rau et al, 1999b; Kronik et al, 1998).

Cu(In,Ga)Se2

297

Solar Cells

(a)

(b)

(c)

Film surface I Grain boundary SCR

Substrata

Substrate

Substrate

Figure 7.12 Illustration of grain boundary and interface charges during the fabrication of a Cu(In,Ga)Se2/CdS heterojunction. (a) On as-prepared films, both grain boundaries and the film surface are positively charged because of the presence of the dangling bonds, (b) Air exposure or air annealing neutralises these charges, (c) The oxide passivation of the surface is removed by the chemical bath deposition process of CdS. The re-established positive charges give rise to a type inversion of the CuInSe2 surface.

The intriguing interplay between surface oxygenation and the deposition of the CdS buffer layer is visualised in Fig. 7.12. In the initial state (Fig. 7.12a), the film surface as well as the grain boundaries are electrically active owing to the positively charged dangling bonds. These charges create a large space-charge region within the grain. Air annealing passivates the dangling bonds at both interfaces. The bands become essentially flat, and space charge essentially vanishes (Fig. 7.12b). Eventually, the chemical bath removes the passivating oxygen and thus re-establishes the beneficial type inversion of the film surface (Fig. 7.12c).

7.3.3 The free Cu(In,Ga)Se2 surface The surface properties of CIGS thin films are especially important, as this surface becomes the active interface of the completed solar cell. However, the band diagram of the ZnO/CdS/Cu(In,Ga)Se2 heterojunction, especially the detailed structure close to the CdS/Cu(In,Ga)Se2 interface, is still under debate. Figure 7.13 depicts three different possibilities corresponding to three different approaches: (a) the ordered defect compound model of Schmid et al. (1993); (b) the surface-state model of Rau et al. (1999a); and (c) the defect layer model of Niemegeers et al. (1998) and Herberholz et al. (1999). The free surfaces of as-grown Cu(In,Ga)Se2 films exhibit two prominent features:

298

U.RauandH. W. Schock

1. The valence band-edge energy Uv lies above the surface-Fermi level £/F by about 1.1 eV for CuInSe2 films (Schmid et al, 1993). This energy is larger than the bandgap energy U"g of the bulk of the absorber material. This was taken as an indication for a widening of band gap at the surface of the film. For example, for the surfaces of Cu(In1_JGax)Se2 thin films it was found that UT-UV = 0.8 eV (almost independent of the Ga content if x > 0) (Schmid et al, 1996b). 2. The surface composition of Cu-poor CuInSe2, as well Cu(In,Ga)Se2 films, corresponds to a surface composition of (Ga + In)/(Ga + In + Cu) of about 0.75 for a range of bulk compositions of 0.5 < (Ga + In)/(Ga + In + Cu) < 0.75. Both observations have led to the assumption that a phase segregation of Cu(In,Ga)3Se5, the socalled Ordered Defect Compound (ODC), occurs at the surface of the films. The segregation of this /3-phase would be compatible with the phase diagram (see Fig. 7.3) and, as this material displays n-type conductivity, could yield the explanation for the surface type inversion. Unfortunately, the existence of a separate phase on top of standard Cu(In,Ga)Se2 thin films has, to our knowledge, not yet been confirmed by structural methods such as X-ray diffraction, high resolution transmission electron microscopy or electron diffraction. Furthermore, if the surface phase exhibited the weak n-type conductivity of bulk Cu(In,Ga)3Se5, simple charge neutrality estimates (Herberholz et al, 1999) show this would not be sufficient to achieve type inversion. The space-charge width in a CIGS absorber of doping density 3 x 1016cm"3 is approximately 300 run. This would require a charge density of 2 x 1018 cm"3 in a 15 nm thick ODC layer to warrant charge neutrality. This required n-type doping density is considerably more than that usually found in Cu(In,Ga)3Se5 compounds. Based on these arguments, another picture of the surface of Cu(In,Ga)Se2 thin films and of junction formation has emerged. As sketched in Fig. 7.13b, the type inversion can be viewed as resulting from the presence of shallow surface donors. This is the classical Bardeen picture (Bardeen, 1947) of Fermi level pinning by electronic states at semiconductor surfaces. Here, a surface-charge density of a few times 1012 cm"2 eV"1 is sufficient to pin the Fermi level at the neutrality level of free semiconductor surfaces. The positively charged surface donors in Fig. 7.13b are expected to be present in the metal-terminated (112) surface of CuInSe2 because of the dangling bond to the missing Se (Cahen and Noufi, 1989). The type inversion vanishes on air exposure because the surface donors are passivated by the reaction of oxygen with the metal-terminated surface, as discussed in Section 7.3.2 in the context of the Cahen-Noufi model.

Cu(In,Ga)Se2

299

Solar Cells

defect layer

a s

Figure 7.13 Models for the CuInSe2 surface: (a) Segregation of a Cu-poor CuInjSes ODC on the surface; (b) band bending due to surface charges; (c) band bending induces Cu-depletion at the surface, creating a surface defect layer. The valence-band energy is lowered due to Cu depletion. The defect layer provides an internal barrier <&i to the electron transport (Niemeegers, 1999).

Surface states play also an important role in the completed heterostructure, where they become interface states at the absorber/buffer interface. As we noted above, one prominent feature of the defect spectrum of the ZnO/CdS/CIGS heterostructure (transition Ni in Fig. 7.5) arises from the charging and discharging of these states. The defect layer model shown in Fig. 7.13c represents in some sense a compromise between the ODC model and the surface defect model, as it takes into account a modification of the band structure due to the Cu deficiency of the surface as well as the presence of positively charged surface states due to the missing surface Se. However, the defect layer model considers the surface layer not as n-type bulk material (as does the ODC model) but as a pMayer (cf. Fig. 7.13c). Furthermore, the defect layer is viewed not as the origin, but rather as the consequence, of the natural surface type inversion. In contrast to the ODC model, surface states are responsible for the band bending. In turn, this band bending leads to the liberation of Cu from its lattice sites and to Cu migration towards the neutral part of the film (Herberholz et al., 1999). The remaining copper vacancies VCu close to the surface result in a high density of acceptor states, i.e., the //-defect layer at the film surface.

300

7.3.4

U.RauandH. W. Schock

Heterojunction formation

The ZnO/CdS/CIGS solar cell structure is usually completed by chemical bath deposition (CBD) of CdS (typically 50 nm) and by the sputter deposition of a nominally undoped (intrinsic) «'-ZnO layer (usually of thickness 50-70 nm) and then by heavily doped ZnO. This three-step process appears at the moment to be mandatory for high efficiency devices, but a convincing explanation of the need for such a relatively complicated three-layer structure, especially the role of the /-ZnO, is not available at the moment. Buffer layer deposition Surface passivation and junction formation is most easily achieved by the CBD deposition of a thin CdS film from a chemical solution containing Cd ions and thiourea. (Kessler et ah, 1992). The benefit of the CdS layer is manifold: 1. CBD deposition of CdS provides complete coverage of the rough polycrystalline absorber surface at a film thickness of only 10 nm (Friedlmeier et al., 1996). 2. The layer provides protection against damage and chemical reactions resulting from the subsequent ZnO deposition process. 3. The chemical bath removes the natural oxide from the film surface (Kessler et al, 1992) and thus re-establishes positively charged surface states and, as a consequence, the natural type inversion at the CdS/Cu(In,Ga)Se2 interface. 4. The Cd ions, reacting first with the absorber surface, remove elemental Se, possibly by the formation of CdSe. 5. The Cd ions also diffuse to a certain extent into the Cu-poor surface layer of the absorber material (Ramanathan et al., 1998b; Wada et al, 1998), where they possibly form Cdc„ donors, thus providing positive charges which support the type inversion of the buffer/absorber interface. 6. As we will discuss below, the open-circuit voltage limitations imposed by interface recombination can be overcome by a low surface recombination velocity in addition to the type inversion of the absorber surface. Thus one might conclude that interface states (except those shallow surface donors responsible for the type inversion) are also passivated by the chemical bath. Owing to the favourable properties of CdS as a heterojunction partner and the chemistry of the CBD process, it is difficult to find a replacement, although avoiding

Cu(In,Ga)Se2 Solar Cells

301

CdS and the chemical bath step would be advantageous from the point of view of production. First, a toxic material such as CdS requires additional safety regulations and, second, the chemical bath deposition does not fit with the vacuum deposition steps of an in-line module fabrication. Research and development in this area relates to two issues: (i) the search for alternative materials for chemical deposition, and (ii) the development of ways to deposit the front electrode without an intermediate step in a chemical bath. Promising materials to replace CdS are In(OH,S), Zn(OH,S) and ZnSe. However, all these materials require additional precautions to be taken for the preparation of the absorber surface or front electrode deposition. Window layer deposition The most commonly used material for the preparation of the front electrode is ZnO doped with B or Al. In some cases, doping with Ga or In is claimed to be advantageous. The first large-area modules produced by ARCO Solar (later Siemens Solar Industries) had a ZnO:B window layer deposited by chemical vapour deposition (CVD). Later production facilities at Boeing and EUROCIS use sputtering processes. Present pilot production lines also favour sputtering. As mentioned above, an undoped /-ZnO layer with a thickness of about 100 ran is needed at the heteroj unction in order to achieve optimum performance. Stability of the ZnO layer in humid environments is a major concern. Indium Tin Oxide (ITO) is superior in this respect.

7.3.5 Module production and commercialisation Monolithic interconnects One inherent advantage of thin-film technology for photovoltaics is the possibility of using monolithic integration for series connection of individual cells within a module. In contrast, bulk Si solar cells must be provided with a front metal grid, and each of these front contacts has to be connected to the back contact of the next cell for series connection. The interconnect scheme, shown in Fig. 7.14, has to ensure that the front ZnO of one cell is connected to the back Mo contact of the next one. Three different patterning steps are necessary to obtain this connection,. The first interrupts the Mo back contact by a series of periodical scribes and thus defines the width of the cells, which is about 0.5-1 cm. For Mo patterning, a laser is normally used. The second patterning step is performed after absorber and buffer deposition, and the final one after window deposition. Scribing of the semiconductor layer is performed by

302

U.RauandH. W. Schock

2nd patterning <

Figure 7.14

• 3rd patterning

Interconnect scheme and patterning of a Cu(In,Ga)Se2 (CIGS) based module.

mechanical scribing or laser scribing. The total width of the interconnect depends not only on the scribing tools, but also on the reproducibility of the scribing lines along the entire module. The typical interconnect width is of the order of 300 fjm. Thus, about 3-5% of the cell area must be sacrificed to the interconnects. Module fabrication Figure 7.15 shows the typical sequence for the production of a Cu(In,Ga)Se2 module. The technologies for absorber, buffer and window deposition used for module production are the same as those discussed above for the production of small laboratory cells. However, the challenge of modules is to transform the laboratoryscale technologies to much larger areas. Basically, the scheme in Fig. 7.15 applies to both of the two concepts currently used for large-area absorber preparation, selenisation and co-evaporation. The selenisation process uses as much off-the-shelf equipment and processing as possible (e.g., sputtering of the metal precursors) for fabricating Cu(In,Ga)Se2 films. For co-evaporation on large areas, the Centre for Solar Energy and Hydrogen Research in Stuttgart (ZSW) has designed its own equipment, shown in Fig. 7.16, for an in-line process. Line-shaped evaporation sources allow continuous deposition of large-area, high-quality Cu(In,Ga)Se2 films. The relatively high substrate temperatures that are necessary for high-quality material impose problems in handling very large area glass sheets. Future process optimisation therefore implies reduction of the substrate temperature.

Cu(In,Ga)Se2

303

Solar Cells

Mo sputtering laser patterning

Figure 7.15

Process sequence for the fabrication process of a Cu(In,Ga)Se2 module.

A bottleneck for the production is the deposition of the buffer layer in a chemical bath. First, it is not straightforward to integrate this process in a line consisting mainly of dry PVD processes. Second, it would be favourable to replace the CdS layer used at multiple sources raw plates

atmosphere

vacuum

deposition

atmosphere

conditioning

Figure 7.16 In-line evaporation system for the deposition of CuInSe2 thin film. The line source evaporates the material from top to bottom. Source: Dimmlcr (1997).

304

U. Rau and H. W.Schock

present. Replacement of CdS by a Zn compound from solution solves part of the problem (Kushyia, 1999). A complete, Cd-free dry process has been demonstrated on a small area, but as yet it does not provide a large enough process window to be used for large-area modules. The transparent ZnO front electrode is put down either by CVD or sputtering. Each method has its specific advantages with respect to process tolerance, throughput, cost and film properties. The widths of the cells within the module—and therefore the relative losses from the patterning—mainly depend on the sheet resistance of the ZnO. Module encapsulation is an important issue because module stability depends on proper protection against humidity. Low-iron cover glasses provide good protection against ambient influences. Hermetic sealing of the edges is mandatory to obtain stable modules. Up-scaling achievements Cu(In,Ga)Se2 has excellent potential to reach more than 15% module efficiency in the near future. Mini-modules ranging in area from 20 to 90 cm that use the process sequence anticipated for use in a commercial module have already reached efficiencies around 14-15%. Recently, Siemens Solar Industries fabricated a 1 x 4 ft2 power module (-44 W) with an NREL-verified efficiency of 12.1%. Using a totally different approach to the deposition of the absorber layer, ZSW fabricated a 30 x 30 cm2 module with a verified efficiency of 12.7%. Table 7.2

Performance of Cu(In,Ga)Se2 cells and modules

Process Co-evaporation

Lab. cell Module eff. (%) eff. (%) 18.8 16.1

11.5 Selenisation

>16

13.9 12.7 9.6 5.6 14.7 12.1 14.7 14.2 11.6

Area (cm2) 90 800 135 240 19 3600 18 50 864

Laboratory/company NREL, USA (Contreras et al, 1999) ZSW/IPE, Germany ZSW, Germany Energy Photovoltaics, USA Global Solar (flexible cells), USA Angstrom Solar Centre, Sweden Siemens, USA and Germany Showa, Japan

Cu(In,Ga)Se2 Solar Cells

305

Table 7.2 shows more results of the different processes. For the future, module sizes up to 60 x 120 cm2 are planned in order to meet cost goals. Because of the promising results from the laboratory scale and the first approaches to up-scaling, several companies other than those mentioned in Table 7.2 now plan commercial production. Stability and radiation hardness Stability appears to be no problem for CuInSe2 modules during long-term outdoor testing and tests at elevated temperatures. The data in Fig. 7.17 obtained by the ZSW prove more than 1000 hours stability in hot (85 C) and humid (85% humidity) atmosphere. Cu(In,Ga)Se2 modules fabricated by Siemens Solar Industries have shown very good stability during more than 10 years outdoor operation (Gay, 1997). This, as well as their operation in space, has proved that there is no intrinsic

1.00

>

c g> 'o

n„«n.3%

I |

85 C, 85 % relative humidity according to IEC1646

0.90

b

0

400

800

1200

1600

Damp heat time/hours

Figure 7.17 Stability of two 30 x 30 cm2 CIS modules under damp heat test. After 1000 hours, the efficiency is still over 95% of the initial value. Data from M. Powalla, ZSW (1999).

mechanism that affects cell performance. On the contrary, cells often improve during operation. The stability inherent to the material system has recently been investigated (Rau et al., 1998a; Guillemoles, 2000). A self-healing mechanism due to defect relaxation with the help of mobile copper makes this material unique. Hence an important prospective application for Cu(In,Ga)Se2 cells is in space, where the main power source is photovoltaics. Space satellites in low-Earth orbits for communication systems require solar cells with high end-of-life efficiencies. Cu(In,Ga)Se2 has proven superior radiation hardness, which could make this type of cell the material of the future for space applications (Schock and Bogus, 1998; Jasenek et al, 2000a). The

306

U.RauandH. W. Schock

challenge for developing CIGS space cells is to reduce the weight by depositing the cells on foil substrates, and at the same time to retain the performance achieved with devices on soda-lime glass.

7.4

Device physics

7.4.1 Band diagram The device physics underlying electronic transport in thin-film solar cells is obviously the same as for (say) bulk silicon solar cells, and this is covered by Schumacher and Wettling in Chapter 2. However, in view of the fact that thin-film Cu(In,Ga)Se2 (and also CdTe cells) are heteroj unction cells, and because the absorber layer is only few times thicker than the space-charge region of the heteroj unction, we shall develop a description which concentrates more on these specific features. The equilibrium band diagram of the ZnO/CdS/Cu(In,Ga)Se2/Mo heterostructure in Fig. 7.18 shows the conduction and valence band energies I/C>v of the Cu(In,Ga)Se2 absorber, the CdS buffer layer and the ZnO window. The latter consists of the intrinsic and the highly Al-doped layer. Here, we completely neglect the polycrystalline nature of the semiconductor materials, which in principle requires a two- or threedimensional band diagram. We will restrict ourselves in the following to the implication of the one-dimensional diagram of Fig. 7.18. Even in the one-dimensional model, some details of the band diagram are still not perfectly clear. The diagram in Fig. 7.18 concentrates on the heteroj unction and does not show the contact between the Mo and Cu(In,Ga)Se2 at the back side of the absorber. Another feature under debate but neglected here is a 10-30 nm thick defect layer on top of the Cu(In,Ga)Se2 absorber, already discussed in Section 7.3.3. The energetic quantities describing the band diagram in Fig. 7.18 are the band gap energies £/*, where x = a,b,w for the absorber, buffer and window, respectively. The conduction/valence band offsets between the semiconductors are denoted MJf/v. The built-in or diffusion voltage of the p-type absorber is Vg whereas that of the n-type window/buffer is the sum of the contributions V£b from the buffer V£b and V^, from the window layer. Note that the quantities Vg/n as drawn in Fig. 7.18 are zero-bias quantities and change when an external voltage is applied. The important barriers £ and ®n„ can be calculated from 4>£ = qVb" + qp and <&nb = AL/f + AUFn = At/"* + AUbl+AUb2, where £p =UF-UV refers to the neutral bulk of the absorber and AfFn =UV -UF to the Cu(In,Ga)Se2/CdS interface. We will discuss the impact of $£ on interface recombination and that of <&"b on the fill factor in Section 7.4.3. A

Cu(In,Ga)Se2

Solar Cells

/-ZnO CdS

Cu(ln,Ga)Se2

—3E_._i

uc

II

Figure 7.18 Band diagram of the CIGS heterojunction showing the conduction and valence band-edge energies Uc and Uv. The quantities 0 £ and <&l are the barriers for holes and electrons, AUF„ is the energy distance between the Fermi level and the conduction band energy at the CdS/CuIn(Ga)Se2 heterointerface, AUJ* and AC,"* are the valence-band and conduction-band discontinuities at the buffer/absorber interface, and wp and w„ are the widths of the space-charge regions in the p-type absorber and the n-type window/buffer respectively.

simplified approach to describing the heterojunction and computing £ and <&"„ is given in Rau et al. (1999a). The most important quantities to be considered in the band diagram are the band discontinuities between the different heterojunction partners. Band discontinuities in terms of valence-band offsets LUyab between semiconductor a and b are usually determined by photoelectron spectroscopy (for a discussion with respect to Cu-chalcopyrite surfaces and interfaces, see Scheer, 1997). The valence-band offset between a (01 l)-oriented Cu(In,Ga)Se2 single crystal and CdS deposited by PVD at room temperature is determined as MJ* = -0.8 ( ± 0.2) eV (Nelson et al, 1993; Loher et al, 1995), and therefore MJ? =U™-U^+MJ? =0.55eV, with the band gaps U™ = 2.4 eV and U™ ~ 1.05 eV of CdS and CuInSe2, respectively. Several authors have investigated the valence-band discontinuity between polycrystalline Cu(In,Ga)Se2 films and CdS, and found values between 0.6 and 1.3 eV with a clear centre of mass around 0.9 eV, corresponding to a conduction-band offset of 0.45 eV

308

U. Rau andH. W. Schock

(Scheer, 1997). Wei and Zunger (1993) calculated a theoretical value of At/f = 1.03eV, which would lead to A t / ? = 0 . 3 e V . The band alignment of polycrystalline CuInSe2 and Cu(In,Ga)Se2 alloys was examined by Schmid et al. (1993, 1996a), who found that the valence-band offset is almost independent of the Ga content. In turn, the increase of the absorber band gap leads to a change of At/"4 from positive to negative values. The conduction-band offset between the CdS buffer and the ZnO window layer was determined by Ruckh et al. (1994b) to be 0.4 eV.

7.4.2

Short-circuit current

Optical losses The short current density ix that can be obtained from the standard lOOmW cm"2 solar spectrum (AM 1.5) is determined, on the one hand, by optical losses, that is, by the fact that photons from a part of the spectrum are either not absorbed in the solar cell or are absorbed without generation of electron-hole pairs. On the other hand, not all photogenerated electron-hole pairs contribute to isc because they recombine before they are collected. We denote these as recombination losses. Figure 7.19 shows the maximum ix that can be obtained from the standard AM 1.5G spectrum (Green, 1995, Appendix C) for photon energies hv> Ug hv>Ug , i.e., the photons that can contribute to the short-circuit current of a semiconductor of band-gap energy Ug. For pure polycrystalline CuInSe2 with Ug = 1.04 eV, this value is 46.8 mA cm"2. For Ug= 1.11 eV, the band-gap energy of the best Cu(In,Ga)Se2 solar cell, J'SC = 43.6 mA cm"2. Now we estimate how much absorber material is needed to achieve this photocurrent. The light absorption in a semiconductor is described by the Lambert-Beer law. The irradiance E decays exponentially with depth x into the semiconductor according to E(x) = E0 exp(-ax)

(7.1)

where E0 is the incident irradiance and a the absorption coefficient. For direct semiconductors, a depends on the photon energy hv according to

a{hv) = a±

£_ hv

(7.2)

Cu(ln,Ga)Se2 Solar Cells

309

50 40 30

1 ^8

20 10 0t

J

•

1

1

1

1

20

•

1

1

L

L—l

I

I

l__J

•

•

2.5

Band-gap energy q/eV

Figure 7.19 Short-circuit current density from the AM 1.5 solar spectrum and corresponding to the band -gap energies of various chalcopyrite compounds, of a typical Cu(In,Ga)Se2 alloy (Ug = 1.12 eV), and of the heterojunction partners. The inset shows the losses that occur if less than 1.25/0.5 [im material is available for light absorption.

The absorption coefficient of Cu(In,Ga)Se2 with a low Ga content is reasonably described by eq. 7.2 and a ~ 8 x 104 eV1/2cm_1. By reorganising eq. 7.2 we can calculate the excess energy Mv = /iv-{/„s =1 — a ' a

(7.3)

of photons that have an absorption coefficient larger than a given a. For instance, with absorption lengths La = a~l = 1.255 /jm and 0.5 /zm, we have A/iv = 10 and 62.5 meV. The inset in Fig. 7.19 shows that the losses A/sc corresponding to the photons that are not absorbed within 1.25 or 0.5 jjm of CuInSe2 are 0.9 mAcm - 2 and 3.0 mAcm , respectively. The lower-energy photons are either absorbed at the backmetal/absorber interface or reflected out of the cell. Thus a typical absorber of thickness 1.5 /an absorbs all the light from the solar spectrum except for a negligible remnant corresponding to less than 1 mA cm - .

310

U. Rau and H. W. Schock

Next, we have to recognise that light absorbed in the ZnO window layer does not contribute to the photocurrent. This loss affects absorption and photogeneration for photons of energy > 3.2 eV (the band-gap energy of ZnO). As shown in Fig. 7.19, this loss of high-energy photons costs about 1.3 mA cm"2. In addition, photons in the energy range hv <1.4 eV are absorbed by free carriers within the conduction band of the window material. High-efficiency heterojunction solar cells avoid free carrier absorption by optimising the conductivity of the front electrode, not by a high concentration, but by the high mobility of the free carriers. In any case, for wide-gap absorbers with Ug > 1.4 eV this loss can be neglected. Another portion of the solar light is absorbed in the buffer layer. If, for instance, a CdS buffer layer caused a sharp cut-off of the spectral response at the band gap of 2.4 eV, only a total of 38 mA cm-2 or 35.5 mA cm""2 would be available for the shortcircuit current of a CuInSe2 or Cu(In,Ga)Se2 (Ug = 1.11 eV) solar cell, respectively. However, measurements of the External Quantum Efficiency (EQE) of a typical ZnO/ CdS/Cu(In,Ga)Se2 heterostructure reveal that the EQE typically drops by a factor of only -0.8 in the wavelength range between the band gap of CdS and that of the ZnO window layer. About 70-80% of the photons in the wavelength range 440-510 nm contribute to i^ because the thin buffer layer does not absorb all photons and about 50% of the electron-hole pairs created in the buffer layer still contribute to the photocurrent (the hole recombination probability at the buffer/absorber interface is relatively low (Engelhardt et al, 1999). Collection loss analysis The most common characterisation method of solar cells other than current-voltage analysis is the measurement of the quantum efficiency. The EQE at a given wavelength X is defined as the number of electron-hole pairs contributing to the photocurrent divided by the number of photons incident on the cell. A quantitative evaluation of the EQE can be used to determine the diffusion length Le if the data are corrected for reflection losses and absorption losses in the window material and if the absorption data of the absorber material are known for the wavelength regime where the absorption length is in the order of Le (Arora et al., 1980). This analysis has been performed in the past by several authors for different types of devices (Klenk and Schock, 1994; Parisi et al, 1998). An alternative way to determine the diffusion length in solar cells is provided by Electron Beam Induced Current (EBIC) measurements. Two approaches are possible: planar EBIC, where the electron beam is scanned over the device surface, and junction EBIC, where the device is cleaved and the beam is scanned along the cross section

Cu(In,Ga)Se2 Solar Cells

311

(Jager-Waldau et al., 1991). For CIGS, the values for Le extracted from EQE and EBIC measurements are -0.5-1.5 /J.m.

7.4.3

Open-circuit voltage

Diode characteristics At open circuit, no current flows across the device and all photogenerated charge carriers have to recombine within the solar cell. The possible recombination paths for the photogenerated charge carriers in the Cu(In,Ga)Se2 absorber are indicated in the band diagram of Fig. 7.20. Here we have considered recombination in the neutral bulk (A) and at the back surface of the absorber (A'), recombination in the space-charge region (B), and recombination at the buffer/absorber interface (C). The dotted lines indicate that the latter two mechanisms may be enhanced by tunnelling in the presence of a high built-in electrical field. Cu(ln,Ga)Sft.

Figure 7.20 Recombination paths in a CdS-Cu(In,Ga)Se2 junction. The paths A and A' represent bulk and back-contact recombination. B and C result from space-charge and interface recombination. The dotted arrows indicate tunnelling.

At the back contact we have drawn the thin MoSe2 layer which forms during the first minutes of absorber deposition. As drawn here, the MoSej has a small conduction-band offset with respect to the Cu(In,Ga)Se2 bulk material and a small Schottky barrier at the Mo back contact. Both features are beneficial for device performance, because the conduction band offset between the Cu(In,Ga)Se2 absorber and the MoSe2 acts as an electronic mirror (the so-called back surface field) for the

U.RauandH. W. Schock

312

photogenerated electrons and diminishes back-surface recombination, and the narrow Schottky barrier provides no substantial resistance for holes between the absorber and the metallic back contact. We emphasise, however, that the details of this band diagram are still under debate. The basic equations for the recombination processes (A-C) can be found in Bube (1992) or Chapters 1 and 2 of this book. All recombination current densities irec for processes A-C can be written in the form of a diode law

W =UexP

PkT t

(7.4)

-1

where V is the applied voltage, /3 the diode quality factor, and kTlq the thermal voltage. The saturation current density i0 is in general a thermally activated quantity and may be written in the form '«. ^ h = 'oo exp kT

(7.5)

where ua is the activation energy and the prefactor j,*, is only weakly temperaturedependent. The quantities «0and /3 depend on the details of each recombination mechanism. Since mechanisms A-C are connected in parallel, the strongest one will dominate the recombination loss. At open circuit, the total recombination current density irec exactly compensates the short-circuit current density iK. Hence we can write the open-circuit voltage in the form

v^SL-BE.^

'L^

(7.6)

where we have assumed that V^ >3/3kT/q, which allows us to consider only the exponential term in eq. 7.4. We have also replaced the activation energy ua by Ua =fiua, which will prove in the following to be the 'true' activation energy of the carrier recombination processes. We shall now discuss the recombination processes A-C in more detail.

313

Cu(In,Ga)Se2 Solar Cells Recombination in the absorber

In the following we shall assume a n -p junction, i.e. that the doping density on the nside is much higher than on the p-side. Shockley's diode equation for such a singlesided junction yields the saturation current density for recombination in the neutral region of the (p-type) absorber. Knowing the square of the intrinsic carrier density n] = NCNV exp(-Ug IkT) we calculate the open-circuit voltage as

U* q

kTjqDeNcN^ q

LN*Le

(7.7)

where De is the diffusion constant for electrons, and Nclv the effective density of states in the conductance/valence band. (For our calculations we have used the values Nc = 6.7 x 1017cm~3 and Nv=l.5x 1019cm"3 resulting from the density-of-states effective masses me =0.09m 0 and mh = 0.71 m0 for electrons and holes, respectively, where m0 is the free electron mass (Neumann, 1986). The quantity NA is the acceptor density, and Le is the diffusion length of the electrons. If this becomes comparable with the thickness d of the quasi-neutral region (QNR) of the absorber, the recombination velocity Sb at the back contact has to be taken into account (recombination path A' in Fig. 7.20), and Lt in eq. 7.7 has to be replaced by cosh sh cosh (z-1) + s i n h ( r l ) where Sb = Sb Le IDe and / = Le Id. Since the width of the space-charge region in thin-film solar cells is comparable with the film thickness, recombination in the space-charge region is important. The Vx -limitation due to recombination in the space-charge region (SCR) of the absorber may be written in a form comparable to eq. 7.7, namely

V.-2L-=II. q

q

kTDenl2^NcNv i E I2

(7.9)

where £m = (2qNAVbm /es) is the electrical field at the position of maximum recombination. The quantity £m depends on the doping density NA, the band bending Vbm, and the dielectric constant es of the absorber. The dependence of eqs. 7.7 and 7.9 on

314

U. Ran and H. W. Schock

the doping density NA is equal in that an increase of NA by one order of magnitude yields an increase of Vx of &VX = (kT In 10) / q ~ 60 raV . However, improving the open-circuit voltage by increasing the doping density is limited by the increased Auger recombination in the QNR and the enhancement of tunnelling in the SCR (Green, 1996a). In eq. 7.9, the activation energy U„ is given by U„ = Aua = 2ua = Ug, whereas in the diode equation for space-charge recombination, the saturation current density is io °= exp (Ug/2kT) and the activation energy is only Ug/2. This demonstrates that we have to correct the activation energies obtained from, for example, Arrhenius plots of the temperature dependence of /„ for the effect of non-ideal diode behaviour in order to obtain the activation energy relevant to VK. Diffusion length L„/ fim

1

05

1

5

' '

' >^1

Sb=10'cmj^£^

.,.'. ...

0.40

0.70 s8

1 0.65

s£^°^S*=\tfcm

y

i 1

Ope

-Cf I

/ /

i

0.60

- 0.45

/

1 i

!

sH

0.50

8

• K.

i:

//

• 0.55

0.55

.... 0.1

.

: i

10

100

0.60

Lifetime t„/ns

Figure 7.21 Correlation of the open-circuit voltage with lifetimes and diffusion lengths for a device with a band gap of 1.12 eV. Solid lines are the results of eq. 7.9 for £,.<0.7/tm. For Le>Q.l fim, eq. 7.7 holds if effective diffusion lengths that take back-surface recombination into account are introduced. The lines with symbols arc the results of a complete device simulation.

In Fig. 7.21 we display the open-circuit voltage limitations given by eqs. 7.7 and 7.9 for a Cu(In,Ga)Se2 solar cell with an absorber layer thickness of 1.5 /jm, a bandgap energy Ug of 1.11 eV and a short-circuit current density /„. of 35.4 mA cm - . The top and the bottom axes, showing the electron diffusion length Le and the lifetime re, are connected by Lr =(z>,T,)"2and a diffusion constant which is here assumed to be De = 2.59 cm2s"'. As the open-circuit voltages in eqs. 7.7 and 7.9 can be shifted by the band-gap energy, we have used the right-hand axis of Fig. 7.21 to display the difference (Ug/q) - V^.

Cu(In,Ga)Se2 Solar Cells

315

For the record Cu(In,Ga)Se2 solar cell (Contreras et al., 1999), this difference is only (1.12- 0.68) V = 0.44 V. The open-circuit voltage of this device requires a lifetime of 30 ns or more, corresponding to a diffusion length of over 2 /an, thus exceeding the absorber thickness. Hence recombination at the back contact also has some influence on V,*—a recombination velocity Sb >105 cm s"1 would hardly allow a VocOfeSOmV. Since the open-circuit voltage of reasonable Cu(In,Ga)Se2 devices (VQC ~ 0.5 V) is just at the threshold between SCR and QNR recombination, we have also conducted some numerical simulations using the software package SCAPS-1D (Niemeegers and Burgelman, 1996). The results for assumed back-surface recombination velocities 5„= 102 cms"1 and 105 cms"1 are also displayed in Fig. 7.21. Here we see that recombination can be well described only outside a transition regime of 1 ns < T„ < 30 ns (0.5 /an < Le < 3 fJm) by the analytical approaches for SCR or QNR recombination. Within this parameter range, the recombination paths A, A' and B contribute to recombination. Note that we have suppressed interface recombination (path C) by setting the recombination velocity for holes at the front contact to Sp = 102 cm s"1 and assuming a hole barrier <X>£ = 1 eV. Effective lifetimes for polycrystalline semiconductors Cu(In,Ga)Se2 solar cells are based on polycrystalline absorbers. Electronic transport in such devices is not completely covered by one-dimensional models. However, quasione-dimensional approaches are possible as long as the influence of grain boundaries on the recombination and charge distribution is not too strong. A first-order approximation is the replacement of the minority-carrier lifetime te by an effective lifetime T^ y , which includes the interface recombination velocity Sg at the grain boundaries. This is given by 1 Tpoly

- U ^ T*

e

(7.10)

where xbe is the minority-carrier lifetime within the grain volume and g denotes the grain size. With the help of eq. 7.10, we can still use eqs. 7.7 and 7.9 if we also use the effective diffusion length Z^Jy for polycrystalline materials, given by .poly

_

'(C°)- 2 + 2S g /(Ds)J" 2

(7.11)

316

U. Rau and H. W. Schock

instead of Leff = L™°"°. For more details, and the limitations of eqs. 7.10 and 7.11, see Green (1996b), Brendel and Rau (1999) and Jensen et al. (2000). Distribution of recombination centres An approach to describing the temperature dependence of current-voltage curves which is useful for Cu(In,Ga)Se2 devices was introduced in Walter et al. (1996b). This approach does not use recombination centres of a single energy within the forbidden gap, but rather a distribution of the form DT(U) = DTOexp(-U I kT*), where the centres are exponentially distributed in energy. The defect density DT(U) has units of cm"3 eV~' and kT* denotes the characteristic energy of the exponential distribution. The characteristic energy U* = kT* is also seen in the defect density spectra obtained from admittance spectroscopy (see Fig. 7.5). A rigorous mathematical treatment for the recombination current density under this assumption is given in Rau et al. (2000). The recombination current density can be written in the form

,exp

(fikTJ exp

fikT

(7.12)

where the pre-exponential term i^ is weakly temperature dependent, and the diode quality factor is given by f

T \

1+ T

(7.13)

The importance of this approach is on the one hand that a defect distribution with a characteristic energy kT* of the order of 100-150 mV is often observed in Cu(In,Ga)Se2 as well as in CuGaSe2. On the other hand, it has been shown by Walter et al. (1996b) and Engelhardt et al. (1998) that the temperature dependence of the current-voltage characteristics of high efficiency Cu(In,Ga)Se2 solar cells in the temperature range 2O0K
Cu(In,Ga)Se2 Solar Cells

317

Interface recombination Here we consider the simple case of an inverted interface, as shown in Fig. 7.13. If the recombination centres are not too close to the conduction band, the recombination rate is dominated by the concentration p \if of free holes at the interface. From the band diagram and taking the interface recombination velocity for holes as Sp and the diffusion potential at the p-side of the heteroj unction as Vg , the recombination current density is given by

( L

= qS N e x p

=
kT

(7.14) kT

with t;=UF-Uv in the QNR of the Cu(In,Ga)Se2 absorber. In general, a voltage applied over the heteroj unction does not drop only across the p-type part of the junction. Rather, a change AK in the externally applied bias is shared between the ptype and n-type part according to &V = (l-a)AV£ +CCAVQ where AVg/AVp is the ratio of the diffusion potential in the /?-type and n-type components. Note that the calculation of the voltage share between the two heteroj unction partners in a complicated heterojunction like that shown in Fig. 7.18 is not straightforward. Here we simply use a linear approach with the coefficient a (0 < a < 1). The diode law can be written equivalently by the use of a voltage-dependent barrier <5>pb (V) = £0 +aV , where
L = 1sPNv e*P

(

fqV kT

exp

|

kT = *S,tf,exp

<J>"

- exp kT

(

qV{\-aj

(7.15)

kT

By comparison of the coefficients we find that the coefficient a is linked to the diode quality factor by f3=(l—a)~l. Finally, we write the open-circuit voltage for interface recombination as

Of

fikT1

•In

(qSpNv

(7.16)

•/sc

where -
318

U.RauandH. W. Schock

CuInSe2 absorber were approximately equal, so the band bending in the two heteroj unction partners was also equal. In terms of eq. 7.16 and the band diagram, the open-circuit voltage was determined by a relatively low value of <&£ and a high value of Sp. The use of a highly doped ZnO layer and a CdS buffer layer of only several tens of nm in thickness in the actual high-efficiency ZnO/CdS/Cu(In,Ga)Se2 devices increases the band bending in the absorber and decreases the recombination velocity for holes at the heterointerface. For these recent state-of-the-art devices, it is believed that recombination in the bulk of the absorber is the dominant loss mechanism. Tunnelling In the presence of high electrical fields, interface recombination as well as spacecharge recombination may be enhanced by tunnelling. As shown in Fig. 7.20, the charge carriers do not have to overcome the entire energetic barrier, but only a part of it in order to recombine. We denote this transport path as tunnelling-enhanced recombination because it can be described in very similar terms to the classical recombination mechanism merely by using modified recombination rates. A useful quantity within the theory of thermally-assisted tunnelling is the tunnelling energy

2

\mes

The tunnelling energy Uoo (commonly encountered in the literature as Em (Padovani and Stratton, 1966)) depends only on the material parameters NA, e, and the effective tunnelling mass m . First we consider the case of tunnelling-enhanced bulk recombination in the SCR. Here the modified recombination rate for tunnelling-assisted recombination can be integrated over the width of the space-charge region. For an exponential distribution of trap states, a convenient form of the diode law gives the diode quality factor as (Rau, 1999)

J__ J_ P~2

1 + —r-

Un

(7.18)

3(My

The recombination current density is again expressed by eq. 7.12, but with the quality factor given by eq. 7.18.

319

Cu(In, Ga)Se2 Solar Cells

The case of tunnelling-enhanced interface recombination can be treated by analogy with the thermionic field emission theory of Schottky contacts (Padovani and Stratton, 1966). The recombination current density is c

.,

ylnqVb(x)Um kTcoshiUgg/kT)

PkT exp

exp

1 kT

-1

(7.19)

and the quality factor u

P = ^ c o t h oo kT kT

(7.20)

is given in Padovani and Stratton (1966). Tunnelling-enhanced interface recombination is often invoked in interpretations of current-voltage curves of wide-band-gap materials such as CuInS2 (Hengel et al., 2000). Recombination loss analysis The analysis of the temperature dependence of current-voltage curves is one way to get closer access to the recombination mechanism relevant to the open-circuit voltage. Usually, current-voltage curves are recorded with different illumination intensities 1.2

j

1.0

0.8

0.6

s 0.4 100 200 Temperature 77K

300

Figure 7.22 The temperature dependence of the open-circuit voltage under various illumination intensities extrapolates at 0 K to the band-gap energy.

320

U. Rau andH. W. Schock

and at different temperatures. One evaluation method makes use of the temperature dependence of the open-circuit voltage. Following eq. 7.6, the open-circuit voltages obtained at different temperatures T but identical illumination intensities (corresponding to roughly the same ix) display a linear temperature dependence. Figure 7.22 shows that the V^T) curves in the temperature range 200-350 K extrapolate to an activation energy of 1.24 eV, indicating that the band-gap energy of the Cu(In,Ga)Se2 absorber material is the relevant activation energy for recombination. Such an evaluation works well as long as the diode quality factor in eq. 7.6 is not too temperature-dependent. An evaluation scheme that is able to invoke also temperature-dependent quality factors makes use of the fact that the saturation current density of all recombination processes discussed above can be written in the form


—

{kT

'ooeXP

{/3kT

(7.21)

Reorganisation of eq. 7.21 yields

/?ln

(i

\

r»)

(7.22)

kT

where Ua = <J>£ for interface and Ua = Ug for bulk recombination. Thus a modified Arrhenius plot of P In j 0 vs. 1/7 should yield a straight line with the relevant activation energy Ua as the slope. Figure 7.23a compares such modified Arrhenius plots of the saturation current densities i0 derived from three different Cu(In,Ga)Se2 based heterojunctions. The three curves stem from absorbers with different Ga contents X = 0, 0.28, and 1. The activation energies Ua =1.04 eV, 1.21 eV and 1.67 eV extracted from the modified Arrhenius plots correspond to the band-gap energies of the respective absorber materials. This indicates that recombination in all these devices takes place in the volume of the absorber. The corrected Arrhenius plots yield the activation energy of the recombination process whether or not the transport processes are enhanced by tunnelling. More specific information on the transport mechanism is obtained from the temperature dependence of the quality factor /?. We evaluate the quality factors with the help of eq. 7.18, as shown in Fig. 7.23b. From the fits we find for the CuInSe2 (x = 0) cell Um = 3.9 meV and kT* = 95 meV, for the Cu(In,Ga)Se2 cell U^ = 6.5 meV and kT' = 74 meV, and for CuGaSe2 Uw = 16 meV and kT' = 281 meV. Thus an increasing Ga

Cu(ln,Ga)Se2 Solar Cells

321

content in Cu(In,Ga)Se2, in general, increases the contribution of tunnelling as expressed by the increase of the characteristic tunnelling energy U^ in Fig. 7.23b. Temperature T(K) 300

200

100

—I—

0.004

0.006

0.008

0.01

Inverse temperature 1/7"(1/K)

0.7 -

CulnSe2

100

150

200

250

300

Temperature T(K) Figure 7.23 (a) Arrhenius plots of the reverse saturation current multiplied by the diode quality factor. Straight lines indicating an activation energy equal to the respective band-gap energy are evidence for bulk recombination; (b) The experimental data of the inverse quality factor vs. temperature can be fitted to the model that includes tunnelling.

However, for CuInSe2 and Cu(In,Ga)Se2 with moderate Ga content at room temperature the tunnelling contribution to recombination is insignificant. In contrast, for CuGaSe2 one extracts significant values of U^ (Nadenau et al., 2000).

322

7.4.4

U. Rau and H. W. Schock

Fillfactor

The fill factor FF of a solar cell can be expressed in a simple way as long as the solar cell is well described by a diode law. Green (1986, p. 96) gives the following phenomenological expression for the fill factor

FF0 =

l-ln(v ^

-0.7) '-

(7.23)

where v,,,. is the dimensionless quantity v,* = qVx ipicT and the fill factor FF0 results from the diode law alone. Thus FFQ depends on temperature as well as on the quality factor of the diode. Effects from series resistance Rs and shunt resistance Rsb also contribute to the fill factor losses. A good approximation is given by (

FF = FF

x

v^+0.7 F p V oc

/V sh

(7.24)

)

where FFX = FF0(1 - rs), rs=Rsix/Vx and rsh = RJx IVx. The description of Cu(In,Ga)Se2 solar cells in terms of eqs. 7.23 and 7.24 works reasonably well. For example, the world record cell has a fill factor of 78.6% and the value calculated from eqs. 7.23 and 7.24 is 78.0% (Voc = 678 mV, ^ = 3 5 . 2 mAcirf2, /?s = 0.2 Qcm 2 , Rsb = 104 Q cm2, j0=1.5: Contreras et a/., 1999). The dependence of the fill factor on the quality factor (bearing in mind that this also determines the open-circuit voltage) highlights the importance of this parameter for the output power of the solar cell. The diode quality factor /? L obtained from the illuminated current-voltage curve is often different from /? D of the dark curve, with /3L > /? D . One explanation for this important fact could be the finite barrier <&nb for the electron transport across the CdS/Cu(In,Ga)Se2 heterointerface. Looking at Fig. 7.18, a band offset AUcab ~ 0.3 eV between the absorber and the buffer layer seems to represent a substantial barrier hindering the collection of photogenerated electrons. The buffer/absorber interface shown in Fig. 7.6 can be roughly looked at as a Schottky barrier with a height nb = AUcttb + At/F„ and the photocurrent as the reverse current over this barrier. A simple calculation shows that the barrier does not provide a large series resistance to the cell as long as OJJ < 0.5 eV . The conductance of a Schottky contact with an effective Richardson constant A*=100 A cm-2 K"2 is G = (qA*T/k) exp [-®nb IkT] = 1.4 S cm-2 for
Cu(In,Ga)Se2

323

Solar Cells

with a reasonable open-circuit voltage. It has been shown by numerical calculations that a conduction band offset All"' below 0.4 eV does not affect the device performance (Niemegeers et al., 1995). However, at lower temperatures or for devices where the type inversion at the Cu(In,Ga)Se2 is less pronounced, i.e., the quantity A£/Fn is larger than 200 meV, the barrier $>"b becomes important as shown by experiments (Schmidt, 2000, p. 682) and numerical simulations (Topic et al., 1997).

7.4.5

Electronic metastabilities

The long time relaxation (over hours and days) of the open-circuit voltage of Cu(In,Ga)Se2 based solar cells during illumination is a commonly observed phenomenon (Ruberto and Rothwarf, 1987; Sasala and Sites, 1993). Fortunately, it turns out that in most cases the open-circuit voltage increases with illumination time, a situation which is more favourable than that encountered in a-Si:H (Staebler and Wronski, 1977). A first model for the open-circuit voltage relaxation of Cu(In,Ga)Se2 solar cells was proposed in Ruberto and Rothwarf (1987). This model relies on the reduction of interface recombination at the CdS/Cu(In,Ga)Se2 interface by additional charges introduced into the CdS buffer layer either by illumination under open-circuit conditions or by application of forward bias in the dark. The model is based on the assumption that interface recombination is the dominant recombination mechanism in the solar cells. The increase of positive charges in the buffer layer is assumed to increase the barrier £ and thus reduce interface recombination. However, as we noted above, the open-

fa)

(b)

Figure 7.24 Illustration of persistent changes of (a) the density of free charge carriers in the bulk, and (b) the charge density in the space-charge region.

324

U.RauandH. W. Schock

circuit voltages of the recent high-efficiency devices are limited by recombination in the bulk {i.e., in the SCR) rather than at the interface. Since these devices also show light-soaking effects, another mechanism, possibly additional to that proposed in Ruberto and Rothwarf (1987), must be at work. An important observation is that of persistent photoconductivity in Cu(In,Ga)Se2 thin films (Rau et al., 1998c) and single crystals (Seifert et al., 1997). Meyer et al. (1999) relate the persistent trapping of electrons as the origin of persistent photoconductivity (Fig. 7.24a) to the persistent increase of the charge density in the SCR of the heterojunction, as shown in Fig. 7.24b. This leads to another model for the open circuit voltage relaxation in Cu(In,Ga)Se2 solar cells: the gradual decrease of the electrical field in the SCR leads to a decrease of space-charge recombination, and finally to the increase of the open-circuit voltage during illumination. The band diagrams in Fig. 7.25 schematically compare the model of Ruberto and Rothwarf

(a)

(b)

Figure 7.25 Metastability effects in CIGS-based heteroj unctions, (a) Light-generated excess positive charges are persistently captured in the buffer layer and lead to an increase of the barrier &pb . The full (dashed) lines correspond to the band diagram before (after) illumination; (b) light-generated excess negative charges persistently trapped in the Cu(In,Ga)Se2 absorber layer lead to a decrease of the width of the p-side part of the space-charge region.

Cu(In,Ga)Se2 Solar Cells

325

(1987) with the more recent suggestion of the consequence of persistent photoconductivity in bulk Cu(In,Ga)Se2 (Meyer et al., 1999). In Fig. 7.25, the solid and dashed lines represent the band diagram before and after illumination, respectively. As shown in Fig. 7.25a, an increase of positive charge in the buffer layer increases the barrier
7.5

Wide-gap chalcopyrites

The system Cu(In,Ga)(S,Se)2 provides the possibility of fabricating alloys with a wide range of band-gap energies between the 1.04 eV of pure CuInSe2 up to 2.4 eV for CuGaS2. The achievement of higher voltages in the individual cells by increasing the band gap of the absorber material is important for thin-film modules. An ideal range for the open-circuit voltage would be between 1.4 and 1.6 eV. This increased voltage would reduce the number of scribes needed for monolithic integration of the cells into a module. The thickness of front and back electrodes could also be reduced because of the reduced current density. Higher band-gap materials also lower the temperature coefficient dPnm IdT of the maximum output power P^^ , i.e., the loss of conversion efficiency with increasing cell temperature. Higher band-gap materials thus also perform better at elevated temperatures. However, the highest efficiency within the chalcopyrite system is achieved with the relatively low band-gap energy Ug of 1.12 eV, and attempts to maintain the high efficiency level at Ug> 1.3 eV have so far failed. Table 7.3 compares the output parameters of the best chalcopyrite-based solar cells. This compilation clearly shows the superiority of Cu(In,Ga)Se2 with a relatively low Ga content, which leads to the actual world champion device. The fact that the best CuInSe2 devices has an efficiency of 3% below that of the best Cu(In,Ga)Se2 device is due not only to the less favourable band-gap energy but also to the lack of the beneficial effect of small amounts of Ga on film growth, discussed above.

326

U.RauandH.

W. Schock

Table 7.3 Operating parameters of the best Cu(In,Ga)Se2, CuInSe2, CuGaSe2 and CuInS2 cells, and the best pentenary Cu(In,Ga)(S,Se)2 cell

VJmV /a/niA cm"2 FFf%

4/cm 2

Ref.

78.6

0.449

1

41.2

72.6

0.38

2

861

14.2

67.9

0.471

3

13.9*

775

24.3

74.0

0.5

4

11.1"

728

21.24

70.9

0.48

5

Material

tfg/eV f?/%

Cu(In,Ga)Se2

1.12

18.8"

678

35.2

CuInSe2

1.04

15.4*

515

CuGaSe2

1.68

8.3"

Cu(In,Ga)(S,Se)2

1.36

CuInS2

1.57

"Confirmed total area values; 'effective area values (not confirmed). References: 1. Contreras et al. (1999); 2. Stolt et al. (1993); 3. Nadenau et al. (1997); 4. Friedlmeier and Schock (1998); 5. Klaer et al. (1998).

The difficulty of obtaining wide-gap devices with high efficiencies is also illustrated by plotting the absorber band gap of a series of chalcopyrite alloys vs. the attained open-circuit voltages. Figure7.26 shows that below Ug = 1.3 eV, the data follow the straight line Voc = (Ug 14) - 0-5 eV, indicating a proportional gain in V^ with increasing Ug, whereas at Ug> 1.3 eV the gain is much more moderate. At the high band-gap end of the scale, the differences between the band-gap energies and the open-circuit voltages of CuInS2 and CuGaSe2 amount to 840 mV and 820 mV, respectively, whereas (Uglq) - Voc is only 434 eV in the record Cu(In,Ga)Se2 device. 1.2

5

!

'

11

(D D)

jS 0.9 §

4'

1 0.8

•

^

^

'o c 0.7 °

-

-?&• m ? •

0.6 W ,

0.5 1.0 CulnSe 2

i

1.1

1

i

I

1.2

1 ,,„!

1.3

1

1.4

, i ...

1.5

Band-gap energy L/g/eV CulnSz

i

1.6

1.7 CuGaSe 2

Figure 7.26 Open-circuit voltages of different Cu-chalcopyrite based solar cells of different band-gap energies. Full squares correspond to Cu(In,Ga)Se2 alloys, open squares to CuIn(S,Se)2, red circles and black crosses to Cu(In,Ga)(S,Se)2, downward triangles to CulnS2, and upward triangles to CuGaSe2.

327

Cu(In, Ga)Se2 Solar Cells

One reason for the large differences in Ug/q - V<>c in wide-gap devices is the less favourable band offset constellation at the absorber/CdS-buffer interface. Figure 7.27 shows the band diagram of a CuGaSe2-based heteroj unction. As the increase of band gap in going from CuInSe2 to CuGaSe2 takes place almost exclusively by increase of the energy of the conduction band, the positive band offset AUf between the absorber and the buffer in Fig. 7.18 turns into a negative one in Fig. 7.27. This implies that the barrier $£ that hinders the holes from the absorber from recombining with the electrons from the buffer does not increase proportionally with increase in the band-gap energy. Thus the importance of interface recombination (dominated by the barrier
ZnO

Figure 7.27 Energy band diagram for the ZnO/CdS/CuGaSe2 heteroj unction.

Using the same arguments with respect to the MoSe2/absorber interface, the backsurface field produced by this type of back contact turns into its contrary because the built-in field acts in the opposite direction when the conduction-band energy of the absorber is increased. At the moment, this drawback of Cu(In,Ga)Se2 with high Ga contents seems to be of minor importance. More substantial are the changes of the electronic quality of the bulk material with increasing x. Figure 7.28 (Hanna et al., 2000) compares the defect density spectra of different CuIni_/ja;cSe2 alloys with x = 0, 0.26, and 0.57. Two features are obvious: 1. The bulk defect at activation energy Ua = 300 meV displays a minimum defect density D for the composition x = 0.26. Higher band-gap material (x = 0.56), as well as lower band-gap material (x = 0), has higher defect densities. It might not be

328

U. Rau andH. W. Schock

incidental that the superior electronic quality of the material is found at that composition which is used to produce the highest efficiency Cu(In,Ga)Se2 devices. 2. The interface-related peak at lower activation energies for the compositions x = 0 and 0.26 is no longer visible at x = 0.56. This is just what we would expect from the band diagram shown in Fig. 7.28b: the Fermi level at the interface moves towards mid-gap and the type inversion is no longer seen in admittance spectroscopy.

Activation energy Ua/eV Figure 7.28 Admittance defect density spectra of solar cells with different CuIni-^Ga^Sei alloy compositions. The Ga content is varied from (a) x = 0 to (c) x = 0.58. The lowest density for the acceptor defect with an activation energy of ~300 meV is found at x = 0.26 (b).

Thus we see that bulk as well as interface properties change when going from low Ga content Cu(In,Ga)Se2 towards higher band-gap compositions over a limit which appears to be at x = 0.3, corresponding to a band-gap energy of 1.3 eV. A similar limit for S/Se alloying is not yet well established. It appears, however, that making use of the full alloy system Cu(In,Ga)(S,Se)2 enables us to go beyond the limit of 1.3 eV while preserving a high efficiency level.

7.5.7

CuGaSe2

CuGaSe2 has a band gap of 1.68 eV and therefore would represent an ideal partner for CuInSe2 in an all-chalcopyrite tandem structure. However, a reasonable efficiency for the top cell of any tandem structure is about 15%, far higher than has been reached by the present polycrystalline CuGaSe2 technology. Despite the recent increase of the record efficiency from 6.3% to confirmed 8.3% (9.3% active area) (Nadenau et al., 1997), we are still in the process of learning about CuGaSe2. The electronic properties of the material are not so far from those of CuInSe2. However, in detail, all the

Cu(In,Ga)Se2 Solar Cells

329

differences quantitatively point in a less favourable direction. In general, the net doping density NA in CuGaSe2 appears too high (Jasenek et al., 2000b). Together with the charge of deeper defects, the high doping density leads to a strong electrical field in the space-charge region, which enhances recombination by tunnelling. Furthermore, a structural difference exists between the bulk CuGaSe2 (chalcopyrite) phase and the surface (defect chalcopyrite) phase. Regardless of whether or not this 135-phase is perfectly formed at the film surface or is a defect layer, a lattice mismatch between the surface layer and the bulk material could account for the increased defect density, which seems operative for CuGaSe2 as well as for Cu(In,Ga)Se2 alloys with high Ga contents (Contreras et al, 1997b).

7.5.2 CuInSj The major difference between CuInS2 and Cu(In,Ga)Se2 is that the former cannot be prepared with an overall Cu-poor composition. Cu-poor CuInS2 displays an extremely low conductivity, making it almost unusable as a photovoltaic absorber material (Walter et al., 1992). Even at small deviations from stoichiometry on the In-rich side, segregation of the spinel phase is observed (Walter and Schock, 1993). Instead, the material of choice is Cu-rich CuInS2. As in the case of CuInSe2, a Cu-rich preparation route implies the removal of the unavoidable secondary phase (here, Cu2_^S) by etching the absorber in KCN (Scheer et al., 1993). Such an etch may involve some damage to the absorber surface as well as the introduction of shunt paths between the front and back electrodes. However, as shown in Table 7.3, the best CuInS2 device has an efficiency above 10% (Klaer et al., 1998). Remarkably, this is achieved by a sulphurisation process rather than by co-evaporation. The higher quality of the material from sulphurisation might be due to the higher activity of sulphur, and the consequently lower concentration of sulphur vacancies. Although results from coevaporated samples come close to those of sulphurised cells, CuInS2 is at the moment the only chalcopyrite for which two-step preparation proves itself to be superior to coevaporation. The main limitation of efficiency, at about 12%, comes from an opencircuit voltage that is too low as compared with the band gap. By adding Zn (Walter et al., 1995) or Ga (Neisser et al., 1999) to the absorber layer, voltages above 0.8 V can be achieved. However, efficiencies are still limited by reduced fill factors or shortcircuit current densities.

330

U. Ran and H. W. Schock

7.5.3 Cu(In,Ga)(S,Se)2 One possible way of overcoming the disadvantages of the ternary wide-gap materials CuInS2 and CuGaSe2 is to use the full pentenary alloy CuIn1_^Gax(Si_),Se>,)2 (Friedlmeier and Schock, 1998). Among the materials listed in Table 7.3, the pentenary system is the only one with an open-circuit voltage larger than 750 mV and an efficiency above 13%, outperforming CuInS2 in both these respects. The advantage of Cu(In,Ga)(S,Se)2 could arise from the mutual compensation of the drawback of CuGaSe2 (too high a charge density) and that of (Cu-poor) CuInS2 (too low a conductivity). Electrical analysis of Cu(In,Ga)(S,Se)2 demonstrates that even with a band-gap energy of 1.3 eV and more, this material still preserves the main features of T

•

J

i

I

'

I

•

r


>, § •D

2 io 1B 0.0

0.1

L

0.2

0.3

0.4

0.5

Activation energy 14/ eV

Figure 7.29 Defect density spectrum obtained from admittance spectroscopy of a Cu(In,Ga)(S,Se)2based heterojunction solar cell.

Cu(In,Ga)Se2. Figure 7.29 shows that the defect density spectrum of a Cu(In,Ga)(S,Se)2 solar cell is close to what we observe in high-efficiency Cu(In,Ga)Se2 cells. Compared with the high-Ga-content Cu(In,Ga)Se2 device shown in Fig. 7.28, the Cu(In,Ga)(S,Se)2 device displays a relatively moderate value of D™" = 2 x 1016 cnT3eV_1 for the maximum bulk defect density at the activation energy Ua fv 300 meV relevant to recombination. As in standard Cu(In,Ga)Se2, an interfacerelated transition appears at energy Ua ~ 200 meV, indicating the preservation of type inversion at the buffer/absorber interface. Thus, it seems that the overall positive features present in Cu(In,Ga)Se2 with Ug > 1.3 eV can be maintained for larger bandgap energies if one makes use of the full alloy system Cu(In,Ga)(S,Se)2.

Cu(In,Ga)Se2 Solar Cells 7.5.4

331

Graded-gap devices

An interesting property of the CuIni^Ga^S^Se^ alloy system is the possibility of designing graded-gap structures that optimise the electronic properties of the final device (Gray and Lee, 1994; Dhingra and Rothwarf, 1996; Gabor et al., 1996; Dullweber et al., 2000). Such band-gap gradings are achieved during co-evaporation by the control of the elemental sources, but selenisation/sulphurisation processes also lead to beneficial compositional gradings. The art of designing optimum band-gap gradings is to push back charge carriers from critical regions, i.e. regions with high recombination probability within the device. Such critical regions are 1) the interfaces between the back contact and the aborber layer; 2) the heterojunction, including the absorber/buffer interface. Figure 7.30 shows a band diagram of a grading structure that fulfils the requirements for minimising recombination losses. 1. To keep the back contact region clear from electrons, one can use a Ga/In grading. The increase of the Ga/(Ga + In) ratio x causes a movement of the conduction-band minimum upward with respect to its position in pure CuInSe2. An increase of x towards the back surface leads to a gradual increase of the conduction-band energy, as illustrated in Fig. 7.30. The resulting back-surface field, as in the Cu(In,Ga)Se2/MoSe2 heterocontact, drives photogenerated electrons away from the metallic back contact towards the buffer/absorber junction. 2. The minimisation of junction recombination, both at the point of equal capture rates of holes and electrons and at the metallurgical interface between absorber and buffer, requires a larger band gap towards the front contact to the absorber. If one had the choice, one would clearly favour a decrease of the valence-band energy, as shown in Fig. 7.30, over an increase of the conduction-band energy. This favours a grading with the help of S/Se alloying, as at least a part of the increasing band-gap energy is supported by a decrease of valence-bandedge energy. The decreased valence-bandedge energy in Fig. 7.30 leads to an increase of the barrier <&pb and thus minimises interface recombination. The favourable combination of In/Ga-grading towards the back contact and S/Se grading towards the front contact appears to be an inherent feature of the combined selenisation/sulphurisation processes, which has led to the highest efficiency devices obtained to date from sequential processing of stacked metallic or stacked elemental precursors.

332

U.RauandH. W. Schock

i-ZnO

Figure 7.30 Band diagram of an optimised graded-gap device with an increasing Ga/ln ratio towards the back surface (region I) and an increasing S/Se-ratio towards the front (region HI). Region II has no grading. The dotted lines correspond to the conduction- and valence-band energies of the non-graded device.

7.6 Conclusions The objective of this chapter was not only to describe the achievements of Cu(In,Ga)Se2-based solar cells, but also to give an account of our present understanding of the physical properties of the materials involved and the electronic behaviour of the devices. The fortunate situation of Cu(In,Ga)Se2, which is in a leading position among polycrystalline thin-film solar cell materials, arises from the benign, forgiving nature of the bulk materials and interfaces. Nevertheless, we want to draw the attention of the reader also to the work that has still to be done. As the trail to efficiencies close to or above 20% could become more and more narrow, a profound knowledge of what is to the left and to the right becomes more and more vital. Knowhow must be transferred into know-why. Three of the four cornerstones 1-4 for the recent achievements mentioned in Section 7.3 concern the growth of the films: the optimised deposition conditions, and the incorporation of Na and Ga. However, no detailed model is available definitively to describe the growth of Cu(In,Ga)Se2, and especially the impact of Na, which in our

Cu(In,Ga)Se2 Solar Cells

333

opinion is the most important of the different ingredients available to tune the electronic properties of the absorber. A clearer understanding of Cu(In,Ga)Se2 growth would allow us to find optimised conditions in the wide parameter space available, and thus to reduce the number of recombination centres and compensating donors and optimise the number of shallow acceptors. The deposition of the buffer layer, or more generally speaking, the formation of the heteroj unction, is another critical issue. The surface chemistry during heteroj unction formation and post-deposition treatments is decisive for the final device performance. Both processes greatly affect not only the surface defects (i.e., recombination and charge), and therefore the charge distribution in the device, but also the defects in the bulk of the absorber. Concentrated effort and major progress in these tasks would not only allow us to push the best efficiencies further towards 20%, but would also provide a sound knowledge base for the various attempts at commercialisation of Cu(In,Ga)Se2 solar cells. After more than twenty years of research, commercialisation is the real last frontier for this fascinating material. We are optimistic that Cu(In,Ga)Se2 can now deliver on its promises and enter the market place on a large scale.

Acknowledgements We thank all our colleagues in the IPE for various discussions and fruitful collaboration. We are especially grateful to J. H. Werner for his continuous support and his interest in this work. We thank H. Kerber, A. Jasenek, T. Dullweber, and G. Hanna for supplying some of the material used in this chapter, as well as I. Kotschau for the help with the bibliography. Critical reading of the manuscript by F. Pfisterer is especially acknowledged. This work is in major part supported by the German Ministry of Economics (BMWi) under contract 0328059F, and by the European Commission under contract JOR3-CT97-0149.

References Abken A., Heinemann F., Kampmann A., Leinkuhler G., Rechid J., Sittinger V., Wietler T. and Reineke-Koch R. (1998), 'Large area electrodeposition of Cu(In,Ga)Se2 precursors for the fabrication of thin film solar cells', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 1133-1136.

334

U. Rau and H. W. Schock

Abou-Elfotouh F. A., Kazmerski L. L., Moutinho H. R., Wissel J. M., Dhere R. G., Nelson A. J. and Bakry A. M. (1991), 'Determination and observation of electronic defect levels in CuInSe2 crystals and thin films', J. Vac. Sci. Technol. A 9, 554-558. Arora N. D., Chamberlain S. G. and Roulston D. J. (1980), 'Diffusion length determination in p-n junction diodes and solar cells', Appl. Phys. Lett. 37, 325327. Bardeen J. (1947), 'Surface states and rectification at a metal/semiconductor contact', Phys. Rev. 71,717-727. Binsma J. J. M. and Van der Linden H. A. (1982), 'Preparation of thin CuInS2 films via a two-stage process', Thin Solid Films 97, 237-243. Birkmire R. W., McCandless B. E., Shafarman W. N. and Varrin R. D. (1989), 'Approaches for high efficiency CuInSe2 solar cells', Proc. 9th. European Photovoltaic Solar Energy Conf., Freiburg, Kluwer Academic Publishers, Dordrecht, 134-137. Braunger D., Hariskos D. and Schock H. W. (1998a), 'Na-related stability issues in highly efficient polycrystalline Cu(In,Ga)Se2 solar cells', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 511-514. Braunger D., Zweigart S. and Schock H. W. (1998b), 'The influence of Na and Ga on the incorporation of the chalcogen in polycrystalline Cu(In,Ga)(S,Se)2 thin-films for photovoltaic applications', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 1113-1116. Brendel R. and Rau U. (1999), 'Injection and collection diffusion lengths of polycrystalline thin-film solar cells', Solid State Phenomena 67-68, 81-86. Bube R. H. (1992), Photoelectronic Properties of Semiconductors, Cambridge University Press, Cambridge. Bube R. H. (1998), Photovoltaic Materials, Imperial College Press, London. Burgelman M., Engelhardt F., Guillemoles J.-F., Herberholz R., Igalson M., Klenk R., Lampert M., Meyer T., Nadenau V., Niemegeers A., Parisi J., Rau U., Schock H.W., Schmitt M., Seifert O., Walter T. and Zott S. (1997), 'Defects in Cu(In,Ga)Se2 semiconductors and their role in the device performance of thin-film solar cells', Prog. Photovoltaic Res. Appl. 5, 121-130. Cahen D. (1987), 'Some thoughts about defect chemistry in ternaries', Proc. 7th. Int. Conf. Ternary and Multinary Compounds, Mat. Res. Soc, Pittsburgh, 433-442. Cahen D. and Noufi R. (1989), 'Defect chemical explanation for the effect of air anneal on CdS/CuInSe2 solar cell performance', Appl. Phys. Lett. 54, 558-560.

Cu(In,Ga)Se2 Solar Cells

335

Chu T. L., Chu S. C , Lin S. C. and Yue J. (1984), 'Large grain copper indium diselenide films', J. Electrochem. Soc. 131,2182-2184. Contreras M. A., Egaas B., Dippo P., Webb J., Granata J., Ramanathan K., Asher S., Swartzlander A. and Noufi R. (1997a), 'On the role of Na and modifications to Cu(In,Ga)Se absorber materials using thin MF (M = Na, K, Cs) precursor layers', Conf. Record. 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 359-362. Contreras M. A., Wiesner H., Tuttle J., Ramanathan K. and Noufi R. (1997b), 'Issues of the chalcopyrite/defect-chalcopyrite junction model for high-efficiency Cu(In,Ga)Se2 solar cells', Solar Energy Mat. Solar Cells 49, 239-247. Contreras M. A., Egaas B., Ramanathan K., Hiltner J., Swartzlander A., Hasoon F. and Noufi R. (1999), 'Towards 20% efficiency in Cu(In,Ga)Se2 polycrystalline solar cells', Prog. Photovoltaic Res. Appl. 7, 311-316. Coutts T. J., Kazmerski L. L. and Wagner S. (1986, ed.), Ternary Chalcopyrite Semiconductors: Growth, Electronic Properties, and Applications, Elsevier, Amsterdam. Devaney W. E., Chen W. S., Steward J. M. and Mickelson R. A. (1990), 'Structure and properties of high efficiency ZnO/CdZnS/CuInGaSe2 solar cells', IEEE Trans. Electron Devices ED-37, 428^133. Dhingra A. and Rothwarf A. (1996), 'Computer simulation and modeling of graded bandgap CuInSe2/CdS based solar cells', IEEE Trans. Electron Devices 43, 613— 621. Dimmler B. and Schock H. W. (1996), 'Scaling-up of CIS technology for thin-film solar modules', Prog. Photovoltaic Res. Appl. 4,425-433. Dullweber T., Hanna G., Shams-Kolahi W., Schwartzlander A., Contreras M. A., Noufi R. and Schock H. W. (2000), 'Study of the effect of gallium grading in Cu(In,Ga)Se2\ Thin Solid Films 361-362,478^181. Eberspacher C , Pauls K. L. and Fredric C. V. (1998), 'Improved processes for forming CuInSe2 films', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 303-306. Engelhardt F., Schmidt M., Meyer Th., Seifert O., Parisi J. and Rau U. (1998), 'Metastable electrical transport in Cu(In,Ga)Se2 thin films and ZnO/CdS/Cu(In,Ga)Se2 heterostructures', Phys. Lett. A 245,489^93.

336

U.RauandH. W. Schock

Engelhardt F., Bornemann L., Kontges M., Meyer Th., Parisi J., Pschorr-Schoberer E., Hahn B., Gebhardt W., Riedl W. and Rau U. (1999), 'Cu(In,Ga)Se2 solar cells with a ZnSe buffer layer—interface characterization by quantum efficiency measurements', Prog. Photovoltaic Res. Appl. 7, 423-436. Friedlmeier T. M., Braunger D., Hariskos D., Kaiser M., Wanka H. N. and Schock H. W. (1996), 'Nucleation and growth of the CdS buffer layer on Cu(In,Ga)Se2 thin films', Conf. Record 25th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 845-848. Friedlmeier T. M. and Schock H. W. (1998), 'Improved voltages and efficiencies in Cu(In,Ga)(S,Se)2 solar cells', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 1117-1120. Gabor A. M., Tuttle J. R., Albin D. S., Contreras M. A., Noufi R. and Hermann A. M. (1994), 'High-efficiency CuInA:Gai_xSe2 solar cells from (In^Gai_^)2Se3 precursors', Appl. Phys. Lett. 65, 198-200. Gabor A. M., Tuttle J. R., Bode M. H., Franz A., Tennant A. L., Contreras M. A., Noufi R., Jensen D. G. and Hermann A. M (1996), 'Band-gap engineering in Cu(In,Ga)Se2 thin films grown from (In,Ga)2Se3 precursors', Solar Energy Mat. Solar Cells 41, 247-260. Gallon P. N. R., Orsal G., Artaud M. C. and Duchemin S. (1998), 'Studies of CuInSe2 and CuGaSe2 thin films grown by MOCVD from three organometallic sources', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 515-518. Gay R. R., 'Prerequisites to manufacturing of thin-film photovoltaics', Prog. Photovoltaic Res. Appl. 5, 337-343. Gray J. L. and Lee Y. J. (1994), 'Numerical modeling of graded band gap CIGS solar cells', Proc. 1st. World Conf. Photovoltaic Solar Energy Conversion, Waikoloa, IEEE Press, Piscataway, 123-126. Green M. A. (1986), Solar Cells: Operating Principles, Technology and System Applications, University of New South Wales, Sydney. Green M. A. (1995), Silicon Solar Cells: Advanced Principles and Practice, University of New South Wales, Sydney. Green M. A. (1996a), 'Depletion region recombination in silicon thin-film multilayer solar cells', Prog. Photovoltaic Res. Appl. 4, 375-380. Green M. A. (1996b), 'Bounds upon grain boundary effects in minority carrier semiconductor devices: A rigorous "perturbation" approach with application to silicon solar cells', J. Appl. Phys. 80, 1515-1521.

Cu(In, Ga)Se2 Solar Cells

337

Guillemoles J.-F. (2000), 'Stability of Cu(In,Ga)Se2 solar cells: a thermodynamic approach', Thin Solid Films 361-362, 338-345. Haalboom T., Godecke T., Ernst F., Rtthle M., Herberholz R., Schock H. W. Beilharz C. and Benz K. W. (1997), 'Phase relations and microstructure in bulk materials and thin films of the ternary system Cu-InSe', Inst. Phys. Conf. Ser. 152E, 249252. Hahn H., Frank G., Klingler W., Meyer A. and Storger G. (1953), 'Uber einige ternare Chalkogenide mit Chalkopyritstruktur', Z anorg. u. allg. Chemie 271, 153-170. Hanna G., Jasenek A., Rau U. and Schock H. W (2000), 'Open-circuit voltage limitations in CuIni_xGajtSe2 thin film solar cells—dependence on alloy compositions', to be published. Hedstrom J., Ohlsen, H., Bodegard M., Kylner A., Stolt L., Hariskos D., Ruckh M. and Schock H.-W. (1993), 'ZnO/CdS/Cu(In,Ga)Se2 thin film solar cells with improved performance', Conf. Record 23rd. IEEE Photovoltaic Specialists Conf., Louisville, IEEE Press, Piscataway, 364-371. Hengel I., Neisser A., Klenk R. and Lux-Steiner C.-M. (2000), 'Current transport in CuInS2:Ga/CdS/ZnO solar cells', Thin Solid Films 361-362,458-462. Herberholz R., Braunger D., Schock H. W., Haalboom T. and Ernst F. (1997a), 'Performance and defects in Cu(In,Ga)Se2 heterojunctions: combining electrical and structural measurements', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 1246-1249. Herberholz R., Nadenau V., Riihle U., Koble C , Schock H. W. and Dimmler B. (1997b), 'Prospects of wide-gap chalcopyrites for thin film photovoltaic modules', Solar Energy Mat. Solar Cells 49, 221-231. Herberholz R. (1998), 'Defect characterisation in chalcopyrite-based heterostructures', Inst. Phys. Conf. Ser. 152E, 733-740. Herberholz R., Igalson M. and Schock H. W. (1998), 'Distinction between bulk and interface states in CuInSe2/CdS/ZnO by space charge spectroscopy', J. Appl. Phys. 83, 318-325. Herberholz R., Rau U., Schock H. W., Hallboom T., Godecke T., Ernst F., Beilharz C , Benz K. W. and Cahen D. (1999), 'Phase segregation, Cu migration and junction formation in Cu(In,Ga)Se2', Eur. Phys. J.: Appl. Phys. 6, 131-139. Heske C , Fink R., Umbach E., Riedl W. and Karg F. (1996), 'Na-induced effects on the electronic structure and composition of Cu(In,Ga)Se2 thin-film surfaces', Appl. Phys. Lett. 68, 3431-3432. Heske C , Richter G., Chen Z., Fink R., Umbach E., Riedl W. and Karg F. (1997), 'Influence of Na and H 2 0 on the surface properties of Cu(In,Ga)Se2 thin films', J. Appl. Phys. 82, 2411-2420.

338

U. Rau and H. W. Schock

Holz J., Karg F. and v. Phillipsborn H. (1994), 'The effect of substrate impurities on the electronic conductivity in CIGS thin films, Proc. 12th. European Photovoltaic Solar Energy Conf., Amsterdam, H. S. Stephens & Associates, Bedford, 15921595. Igalson M., Bacewicz R. and Schock H. W. (1995), ' "Dangling bonds" in CuInSe2 and related compounds', Proc. 13th. European Photovoltaic Solar Energy Conf., Nice, H. S. Stephens & Associates, Bedford, 2076-2079. Igalson M. and Schock H. W. (1996), 'The metastable changes of the trap spectra of CuInSe2-based photovoltaic devices', /. Appl. Phys. 80, 5765-5769. Jager-Waldau G., Schmid D. and Jager-Waldau A. (1991), 'Diffusion length measurement of heterojunction thin films by junction-EBIC, /. Phys. IV 1, C 6 131-132. Jasenek A., Rau U., Hahn T., Hanna G., Schmidt M., Hartmann M., Schock H. W., Werner J. H., Schattat B., Kraft S., Schmid K.-H. and Boise W. (2000a), 'Defect generation in polycrystalline Cu(In,Ga)Se2 by high-energy electron irradiation', Appl. Phys. A (in press). Jasenek A., Rau U., Nadenau V. and Schock H. W. (2000b), 'Electronic properties of CuGaSe2-based heterojunction solar cells. Part II—Defect spectroscopy', J. Appl. Phys. 87, 594-602. Jensen N., Rau U., Hausner R. M., Uppal S., Oberbeck L., Bergmann R. B. and Werner J. H. (2000), 'Recombination mechanisms in amorphous silicon/ crystalline silicon heterojunction solar cells', J. Appl. Phys. 87, 2639-2645. Kapur V. K., Basol B. M. and Tseng E. S. (1987), 'Low-cost methods for the production of semiconductor films for CuInSe2/CdS solar cells', Solar Cells 21, 65-72. Karg F. (1999), 'Development and manufacturing of CIS thin film modules', Tech. Digest 11th. Int. Sci. Eng. Conf., Tokyo University of Agriculture and Technology, Tokyo, 627-630. Kazmerski L. L. and Wagner S. (1985), 'Cu-ternary chalcopyrite solar cells', in Coutts T. J. and Meakin J. D., Current Topics in Photovoltaics, Academic Press, Orlando, pp. 41-110. Kessler J., Velthaus K. O., Ruckh M., Laichinger R., Schock H. W., Lincot D., Ortega R. and Vedel J. (1992), 'Chemical bath deposition of CdS on CuInSe2, etching effects and growth kinetics', Proc. 6th. Int. Photovoltaic Solar Energy Conf, New Delhi, Oxford IBH Publishing, New Delhi, 1005-1010. Kessler J., Bodegard M., Hedstrbm J. and Stolt L. (1999), 'Baseline Cu(In,Ga)Se2 device production: control and significance', Tech. Digest 11th. Int. Sci. Eng. Conf, Tokyo University of Agriculture and Technology, Tokyo, 631-632.

Cu(In,Ga)Se2 Solar Cells

339

Keyes B. M., Hasoon F., Dippo P., Balcioglu A. and Aboulfotuh F. (1997), 'Influence of Na on the electro-optical properties of Cu(In,Ga)Se2\ Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 479^482. Klaer J., Bruns J., Henninger R., Topper K., Klenk R., Ellmer K. and Braunig D. (1998), 'A tolerant two-step process for efficient CuInS2 solar cells', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 537-540. Klein A. and Jaegermann W. (1996), 'Chemical interaction of Na-cleaved (001) surfaces of CuInSe2', J. Appl. Phys. 80, 5039-5043. Klenk R., Walter T., Schock H. W. and Cahen D. (1993), 'A model for the successful growth of polycrystalline films of CuInSe2 by multisource physical vacuum evaporation', Adv. Mat. 5, 114-119. Klenk R. and Schock H. W. (1994), 'Photocurrent collection in thin film solar cells— calculation and characterization for CuGaSe2/(Zn,Cd)S', Proc. 12th. European Photovoltaic Solar Energy Conf, Amsterdam, H. S. Stephens & Assoicates, Bedford. Kronik L., Cahen D. and Schock H. W. (1998), 'Effects of sodium on polycrystalline Cu(In,Ga)Se2 and its solar cell performance', Adv. Mat. 10, 31-36. Kronik L., Rau U., Guillemoles J.-F., Braunger D., Schock H. W. and Cahen D. (2000), 'Interface redox engeering of Cu(In,Ga)Se2-based solar cells: oxygen, sodium, and chemical bath effects', Thin Solid Films 361-362, 353-359. Kushyia K., Tachiyuli M., Nagoya Y., Fujimaki A., Sang B., Okumara D., Satoh M. and Yamase O. (1999), 'Progress in large-area Cu(In,Ga)Se2-based thin-film modules with a Zn(0,S,OH)^ buffer layer', Tech. Digest 11th. Int. Sci. Eng. Conf, Tokyo University of Agriculture and Technology, Tokyo, 637-640. Lincot D., Guillemoles J.-F., Cowache P., Marlot A., Lepiller C , Canava B., Yousfi F. B. and Vedel J. (1998), 'Solution deposition technologies for thin film solar cells: status and perspectives', Proc. 2nd. World Conf Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 440-445. Loher T., Jaegermann W. and Pettenkofer C. (1995), 'Formation and electronic properties of the CdS/CuInSe2 (001) heterointerface studied by synchrotroninduced photoemission', J. Appl. Phys. 77, 731-737. Mauch R. H., Ruckh. M., Hedstrom J., Lincot D., Kessler J., Klinger R., Stolt L. Vedel J. and Schock H. W. (1991), 'High efficiency Zno/CdS/CuInSe2 solar cells—recent results of EUROCIS collaboration', Proc. 9th. European Photovoltaic Solar Energy Conf, Lisbon, Kluwer Academic Publishers, Dordrecht, 1415-1419.

340

U. Rau and H. W. Schock

Meyer T., Schmidt M., Engelhardt F., Parisi J. and Rau U. (1999), 'A model for the open-circuit voltage relaxation in Cu(In,Ga)Se2 heterojunction solar cells', Eur. Phys. J.: Appl. Phys. 8, 43-52. Mickelsen R. A. and Chen W. S. (1980), 'High photocurrent polycrystalline thin-film CdS/CuInSe2 solar cell', Appl. Phys. Lett. 36, 371-373. Mickelsen R. A. and Chen W. S. (1982), 'Polycrystalline thin-film CuInSe2 solar cells', Conf. Record 16th. IEEE Photovoltaic Specialists Conf., San Diego, IEEE Press, Piscataway, 781-785. Migliorato P., Shay J. L., Kasper H. M. and Wagner S. (1975), 'Analysis of the electrical and luminescent properties of CuInSe2', J. Appl. Phys. 46, 1777-1782. Mitchell K. C , E., Ermer J. and Pier D. (1988), 'Single and tandem junction CuInSe2 cell and module technology', Conf. Record 20th. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 1384-1389. Nadenau V., Hariskos D. and Schock H. W. (1997), 'CuGaSe2 based thin film solar cells with improved performance', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford. Nadenau V., Rau U., Jasenek A. and Schock H. W. (2000), 'Electronic properties of CuGaSe2-based heteroj unction solar cells. Part I—Transport analysis', J. Appl. Phys. 87, 584-593. Nakada T., Iga D., Ohbo H. and Kunioka A. (1997), 'Effects of sodium on Cu(In,Ga)Se2-based thin films and solar cells', Jpn. J. Appl. Phys. 36, 732-737. Nakada T., Mise T., Kume T. and Kunioka A. (1998), 'Superstate type Cu(In,Ga)Se2 thin film solar cells with ZnO buffer layers—a novel approach to 10% efficiency', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 413-417. Neisser A., Hengel I., Klenk R., Matthes Th. W., Alvarez-Garcia J., Perez-Rodriguez A., Romano-Rodriguez A. and Lux-Steiner M. Ch. (1999), 'Effect of Ga incorporation in sequentially prepared CuInS2 thin film absorbers', Tech. Digest 11th. Int. Sci. Eng. Conf, Tokyo University of Agriculture and Technology, Tokyo, 955-956. Nelson A. J., Schwerdtfeger C. R., Wei S.-H., Zunger A., Rioux D., Patel R. and Hochst H. (1993), 'Theoretical and experimental studies of the ZnSe/CuInSe2 heterojunction band offset', Appl. Phys. Lett. 62, 2557-2559 Neumann H. (1983), 'Vacancy formation in AIBinC2VI chalcopyrite semiconductors', Cryst. Res. Technol. 18,901-906. Neumann H. (1986), 'Optical properties and electronic band structure of CuInSe2', Solar Cells 16, 317-333.

Cu(In,Ga)Se2 Solar Cells

341

Niemegeers A., Burgelman M. and De Vos A. (1995), 'On the CdS/CuInSe2 conduction band discontinuity', Appl. Phys. Lett. 67, 843-845. Niemeegers A. and Burgelman M. (1996), 'Numerical modelling of AC characteristics of CdTe and CIS solar cells', Conf. Record 25th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 901-904. Niemegeers A., Burgelman M., Herberholz R., Rau U., Hariskos D. and Schock H. W. (1998), 'Model for electronic transport in Cu(In,Ga)Se2 solar cells', Prog. Photovoltaic Res. Appl. 6,407-421. Niki S., Fons P. J., Yamada A., Suzuki R., Ohdaira T., Ishibashi S. and Oyanagai H. (1994), 'High quality CuInSe2 epitaxial films—molecular beam epitaxial growth and intrinsic properties', Inst. Phys Conf. Ser. 152E, 221-227. Niles D. W., Ramanathan K., Haason F., Noufi R., Tielsh B. J. and Fulghum J. E. (1997), 'Na impurity chemistry in photovoltaic CIGS thin films: Investigation with X-ray photoelectron spectroscopy', J. Vac. Sci. Technol. A 15, 3044-3049. Niles D. W., Al-Jassim M. and Ramanathan K. (1999), 'Direct observation of Na and O impurities at grain surfaces of CuInSe2 thin films', /. Vac. Sci. Technol. A 17, 291-296. Noufi R., Axton R., Herrington C. and Deb S. K. (1984), 'Electronic properties versus composition of thin films of CuInSe2', Appl. Phys. Lett. 45, 668-670. Padovani F. A. and Stratton R. (1966), 'Field and thermionic field emission in Schottky barriers', Solid State Electron 9, 695-707. Parisi J., Hilburger D., Schmitt M. and Rau U. (1998), 'Quantum efficiency and admittance spectroscopy on Cu(In,Ga)Se2 solar cells', Solar Energy Mat. Solar Cells 50, 79-85. Potter R. R., Eberspacher C. and Fabick L. B. (1985), 'Device analysis of CuInSe2/ (Cd,Zn)S solar cells', Conf. Record 18th. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 1659-1664. Probst V., Karg F., Rimmasch J., Riedl W., Stetter W., Harms H. and Eibl O. (1996), 'Advanced stacked elemental layer progress for Cu(InGa)Se2 thin film photovoltaic devices', Mat. Res. Soc. Symp. Proc. 426, 165-176. Ramanathan K., Bhattacharya R. N., Granata J., Webb J., Niles D., Contreras M. A., Wiesner H., Haason F. S. and Noufi R. (1998a), 'Advances in CIGS solar cell research at NREL', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 319-325.

342

U. Rau and H. W. Schock

Ramanathan K., Wiesner H., Asher S., Niles D., Bhattacharya R. N., Keane J., Contreras M. A. and Noufi R. (1998b), 'High efficiency Cu(In,Ga)Se2 thin film solar cells without intermediate buffer layers', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 477-481. Rau U., Jasenek A., Herberholz R., Schock H. W., Guillemoles J. F., Lincot D. and Kronik L. (1998a), 'The inherent stability of Cu(In,Ga)Se2-based solar cells', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 428-433. Rau U., Schmitt M., Engelhardt F., Seifert O., Parisi J., Riedl W., Rimmasch J. and Karg F. (1998b), 'Impact of Na and S incorporation on the electronic transport mechanisms of Cu(In,Ga)Se2 solar cells', Solid State Commun. 107, 59-63. Rau U., Schmitt M., Parisi J., Riedl W. and Karg F. (1998c), 'Persistent photoconductivity in Cu(In,Ga)Se2 heteroj unctions and thin films prepared by sequential deposition', Appl. Phys. Lett. 74, 111-113. Rau U. (1999), 'Tunneling-enhanced recombination in Cu(In,Ga)Se2 heteroj unction solar cells', Appl. Phys. Lett. 74, 111-113. Rau U. and Schock H. W. (1999), 'Electronic properties of Cu(In,Ga)Se2 heterojunction solar cells—recent achievements, current understanding and future challenges', Appl. Phys. A 69, 131-147. Rau U., Braunger D., Herberholz R., Schock H. W., Guillemoles J.-F., Kronik L. and Cahen D. (1999a), 'Oxygenation and air-annealing effects on the electronic properties of Cu(In,Ga)Se2 films and devices', J. Appl. Phys. 86,497-505. Rau U., Braunger D. and Schock H. W. (1999b), 'Air-annealing effects on polycrystalline Cu(In,Ga)Se2 heterojunctions', Solid State Phenomena 67-68, 409414. Rau U., Jasenek A., Schock H. W., Engelhardt F. and Meyer T. (2000), 'Electronic loss mechanisms in chalcopyrite based heteroj unction solar cells', Thin Solid Films 361-362, 298-302. Rincon C , Bellabarba C , Gonzalez J. and Sanchez Perez G. (1986), 'Optical properties oand characterization of CuInSe2', Solar Cells 16, 335-349. i in vi Rincon C. and Wasim S. M. (1987), 'Defect chemistry of A B C2 chalcopyrite semiconducting compounds', Proc. 7th. Int. Conf. Ternary and Multinary Compounds, Mat. Res. Soc, Pittsburgh, 443-452. Rockett A. and Birkmire R. W. (1991), 'CuInSe2 for photovoltaic applications', J. Appl. Phys. 70, 81-97. Ruberto M. N. and Rothwarf A. (1987), 'Time-dependent open-circuit voltage in CuInSe2/CdS solar cells: theory and experiment', /. Appl. Phys. 61,4662-4669.

Cu(In, Ga)Se2 Solar Cells

343

Ruckh M., Schmid D., Kaiser M., Schaffler R., Walter T. and Schock H. W. (1994a), 'Influence of substrates on the electrical properties of Cu(In,Ga)Se2 thin films', Proc. 1st. World Conf. Photovoltaic Solar Energy Conversion, Waikoloa, IEEE Press, Piscataway, 156-159. Ruckh M., Schmid D. and Schock H. W. (1994b), 'Photoemission studies of the ZnO/CdS interface', J. Appl. Phys. 76, 5945-5948. Sasala R. A. and Sites J. R. (1993), 'Time dependent voltage in CuInSe2 and CdTe solar cells', Conf. Record 23rd IEEE Photovoltaic Specialists Conf., Louisville, IEEE Press, Piscataway, 543-548. Scheer R., Walter T., Schock H. W., Fearheiley M. L. and Lewerenz H. J. (1993), 'CuInS2 based thin film solar cells with 10.2% efficiency', Appl. Phys. Lett. 63, 3294-3296. Scheer R. (1997), 'Surface and interface properties of Cu-chalcopyrite semiconductors and devices', Research Trends in Vacuum Sci. Technol. 2, 77-112. Schmid D., Ruckh M., Grunwald F. and Schock H. W. (1993), 'Chalcopyrite / defect chalcopyrite heterojunctions on basis of CuInSe2', J. Appl. Phys. 73, 2902-2909. Schmid D., Ruckh M., Grunwald F. and Schock H. W. (1996a), 'Photoemission studies on Cu(In,Ga)Se2 thin films and related binary selenides', Appl. Surf. Sci. 103,409-429. Schmid D., Ruckh M. and Schock H. W. (1996b), 'A comprehensive characterization of the interfaces in Mo/CIS/CdS/ZnO solar cell stuctures', Solar Energy Mat. Solar Cells 41,281-294. Schmitt M., Rau U. and Parisi J. (1995), 'Investigation of deep trap levels in CuInSe2 solar cells by temperature dependent admittance measurements', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S. Stephens & Associates, Bedford, 1969-1972. Schock H. W. (1996), 'Thin film photovoltaics', Appl. Surf. Sci. 92, 606-616. Schock H. W. and Bogus K. (1998), 'Development of CIS solar cells for space applications', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 3586-3589. Schroeder D. J. and Rockett A. A. (1997), 'Electronic effects of sodium in epitaxial CuIni_^Ga^Se2', J. Appl. Phys. 82, 5982-5985. Seifert O., Engelhardt F., Meyer Th., Hirsch M. T., Parisi J., Beilharz C , Schmitt M. and Rau U. (1997), 'Observation of a metastability in the DC and AC electrical transport properties of Cu(In,Ga)Se2 single crystals, thin films and solar cells', Inst. Phys. Conf. Ser. 152E, 253-256.

344

U. Rau andH. W. Schock

Shay J. L. and Wernick J. H. (1975), Ternary Chalcopyrite Semiconductors: Growth, Electronic Properties, and Applications, Pergamon Press, Oxford. Staebler D. L. and Wronski C. R. (1977), 'Reversible conductivity changes in discharge-produced amorphous Si', Appl. Phys. Lett. 31, 292-294. Stolt L., Hedstrom J., Kessler J., Ruckh M., Velthaus K. O. and Schock H. W. (1993), 'ZnO/CdS/CuInSe2 thin-film solar cells with improved performance', Appl. Phys. Lett. 62, 597-599. Takei R., Tanino H., Chichibu S. and Nakanishi H. (1996), 'Depth profiles of spatially-resolved Raman spectra of a CuInSe2-based thin-film solar cell', J. Appl. Phys. 79, 2793-2795. Topic M., Smole F. and Furlan J. (1997), 'Examination of blocking current-voltage behavior through defect chalcopyrite layer in ZnO/CdS/Cu(In,Ga)Se2 solar cell', Solar Energy Mat. Solar Cells, 311-317. Wada T., Kohara N., Negami T. and Nishitani M. (1996), 'Chemical and structural charakterization of Cu(In,Ga)Se2/Mo Interface in Cu(In,Ga)Se2 solar cells', Jpn. J. Appl. Phys. 35, 1253-1256. Wada T., Hayashi S., Hashimoto Y., Nishiwaki S., Sato T., Negami T. and Nishitani M. (1998), 'High efficiency Cu(In,Ga)Se2 (CIGS) solar cells with improved CIGS surface', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 403-408. Wagner S., Shay J. L., Migliorato P. and Kasper H. M. (1974), 'CuInSe2/CdS heterojunction photovoltaic detectors', Appl. Phys. Lett. 25,434-435. Walter T., Braunger D., Hariskos D., Koble C. and Schock H. W. (1995a), 'CuInS2: film growth, devices, submodules and perspectives', Proc. 13th. European Photovoltaic Solar Energy Conf., Nice, H. S. Stephens & Associates, Bedford, 597-600. Walter T., Content A., Velthaus K. O. and Schock H. W. (1992), 'Solar cells based on CuIn(Se,S)2', Solar Energy Mat. Solar Cells 26, 357-368. Walter T. and Schock H. W. (1993), 'Structural and electrical investigations of the anion exchange in polycrystalline CuIn(S,Se)2 thin films', Jpn. J. Appl. Phys. 323, 116-119. Walter T., Herberholz R., Muller C. and Schock H. W. (1996a), 'Determination of defect distributions from admittance measurements and application to Cu(In,Ga)Se2 based heterojunctions', J. Appl. Phys. 80, 4411^420. Walter T., Herberholz R. and Schock H. W. (1996b), 'Distribution of defects in polycrystalline thin films', Solid State Phenomena 51, 309-316. Wei S. H. and Zunger A. (1993), 'Band offsets at the CdS/CuInSe2 heteroj unction', Appl. Phys. Lett. 63, 2549-2551.

Cu(In,Ga)Se2 Solar Cells

345

Wolf D., Muller G., Stetter W. and Karg F. (1998), 'In-situ investigation of Cu-In-Se reactions: impact of Na on CIS formation', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 2426-2429. Zhang S. B., Wei S. H. and Zunger A. (1997), 'Stabilization of ternary via ordered arrays of defect pairs', Phys. Rev. Lett. 78,4059^1062. Zhang S. B., Wei S. H., Zunger A. and Katayama-Yoshiba H. (1998), 'Defect physics of the CuInSe2 chalcopyrite semiconductor', Phys. Rev B. 57, 9642-9656. Zweibel K. (1995), 'Thin films: past, present, future', Prog. Photovoltaic Res. Appl. 3, 279-293.

CHAPTER 8

SUPER-HIGH EFFICIENCY III-V TANDEM AND MULTIJUNCTION CELLS MASAFUMI YAMAGUCHI Toyota Technological Institute, Nagoya 468-8511, Japan masafumi @ toyota-ti. ac.jp

The reasonable man adapts himself to the world: the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man. George Bernard Shaw, Man and Superman, 1903.

8.1 Introduction The development of high-performance solar cells offers a promising pathway toward achieving high power per unit cost for many applications. Substantial increases in conversion efficiency can be realised by multijunction solar cells in comparison with single-junction cells. The principles of multijunction cells were suggested in 1955 (Jackson, 1955) and multijunction solar cells have been investigated since 1960 (Wolf, 1960), as shown in Table 8.1. However, no significant progress was made in multijunction cell conversion efficiency during the period 1960-75 because of poor thin-film technologies. Based on progress in technologies for liquid-phase epitaxy, vapour-phase epitaxy and optical devices such as semiconductor lasers, AlGaAs/GaAs tandem cells, including tunnel junctions (Hutchby et al., 1985) and metal interconnections (Ludowise et al., 1982; Flores, 1983; Chung et al., 1989), were developed during the 1980s. At that time, the predicted efficiency of close to 30% was not obtained because of difficulties in making high performance and stable tunnel junctions, and oxygen-related defects in the AlGaAs materials (Ando et al., 1987). High performance and stable tunnel junctions with a double-hetero (DH) structure (the GaAs tunnel junction is sandwiched between AlGaAs layers) were proposed by Sugiura et al. (1988) of the Nippon Telegraph and Telephone Corporation (NTT) of Japan. Use of InGaP material for the top cell was proposed by Olson et al. (1990) of the National Renewable Energy Laboratory (NREL) of the United States, and as a result, a GalnP/GaAs tandem cell with an efficiency of 29.5% but small area of 0.25 cm2 was reported (Bertness et al.,

347

348

M. Yamaguchi

1994). More recently, monolithically grown InGaP/GaAs two-junction solar cells have reached a highest efficiency of 30.3% (Takamoto et al., 1997a) at 1 Sun AMI.5. For concentrator systems, over 30% efficiency has been attained by GaAs/GaSb mechanically stacked cells (Fraas et al., 1990). Table 8.1

Progress of III-V tandem solar cell technologies

Date

Milestone

Location

1955 1960 1982 1982

Jackson Wolf MIT RTI Varian NTT

1994 1996 1997

Principle of multijunction solar cells Proposal of multijunction solar cells Optimal design for high efficiency 15.1% AlGaAs/GaAs tandem cell 15.7% AlGaAs/GaAs tandem cell 20.2% AlGaAs/GaAs tandem cell (double heterostructure tunnel junction) 27.6% AlGaAs/GaAs tandem cell (metal interconnector) 34.2% GaAs/GaAs tandem cell (mechanically stacked xlOO concentrator 27.3% InGaP/GaAs tandem cell 29.5% InGaP/GaAs tandem cell 30.3% InGaP/GaAs tandem cell 33.3% InGaP/GaAs/InGaAs mechanically stacked cell

1997

First satellite using InGaP/GaAs cells

1987 1989 1990

Varian Boeing NREL NREL J-Energy J-Energy Sumitomo Hughes Spectrolab

Figure 8.1 shows the chronological improvement in AM1.5 efficiencies of two-junction solar cells under 1 Sun and concentrator operation. As a result of improvements in device structures and epitaxial technologies, based on understanding the physics and materials science of multijunction cells, remarkable improvements in the efficiencies of two-junction cells have been obtained. This chapter describes the principles of multijunction solar cells, candidate materials, epitaxial technologies, multijunction cell configurations and cell interconnection technologies. Recent progress with laboratory cells is also reviewed and predictions for the future of multijunction cells are discussed.

Super-High Efficiency III-V

Tandem and Multijunction

349

Cells

40

10-

1980

1985

1990

1995

2000

Year

Figure 8.1 Chronological improvements in AM1.5efficienciesoftwo-junctionsolarcellsunder 1 Sunand concentrator operation.

8.2 Principles of super-high efficiency multijunction solar cells 8.2.1 Conversion efficiency analysis While single-junction cells may be capable of attaining AM 1.5 efficiencies of up to 27%, multijunction structures appear capable of realising efficiencies of up to 35% (Hutchby et al., 1985). As shown in Fig. 8.2, solar cells with different band gaps are stacked in tandem so that the cell facing the Sun has the largest band gap. This top cell absorbs all the photons at and above its band-gap energy and transmits the less energetic photons to the cells below. The next cell in the stack absorbs all the transmitted photons with energies equal to or greater than its band-gap energy, and transmits the rest downward in the stack, etc. As shown in Fig. 8.3 as one example, the predicted spectral response for an AlGaAs/GaAs/InGaAs monolithic, two-terminal three-junction cell (MacMillan etal., 1989) suggests a wideband photoresponse for multijunction cells. In principle, any number of cells can be used in tandem. Computer analysis of the performance of multijunction solar cells has been carried out by several researchers (Loferski, 1976; Lamorte and Abbott, 1980; Mitchell, 1981; Fan et al., 1982; Nell and Barnett, 1987; Amano et al., 1989). The following explanations are based on the results of Fan et al. (1982). Figure 8.4 shows AMI and AM0 iso-efficiency plots for a two-cell, two-terminal tandem structure at 27 C and 1 Sun. The maximum theoretical efficiencies for this system are 36.2% at AMI and 32.4% at AM0. For optimal efficiencies in the two-terminal structure, the allowable range of band gaps for the top and bottom cells is very narrow. For both AMI and AM0,

350

M.

Yamaguchi

o A

? "

1

Cell 1

i\ I

Cell 2

u,

••jy.

Cells

Ugl > Ug2 > Ug3

Figure 8.2

Spectrum-splitting scheme for achieving high conversion efficiency using tandem cells. Wavelength (nm) 1400 1000 800

600

500

0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0

U(eV) Figure 8.3 Predicted spectral response for an AlGaAs/GaAs/InGaAs monolithic, two-terminal three-junction cell (MacMillan el id., 1989).

the top cell must have a band gap of about 1.75 eV, and the bottom cell about 1.1 eV. The four-terminal structure has an advantage over the two-terminal structure in that the photocurrents of the two elements do not have to match and a wide range of band-gap energies is allowable for the top and bottom cells. Figure 8.5 shows the AMI iso-efficiency curve at 27 C for tandem structures where the two cells have separate electrical connections. The maximum calculated efficiencies for this system are 36.6% at AMI and 32.9% at AM0, slightly higher than those of the two-terminal structure.

Super-High Efficiency III-V

> %

Tandem and Multijunction

1.50

//

1.25

E

Cells

351

TWO TERMINALS 27 C AM1

1.00

0.75

1.50

1.75

2.00

2.25

2.50

Top-cell band gap (eV)

1.50 TWO TERMINALS 27 C AMO 1.25

1.00

0.75 1.50

1.75

2.00

2.25

2.50

Top-cell band gap (eV)

Figure 8.4 Calculated AMI and AMO iso-efficiency plots for the two-cell, two-terminal tandem structure at 27 C and 1 Sun (Fan el at., 1982).

Although the optimum combination of 1.8 eV and 1.1 eV is very close to the optimal selection of 1.75 eV and 1.1 eV for the two-terminal tandem structure, there is a wide selection of acceptable band-gap energies. Figure 8.6 shows the AMI iso-efficiency plots at 27 C for the three-cell tandem structure with the cells connected in series. The curves are plotted for a bottom-cell band gap of 1.0 eV, because the maximum calculated efficiencies are obtained for a range of 0.95-1.0 eV. The optimal top/middle/bottom cell band-gap combination is 1.95/1.45/1.0 eV for the series-connected structure, and 2.15/1.55/1.0 eV for the separately connected one. The maximum theoretical efficiency at AMI is 41.1% for the series-connected structure and 42.5% for the separately connected structure. Computer analysis for two-junction solar cells under concentrator operation has also been carried out by some researchers (Mitchell, 1981; Wanlass et al., 1990). Figure 8.7 shows the AM 1.5 iso-efficiency curve for series-connected two-junction concentrator (500 Suns) tandem cells at 50 C (Wanlass et al., 1990). The optimum band-gap combinations and the maximum AM 1.5 efficiencies are 1.54/0.94 eV and

352

M.

Yamaguchi

1.50

> 1.25

1

1.00 AM1 Four terminals 27 C 0.75

1.50

1.75

2.00

2-25

2.50

Top-cell band gap (eV)

Figure 8.5 Calculated AM 1 iso-efficiency curve at 27 C and 1 Sun for tandem structures where the two cells are separately connected (Fan el at., 1982).

2.00Three cells Bottom-cell band gap = 1.00 eV Series-connected AM1

>

27 C

1.75-

1.50

1.25

1.75

2.00

2.25

2.50

Top-cell band gap (eV)

Figure 8.6 Calculated AM 1 iso-efficiency plots at 27 C and 1 Sun for the three-cell tandem structure with the cell connected in series. The bottom cell has a fixed energy gap of 1.0 eV (Fan el al., 1982). 4 2 . 0 % , and

1.42/0.72 eV and o v e r 4 1 % for t h e series-connected structure, and

1.63/0.94 eV and 42.1%, and 1.43/0.72 eV and over 41% for the separately connected one. Those for the two-junction cells at 80 C are 1.54/0.94 eV and 39.4% for the series-connected structure, and 1.64/0.94 eV and 39.7% for the separately connected one. The optimum band-gap combinations and the maximum AMO efficiency for the

Super-High Efficiency III-V

Tandem and Multijunction

Cells

353

two-junction cells at 100 Suns and 80 C are 1.70/1.04 eV and 35.0% for the series-connected structure, and 1.81/1.02 eV and 35.3% for the separately connected one. 2.0 i i • i i i i i i i i i i i i i i i i • i • • • i i i ' i i i i i i i i i i • • i i i •

Bottom-cell band gap (eV) Figure 8.7 Computed AM 1.5 iso-efficiency curve for series-connected two-junction concentrator tandem cells at 500 Suns and 50 C (Wanlass et al., 1990).

8.2.2

Cost

analysis

The allowable cost per unit area of solar cell modules depends strongly on module efficiency (Bowler and Wolf, 1980; Yamaguchi et al., 1994). Figure 8.8 shows the calculated tradeoff between the allowable cost of solar cells in dollars per square metre or dollars per peak watt and cell module efficiency for flat-plate modules (Bowler and Wolf, 1980). For example, a 30% efficient cell costing 3.5 times as much as a 15% efficient cell of the same area will yield equivalent overall photovoltaic system costs. Therefore high-efficiency cells will have a substantial economic advantage over low-efficiency cells, if the cost of fabricating such cells is low enough. For space applications, high-efficiency cells also have significant payload advantages. The use of concentrating systems can further enhance the cost advantage of high-efficiency cells. Figure 8.9 shows cost estimates for monolithically integrated two-junction solar cells with a III-V compound solar cell and a Si cell, fabricated by metal-organic chemical, vapour deposition (MOCVD), molecular beam epitaxy (MBE)

M. Yamaguchi

354

or chemical beam epitaxy (CBE) (Yamaguchi et al., 1994). As a result of source material cost reduction combined with the use of the tandem structure cell and a concentrator system, high-efficiency cells at a cost of less than $1/Wp should be possible.

10 12 14 16 18 20 22 24 26 28 30 AM1 module efficiency (%) Figure 8.8 Calculated curves relating the tradeoff between the allowable cost of solar cells in dollars per square metre or dollars per peak watt and cell module efficiency for flat-plate modules (Bowler and Wolf, 1980).

40 xu.1

• Factory O Equipment

30

• Labour BOthers im Dopant

20

0 V source

I o

S V source

xO.1

• Substrate

10

—

Present

Mass production

MOCVD (future)

1

—

MBE (future) CBE (future)

CBE + concentrator

Figure 8.9 Cost estimates for monolithically integrated two-junction solar cells consisting of a III-V compound solar cell combined with a Si cell, fabricated by metal-organic chemical vapour deposition (MOCVD), molecular beam epitaxy (MBE) or chemical beam epitaxy (CBE) (Yamaguchi el al., 1994).

Super-High Efficiency 1II-V Tandem and Mullijunciion

355

Cells

8.3 Candidate materials for multijunction cells and their present status 8.3.1 Lattice-matched systems The lattice parameter of the materials used to make the top and bottom cells must be closely matched in order to minimise misfit dislocations in the materials. In order to realise high performance, it is therefore important to consider the relation between the band gap and lattice constant shown in Fig. 8.10. The ideal band gap combination of band-gap energies of 1.7 and 1.1 eV yields a maximum AM 1.5 conversion efficiency of 36%. However, there does not exist within the III-V compounds a material that is lattice-matched to an inexpensive, commercially available bottom cell substrate material with a band gap of 1.1 eV. As the next possible band-gap combinations, 1.9/1.4 eV yields a maximum theoretical AM 1.5 efficiency of 34% at 1 Sun and 1.4 eV/0.73 eV yields 41% efficiency under 500 Suns AM 1.5. 2.5 2.0

«. 1.5

Q.
S m

1.0 0.5

'5.2

5.4

5.6 5.8 6.0 6.2 Lallice parameter (A)

6.4

6.6

Figure 8.10 Band-gap energies and lattice constants for III—V compounds and elemental semiconductors. AlGaAs/GaAs

two-junction

cell

T h e Alo.4Gao.6As/GaAs 2-junction cell, which h a s a

theoretical AM 1.5 efficiency of 34%, was favoured in the past (Bedair et al., 1981; Amano et al., 1987; Chung et al., 1989) because the device fabrication technologies for this system were fairly well developed. A 1.93 eV AlGaAs/GaAs metal-interconnected, two-terminal two-junction cell with an efficiency of 27.6% under 1 Sun AM1.5G and 23% at 1 Sun AM0 (Chung et al., 1989) has been reported. More recently, an AlGaAs/GaAs two-terminal two-junction cell with a tunnel-junction interconnector and an efficiency of 28.8% under 1 Sun AM1.5G has been demonstrated (Takahashi et al., 1998). However, there are problems with the growth of high-quality, oxygen-free AlGaAs and the fabrication of a high-conductance cell interconnect.

356

M. Yamaguchi

InGaP/GaAs two-junction cell The tandem combination of an Irio .5Gao.5P top cell with a direct band gap of ~ 1.9 eV and a GaAs bottom cell with a direct band gap of 1.43 eV has a theoretical efficiency of 34 % (Fig. 8.4). As in the case of InGaP/GaAs, the lattice parameters and current of these cells can both be matched. In addition, the absence of oxygen-related defects has enabled the development of high-efficiency InGaP/GaAs two-terminal, two-junction tunnel junction-connected cells with AM 1.5 efficiencies of 29.5% (Bertness et al., 1994) and later 30.3% (Takamoto et al., 1997a). Figures 8.11 and 8.12 show the structure and the I-V curve of the 4 cm2 InGaP/GaAs tandem solar cell with an efficiency of 30.3% under AM1.5G illumination. This high efficiency was achieved by increasing the InGaP top cell efficiency by improvements in the epitaxial growth process and introduction of a DH-structure InGaP tunnel junction with AllnP barriers. The InGaP/GaAs cells also have great potential for space applications, and cells with an AMO efficiency of 26.9% show superior radiation-resistant properties in comparison with single-junction GaAs or Si cells (Yamaguchi et al., 1997).

n"-lnGaP:0.050 um

2.0x10" cm - * (Si doped)

p-lnGaP: 0.550 j i m

1.5x10" cm" 3 (Zn doped)

p'-lnGaP:0.030 nm

2.0x10'" cm" 3 (Zn doped)

p'-AllnP: 0.030 nm <1.0x10" cm" 3 (Zn doped) p*-lnGaP:0.015jim 8.0x10 cm (Zn doped) rMnGaP:0.015um 1.0x10"cnV* (Si doped)

InGaP top cell

J InGaP tunnel junction —I

n'-AllnP: 0.050 urn

1.0x10" cm" 3 (Si doped)

n*-GaAs:0.100um

2 . 0 x 1 0 " cm" 3 (Si doped)

1

p-GaAs: 3.000 urn

1.0x10" cm" 3 (Zn doped)

GaAs bottom cell

p*-lnGaP:0.100|im

2.0x10"cm"

3

p*-GaAs: 0.300 urn

7.0x10'" cm" 3 (Zn doped)

p*-GaAs substrate

< 1 x 1 0 " c m " 3 (Zn doped)

(Zndoped)

Au

Figure 8.11 Structure of a 4 cm 2 InGaP/GaAs tandem solar cell, with an InGaP tunnel junction and AllnP barrier layers, with an efficiency of 30.3% (Takamoto el al., 1997a).

Super-High Efficiency I1I-V

Tandem andMultijunction

357

Cells

Cell area: 4 c m 2 AM1.5 global lOOmW/cm 2 25.3°C Isc

56.88mA

Voc

2.488V

FF

85.6%

Eff

30.28%

VOLTAGE (V) Figure 8.12 Current-voltage and power- voltage characteristics of a high efficiency InGaP/GaAs tandem cell under AMI.5 global illumination measured at JQA (Japan Quality Assurance Organization). Source: Takamoto etal. (1997a).

InP/lnGaAs two-junction cell The combination of InP with a band-gap energy of 1.35 eV and Ino.53 Gao^As with a band-gap energy of 0.75 eV is a lattice-matched system and has a theoretical conversion efficiency of 37% under 500 Suns AM1.5G and 33% under 500 Suns AM0 at 80 C (Wanlass et al., 1989). This system has also great potential for space applications, because InP cells are more radiation-resistant than other cells such as Si, GaAs cells and so forth (Yamaguchi et al., 1984). 31.8 % has been attained with an InP/InGaAs metal-interconnected three-terminal, monolithic two-junction cell under 50 Suns AM1.5 at 25 C (Wanlass et al, 1991). GaAs/Ge two-junction cell The combination of GaAs with a band-gap energy of 1.43 eV and Ge with a band-gap energy of 0.73 eV is also a lattice-matched system and yields 41% efficiency under 500 Suns AM1.5 at 50 C. A GaAs/Ge two-junction cell with efficiencies of 24.1% at 1 Sun AM1.5 and 21.3% at 1 Sun AM0 was fabricated by MOCVD (Vernon et al., 1989). The p-n junction in the Ge bottom cell was formed by in-diffusion of gallium and arsenic into the Ge from the growing GaAs cell structure during the MOCVD run and the cell interconnect was formed by a highly doped GaAs-Ge heteroj unction.

358

M. Yamaguchi

InGaP/GaAs/Ge three-junction cell The three-junction combination of an Ino.5Gao.5P top cell with a direct band gap of about 1.9 eV, a GaAs middle cell with a direct band gap of 1.43 eV and a Ge bottom cell with a band gap of 0.73 eV is lattice-matched and has a theoretical efficiency of 38% under 1 Sun AM1.5 and 42% under 400 Suns AM1.5 at 25 C (MacMillan et al., 1989). An InGaP/GaAs/Ge two-terminal threejunction cell using a tunnel junction with an efficiency of 26.5% under 1 Sun AMO has been demonstrated (Chiang et al, 1994). 2.0 eV/1.43 eV/1.05 eV/0.66 eV monolithic 4-junction cell The 4-junction combination of an InGaAlP (Ug = 2.0 eV) top cell, a GaAs (Ug = 1.43 eV) second-layer cell, a material third-layer cell with a band gap of 1.05 eV made of, for example, InGaAsP, and a Ge (Ug = 0.66 eV) bottom cell is lattice-matched and has a theoretical 1 Sun AMO efficiency of 42.3%. Although this system is ideal for maximum theoretical efficiency, the selection of the third-layer cell material and improvement in the material quality are problems still to be overcome.

8.3.2 Lattice-mismatched systems The attractive feature of lattice-mismatched materials is the availability of semiconductor alloys with precisely the band-gap energies needed to obtain the maximum efficiencies for multijunction solar cells. However, one of the major problems in making high-efficiency multijunction cells using lattice-mismatched materials is preventing the generation of dislocations and other defects that act as recombination centres and reduce device performance. Figure 8.13 shows the calculated dislocation density dependence of the conversion efficiency of single-junction and two-junction cells in comparison with experimental results (Yamaguchi and Amano, 1985). It is evident that high-efficiency two-junction cells with efficiencies of >30% should be possible using lattice-mismatched materials if the dislocation density can be reduced to <5 x 105 cm-2. The mechanically stacked configuration shows promise for realising high efficiency because it reduces dislocation problems in lattice-mismatched systems. The best results to date achieve a dislocation density of 2 x 106 cm-2 in GaAs grown on Si (Yamaguchi, 1991).

Super-High

Efficiency II1-V

Tandem and Multijunction

Cells

359

Grain size (u.m) 1

0.1

0.01

50 % c 0)

1

40

30

o .2 20 i2

Concentrator " \ tandem cells

n

^Tandem cells

-* Single-junction cells ^ s „ x

I «>
o

"'I

10 3

10 4

'

I

J-_l.4-IJI]l!

10 5

10 6

I ' ' 'Hill

10 7

I

10 8

Dislocation density (cm - 2 ) Figure 8.13 Calculated dislocation density and grain size dependence (solid lines) of the conversion efficiency of single-junction and two-junction cells under 1 Sun and concentrator conditions, compared with experimental results (crosses, triangles and circle). Source: Yamaguchi and Amano (1985).

AlGaAs/Si monolithic two-junction cell An attractive combination of materials for obtaining high-efficiency multijunction cells is either AlxGai_xAs or GaAsi_yPy with x~0.3, y~0.27 (band-gap energy -1.75 eV) for a top cell, together with Si as a bottom cell. In addition to an ideal combination of band gaps, the Si substrate and cell technologies are very well developed and quite inexpensive. An Alo.15Gao.85As/Si two-junction cell with an AM0 efficiency of 21.2% active area (the area excluding bus and fingers) fabricated on a Si substrate using MOCVD has been reported (Umeno et al., 1998). The major problem for improving cell efficiency is reducing the high density of dislocations, which is currently 2 x 107 cm"2 in the AlGaAs layer on Si. InGaP/InGaAs monolithic cell An InGaP (Ug = 1.7 eV) top cell and an InGaAs (Ug =1.1 eV) bottom cell on a GaAs substrate is also an ideal combination of band gaps for the maximum theoretical 1 Sun AM0 efficiency of 33.1%, but this is a lattice-mismatched system. Although there have been some approaches to InGaP/InGaAs monolithic two-junction cells, it is necessary to overcome the problem of dislocations induced by lattice mismatch in the interface between InGaP and InGaAs, and also that between InGaAs and GaAs. GaAs/CuInSe2 mechanically stacked cell An Alo.3Gao.7As/CuInSe2 cell also has an ideal combination of band gaps (1.75/1.1 eV) and its theoretical efficiency is 32.4% at AM0. Because, in addition, CuInSe2 has high radiation resistance, AlGaAs/CuInSe2

360

M. Yamaguchi

cells have potential for space applications. In a similar approach, a GaAs-CLEFT (Cleavage of Lateral Epitaxial Films for Transfer)/CuInSe2 mechanically stacked cell achieved an AMO efficiency of 23.31% (Gale etal., 1990). GaAs/Si mechanically stacked cell The GaAs/Si cell, while certainly not offering an ideal combination of band gaps, does yield a theoretical efficiency of 38% under 500 Suns AM 1.5 at 50 C. In addition, both the cells are very well developed and Si is quite inexpensive. A GaAs/Si mechanically stacked cell with an efficiency of 31.0% under 347 Suns AM1.5 has been reported by Gee and Virshup (1988). GaAs/GaSb mechanically stacked cell A GaAs/GaSb cell also possesses a near-ideal combination band gaps (1.43/0.73 eV) and has a theoretical efficiency of 41% under 500 Suns AM 1.5 at 50 C. A GaAs/GaSb mechanically stacked cell with efficiencies of 34.2% under 100 Suns AM1.5 and 32.2% under 100 Suns AMO has been reported by Fraasefa/. (1990). GaAs/lnGaAsP mechanically stacked cell A GaAs/InGaAsP (0.95 eV) cell and GaAs/InGaAs (0.73 eV) cell also have near-ideal band-gap combinations of 1.43/0.95 eV and 1.43/0.73 eV, and both have a theoretical efficiency of 41% under 500 Suns AM 1.5 at 50 C. A GaAs/InGaAs mechanically stacked cell with an efficiency of 30.2% under 350 Suns AM1.5 has been made by Gee and Virshup (1988). More recently, a GaAs/InGaAs mechanically stacked cell with an efficiency of 28.8% under 1 Sun AM 1.5 has been fabricated by Matsubara et al. (1998). Other two-junction cells Other approaches to two-junction cells with optimum band-gap combinations are the vacuum MOCVD-grown GaAso.77Po.23 (1.6 eV) /GaAso.84Sb0.i6(l.l eV) three-terminal, two-junction cell with an active area efficiency of 21.1% under 133-360 Suns AMI on a GaAs substrate (Fraas et al., 1984), and a liquid phase epitaxy (LPE)-grown AlGaAsSb (1.8 eV)/GaAsSb (1.2 eV) tunnel-junction interconnected, two-terminal cell with an AMO efficiency of 1% (Timmons etal., 1981). lnGaP/GaAs/lnGaAs(P) and AlGaAs/GaAs/InGaAs(P) mechanically-stacked threejunction cells A 1.93 eV-AlGaAsorInGaP/1.43 eV-GaAs/0.95 eV-InGaAs(P)cell has predicted efficiencies of 37.5% at 1 Sun AM1.5 and 46% at 400 Suns AM1.5 (MacMillan et al, 1989). An 1.92 eV-AlGaAs/GaAs/0.95 eV-InGaAsP (0.95 eV) two-terminal, three-junction cell with an AMO efficiency of 25.2% has also been

Super-High Efficiency III-V Tandem and Multijunction Cells

361

reported (Chung et al, 1991). This is a mechanically stacked cell composed of an AlGaAs/GaAs metal-interconnected two-junction cell and an InGaAsP bottom cell. More recently, a mechanically stacked three-junction cell consisting of a monolithically grown InGaP/GaAs two-junction cell and an InGaAs bottom cell has reached 33.3% at 1 Sun AM1.5 (Takamoto etal, 1997b). Figure 8.14 shows a schematic cross section of this cell, and Figure 8.15 shows the /-V curves of the components under AM 1.5 global illumination. InGaP/GaAs dual-junction cell

Figure 8.14 Schematic cross section of a mechanically stacked three-junction cell consisting of an InGaP/GaAs two-junction cell and an InGaAs bottom cell (Takamoto et al., 1997b). 30

•

I ' •

Total: 33.3%

• InGaAs (6.2%)

InGaP/GaAs (27.1%)

i . . . .

0.5

i .

1 1.5 Voltage (V)

'/.'.>

Figure 8.15 l-V curves of a mechanically stacked three-junction cell, consisting of an InGaP/GaAs two-junction cell and an InGaAs bottom cell, under 1 Sun AMI.5G illumination (Takamoto et al., 1997b).

M. Yamaguchi

362

AlGaAs (InGaP)/GaAs/InGaAsP/InGaAs four-junction cell Beaumont et al. (1990) have proposed a mechanically stacked four-junction cell consisting of an AlGaAs/GaAs two-junction cell on a GaAs substrate and an InGaAsP/InGaAs two-junction cell on an InP substrate, and Sharps et al. (1997) a mechanically stacked four-junction cell consisting of an InGaP/GaAs two-junction cell on a GaAs substrate and an InGaAsP/InGaAs two-junction cell, also on an InP substrate. The maximum theoretical efficiencies for a 1.91/1.43/1.05/0.75 eV four-junction cell are 35.1% at 1 Sun AMO, 39.3% at 1 Sun AMI and 44.5% at 500 Suns AMI. Figure 8.16 shows a cross section of a proposed InGaP/GaAs/InGaAsP/InGaAs four-junction solar cell, with a projected efficiency of 34.8% at 1 Sun AMO (Sharps et al, 1997). AR coaling V

I n*-GaAs

n-AI!nP n-GalnP p-GalnP

• contact metal • cap layer window layer emitter layer base layer back surface field

High band-gap tandem

• tunnel junction n-GaAs p-GaAs p-GalnP

window layer emitter layer base layer back surface field ' substrate

p-GaAs

n-lnP n-GalnAsP p-GalnAsP

• metal-bond interconnect window layer emitter layer base layer back surface field tunnel junction

Low band-gap landem

n-lnP n-GalnAs p-GalnAs

- window layer - emitter layer - base layer back surface field

p-lnP p-lnP

Figure 8.16

-substrate -contact metal

Schematic cross section of a proposed InGaP/GaAs/InGaAsP/InGaAs four-junction solar cell

(Sharps era/., 1997).

Super-High Efficiency III-V Tandem and Multifunction Cells

363

8.4 Epitaxial technologies for growing III-V compound cells Figure 8.17 shows chronological improvements in the efficiencies of GaAs solar cells fabricated by the LPE, MOCVD and MBE methods. LPE was used to fabricate GaAs solar cells in 1972 because it produces high quality epitaxial film and has a simple growth system. However, it is not as useful for devices that involve multilayers because of the difficulty of control over layer thickness, doping, composition and speed of 30

& |

20

|

15

|

10

1970

1975

1980

1985

1990

1995

Year Figure 8.17 Chronological improvements in the efficiencies of GaAs solar cells fabricated by the LPE, MOCVD and MBE methods.

throughput. Since 1977, MOCVD has been used to fabricate large-area GaAs solar cells because it is capable of large-scale large-area production and has good reproducibility and controllability. Using large MOCVD systems (for example, AIXTRON AIX-3000 or EMCORE Enterprise 400) which can simultaneously process up to 25 wafers, each of 4 inch diameter, two-junction InGaP/GaAs cells and three-junction InGaP/GaAs/Ge cells are now commercially produced by TECSTAR (Yeh et al., 1996) and Spectrolab (Chiang et al., 1996). In the research stage, InGaP/GaAs two-junction solar cells with efficiencies of 30.3% at 1 Sun AMI.5 and 26.9% at 1 Sun AM0 have been fabricated using the MOCVD method (Takamoto et al., 1997), while an efficiency of 21.1% at 1 Sun AM0 has been reported for MBE-grown InGaP/GaAs two-junction cells (Lammasniemi etal., 1997) and efficiencies of 27.5% at 140 Suns AM1.5 and 24.6% at 100 Suns AM0 have been reported for LPE-grown AlGaAs/GaAs two-junction cells (Andreevefa/., 1997).

364

M. Yamaguchi

Table 8.2 compares the advantages and disadvantages of the various epitaxial technologies. While the LPE method can produce high-quality epitaxial films, the MOCVD method is effective for large-scale large-area production of solar cells. MBE and CBE are advantageous for realising novel multilayer structures such as multijunction solar cells because they provide excellent controllability of monolayer abruptness and thickness due to the nature of beam (Yamaguchi et al., 1994). However, there have been few reports of CBE-grown solar cells. Table 8.2

Advantages and disadvantages of epitaxial technologies

Characteristics

LPE

MOCVD

MBE

CBE

Quality MQW Abrupt interface Heavy doping Large-area Throughput Efficient use of Source materials Equipment cost

**** * * ** * *** ***

*** *** *** #* **** **** **

** **** **** *** ** ** **

*** **** **** **** *** *** ####

#*#

**

**

*

**** Excellent, *** very good, ** fairly good, * bad.

8.5 Monolithic vs. multi-terminal connection modes Figure 8.18 shows various configurations of two-junction cells. For example, in the case of two-junction cells, two cells can be connected to form either two-terminal, three-terminal or four-terminal devices. In a monolithic, two-terminal device, the cells are connected in series with an optically transparent tunnel junction intercell electrical connection. In a two-terminal structure, only one external circuit load is needed, but the photocurrents in the two cells must be equal for optimal operation. Key issues for maximum efficiency monolithic cascade cells (two-terminal multijunction cells series connected with tunnel junction) are the formation of tunnel junctions of high performance and stability for cell interconnection, and growth of optimum band-gap top-and bottom-cell structures on lattice-mismatched substrates, without permitting propagation of deleterious misfit and thermal stress-induced dislocations.

Super-High Efficiency Ill-V Tandem and Multifunction Cells

metal interconnect

Z=E P

365

c ^///////////////W/. o — "I

n

n

-o -

P' n

r 2-terminal cell

3-terminal cell

4-terminal cell

Figure 8.18 Schematic diagrams of various configurations of two-junction cells. In contrast, in three-and four-terminal cells the photocurrents do not have to be equal. However, because it is very difficult to connect three-terminal devices in series, three-terminal tandem cells do not appear to be viable. In the four-terminal case, two separate external circuit loads are used. Since the two individual cells are not coupled, the photocurrents do not have to be the same. Consequently, a much larger selection of band-gap energy combinations is possible, and the changes in photocurrents with changing solar spectral distributions do not pose serious limits. This approach avoids the problem of lattice-mismatched epitaxial growth, current matching and the internal electrical connection of the two-terminal device. Important issues for obtaining high efficiency mechanically stacked cells are the development of multijunction cell fabrication techniques such as thinning the top cell, bonding the bottom cell to the top cell, and cell connections

8.6 Cell interconnection One of the most important factors in making high-efficiency monolithic-cascade type multijunction cells is to achieve optically and electrically low-loss interconnection of two or more cells. There are two main approaches to providing low-resistance intercell ohmic contacts: degenerate doping (tunnel-junction interconnection) and localised metallisation (metal interconnection). The use of a degenerately doped p*/n* tunnel junction is attractive because it only involves one extra step in the growth process. To minimise optical absorption, formation of thin, wide-band-gap tunnel junctions is

366

M.

Yamaguchi

10000

f

0

1

1000

< >. —

100

-

10

-

ID

ti

-o r

A

CBE

A MOCVD

^ v calculated

0

MBE

1

P

*o

0.1

-

0.01

-

&

..Y:

S P c Q.

-1

H

< > \

0.001 0.0001 0.5

i

.

.

.

.

i

. . . . < >

1.5

l/„(eV) Figure 8.19 Changes in tunnel peak current densities of III—V compound tunnel diodes fabricated by various epitaxial growth methods as a function of band-gap energy of lll-V materials (Yamaguchi el al., 1994).

E o

top cell

P"-GaAs /i"-GaAs "•AI,Ga,.,As n-GaAs sub bottom cell

13

500

600 700 800 Annealing temperature (C)

Figure 8.20 Dependence of tunnel peak current density on annealing temperature for double hetero (DH) structure GaAs tunnel diodes; x indicates the Al mole fraction of the sandwichingAl.tGai_rAs layers.

Super-High Efficiency III-V Tandem and Multifunction Cells

367

necessary. However, the formation of a wide-band-gap tunnel junction is very difficult, because the tunnelling current decreases exponentially with increase in band-gap energy, as shown in Fig. 8.19. In addition, impurity diffusion into a highly doped tunnel junction during overgrowth of the top cell increases the resistivity of the tunnel junction and degrades the top cell performance. This was a severe problem in the past, but it has been reduced by the use of lower growth temperatures, as shown in Fig. 8.20, the advent of new dopants including carbon and the introduction of the double-hetero (DH) structure (Sugiura et al., 1988).

Grid Cap

Top metal 0.50fim

n'-GaAs

0.14nm

n'-AlxGa.xAs

0.15um

n • AI0.37G ao.63As

3.80nm

p

O.Uym

-AlosvGaosiAs

p'-AUGai.xAs

jfo/$0$0%&0fr

0.12nm

n*-AlxGai.xAs

&##:••••*"*»*»*> n*-Alo»Gao.iAs

Window _Ernjtter_ Upper ce.i bottom bottom metal metal Lower cell \ \ top metal V /T\ I f~\ S/^^J

Base Dd&e / J

StopetcrT

Slop etch Lower cetl-top-cohlct Window

0.15(im

n-GaAs

Emitter

3.60|un

p

-QaAs

Base

Figure 8.21 Cross section of an AlGaAs/GaAs tandem cell, showing details of the metal interconnecter (MacMillan et al., 1989).

368

M. Yamaguchi

The second approach uses a metallisation scheme to connect the top and bottom cells in the metal-interconnected cascade cell (MacMillan et al., 1989), as shown in Fig. 8.21. The grooves are formed using a sequence of wet chemical etches to remove the epitaxial layers selectively. The metal-interconnect approach has the problem of a complex fabrication process and difficulty in obtaining a low-resistance, reliable contact to the top cell materials. However, this approach may be effective when formation of a tunnel-junction interconnection is difficult. For mechanically stacked structures, adhesive bonding is used to connect the cells. Adhesive materials must be optically transparent over the wide wavelength range 350-1700 nm, have a high thermal conductivity and be mechanically resistant.

8.7 Possible applications of multijunction cells Concentrator operation of two-junction and three-junction cells fabricated on inexpensive substrates such as Ge, Si and polycrystalline materials are being considered as a way to achieve high-efficiency and low-cost cells. In Japan, the super-high efficiency solar cell R&D project including multijunction cells started in fiscal year 1990 (Yamaguchi and Wakamatsu, 1996). The objective of the project is to reach conversion efficiencies of about twice the 1990 values at the laboratory level by the year 2000 and production of such cells for terrestrial applications by 2010. As markets for direct-to-home broadcast, mobile telephone and data communications are growing, commercial satellite power requirements have increased by 200-400% during the early 1990s, and this increasing demand requires continuous efforts to improve solar cell performance and reduce solar cell array cost. In September 1995, the US Air Force Joint Wright Lab./Phillips Lab./NASA Lewis Multijunction Solar Cell Manufacturing Technology (the so-called Man Tech) Program for the development and fabrication of large-area InGaP/GaAs/Ge two-junction and threejunction cells started (Keener et al., 1997). This aims to improve InGaP/GaAs/Ge cell performance (average efficiency > 24-26%) and scale up to production size, quantity and yield while limiting the production cost per watt to not more than 15% over GaAs cells. The average efficiencies of InGaP/GaAs-on-Ge two-junction cells and InGaP/GaAs/Ge three-junction cells made to date were 22.4% and 24.2% at AM0, respectively. OANAMSAT5, the first satellite powered by InGaP/GaAs two-junction cells on Ge substrates, was launched in August 1997 into geosynchronous orbit and is operating nominally with 10 kW of multijunction power (Brown et al, 1997). The average two-junction cell efficiency of this array is 21.6%.

Super-High Efficiency II1-V Tandem and Multifunction Cells

369

8.8 Predictions For super-high-efficiency cells to come into wider use, it will be necessary to improve their conversion efficiency and reduce their cost. In this section, the possibility of obtaining efficiencies of over 40% by using multijunction cell structures and thin-film technologies on inexpensive substrates such as Si and polycrystalline materials is discussed. Figure 8.22 shows the theoretical and realistically expected conversion efficiencies of single-junction and multijunction solar cells reported in the past by some researchers (Fan et a/., 1982; Wanlass et al., 1989; Kurtz et al., 1997) compared with experimentally realised efficiencies. Clearly, concentrator three-junction solar cells have great potential for realising efficiencies of over 40%. 55 50

5 •

>

theoretical values

45

•

0

\ 40 a % 35

r

O

realistically expected values

lf>

$

30

experimental values

lb 1

1.5

2 2.5 3 3.5 Number of junctions

4

4.5

Figure 8.22 Theoretical and realistically expected conversion efficiencies of single-junction and multijunction solar cells reported by Fan et al. (1982), Wanlass et al. (1989) and Kurtz et al. (1997). compared with experimental values.

So far, we have focused mainly on improvement of conversion efficiency. To reduce costs, the use of cheaper substrates is necessary. Figure 8.13 shows the calculated grain size d and dislocation density iVd dependencies of the AM 1.5 conversion efficiency of GaAs single-junction cells, two-junction cells and concentrator two-junction cells compared with experimental values. The calculations were carried out using the following expressions for minority-carrier diffusion length L as a function of d and NA (Yamaguchi and Itoh, 1986): 1/L2 = 1/Lo2 + ASIDd UL2 = l/Lo2 + x3Nd/4

370

M. Yamaguchi

where L is the minority-carrier diffusion length in the solar cell active layers, Lo is the radiative-recombination-limited value of L, S is the surface recombination velocity at the edge of the grain boundary depletion region (assumed to be 5 x 106 cm s_l in the case of GaAs), and D is the minority-carrier diffusion coefficient. It follows that concentrator thin-film multijunction solar cells fabricated on inexpensive substrates such as Si and polycrystalline materials have great potential for realising efficiencies of more than 35% at low cost if one can reduce the dislocation density to less than 5 x 105 cm"2 and increase the grain size to more than 0.1 cm. Cost reduction of III-V compound solar cells is also necessary for their widespread application. To this end, cell fabrication using inexpensive substrates such as Si and Ge, large-scale epitaxial growth equipment and concentrator systems are needed. In addition, an increase in conversion efficiency reduces the cell cost per Wp directly. Figure 8.23 shows an analysis of the energy cost of a 50 MW concentrator system (Whisnant et al., 1994). This suggests that tandem solar cells on Ge substrates under concentrator operation with efficiencies as high as 35% are promising for cost reduction.

T

H

-

-

II GaAs 1-junclion 25%/27.5%/30% 'A

Si point contact 27.4%

!l

i

sc-GaAs sc-Ge pc-Ge

I

GaAs 2-junctio i 30% / 32.5% / 3 5%

sc-GaAs sc-Ge pc-Ge

Si

s c : single-crystalline p c : poly-crystalline

Figure 8.23

An analysis of the energy cos! of a 50 MW concentrator system (Whisnant et al., 1994).

Super-High Efficiency III-V Tandem and Multifunction Cells

371

Acknowledgment This work was partly supported by the New Energy and Industrial Technology Development Organization as part of the New Sunshine Program under the Ministry of International Trade and Industry, Japan.

References Amano C , Sugiura H., Yamamoto A. and Yamaguchi M. (1987), '20.2% efficiency Alo.4Gao.6As/GaAs tandem solar cells grown by molecular beam epitaxy', Appl. Phys. Lett. 51, 1998-2000. Amano C , Sugiura H., Yamaguchi M. and Hane K. (1989), 'Fabrication and numerical analysis of AlGaAs/GaAs tandem solar cells with tunnel interconnections', IEEE Trans. Electron Devices ED-36, 1026-1035. Ando K., Amano C., Sugiura H., Yamaguchi M. and Saletes A. (1987), 'Nonradiative e-h recombination characteristics of mid-gap electron trap in AljGai., As (x = 0.4) grown by molecular beam epitaxy', Jpn. J. Appl. Phys. 26, L266-L269. Andreev V. M., Khvostikov V. P., Rumyantsev V. D., Paleeva V. E. and Shvarts M. Z. (1997), 'Monolithic two-junction AlGaAs/GaAs solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 927-930 Beaumont B., Garabedian P., Nataf G., Guillaume J.-C, Gibart P. and Verie C. (1990), 'Mechanically stacked two-tandem consolar cell concept', Conf. Record 21st. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 47-52. Bedair S. M., Hutchby J. A., Chiang J. P. C , Simons M. and Hauser J. R. (1981), 'AlGaAs/GaAs high efficiency cascade solar cells', Conf. Record 15th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 21-26. Bertness K. A., Kurtz S. R., Friedman D. J., Kibbler A. E., Kramer C. and Olson J. M. (1994), '29.5%-Efficiency GalnP/GaAs tandem solar cells', Appl. Phys. Lett. 65, 989-991. Bowler D. L. and Wolf M. (1980), 'Interactions of efficiency and material requirements for terrestrial silicon solar cells', IEEE Trans. Components, Hybrids Manufacturing Technol., CHMT-3, 464-472. Brown M. R., Goldhammer L. J., Goodelle G. S., Lortz C. U., Perron J. N., Powe J. S. and Schwartz J. A. (1997), 'Characterization testing of dual junction GaInP2/ GaAs/Ge solar cell assemblies', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 805-810.

372

M. Yamaguchi

Chiang P. K., Kurt D. D., Cavicchi B. T., Bertness K. A., Kurtz S. R. and Olson J. M. (1994), 'Large-area GaInP2/GaAs/Ge multijunction solar cells for space applications', Proc. 1st. World Conf. Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 2120-2123. Chiang P. K., Ermer J. H., Nishikawa W. T., Krut D. D., Joslin D. E., Eldredge J. W., Cavicchi B. T. and Olson J. M. (1996), 'Experimental results of GaInP2/GaAs/Ge triple junction cell development for space power systems', Conf. Record 25th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 183— 186. Chung B.-C, Virshup G. F., Hikido S. and Kaminar N. R. (1989), '27.6% efficiency (1 Sun, air mass 1.5) monolithic Alo.37Gao.63 As/GaAs two-junction cascade solar cell with prismatic cover glass', Appl. Phys. Lett. 55, 1741-1743. Chung B.-C, Virshup G. F., Klausmeier-Brown M., Ristow M. L. and Wanlass M. W. (1991), '25.2%-efficiency (1 Sun, air mass 0) AlGaAs/GaAs/InGaAsP threejunction, two-terminal solar cell', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 54-57. Fan J. C C , Tsaur B. Y. and Palm B. J. (1982), 'Optical design of high-efficiency tandem cells', Conf. Record 16th. Photovoltaic Specialists Conf, San Diego, IEEE Press, Piscataway, 692-701. Flores C. (1983), 'A three-terminal double junction GaAs/GaAlAs cascade solar cells', IEEE Electron Device Lett. EDL-4, 96-99. Fraas L. M , McLeod P. S., Cape J. A. and Partain L. D. (1984), 'Monolithic two-color, three-terminal GaAsP/GaAsSb solar cells', Conf. Record 17th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 734-739. Fraas L. M., Avery J. E., Martin J., Sundaram V. S., Girard G., Dinh V. T., Davenport T. M., Yerkes J. W. and O'Neill M. J. (1990), 'Over 35-percent efficient GaAs/GaSb tandem solar cells', IEEE Trans. Electron Devices 37, 443-449. Gale R. P., McClelland R. W., Dingle B. D., Gormley J. V., Burgess R. M., Kim N. P., Mickelsen R. A. and Stanbery B. J. (1990), 'High-efficiency GaAs/CuInSe2 thin-film tandem solar cells', Conf. Record 21st. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 53-57. Gee J. M. and Virshup G. F. (1988), 'A 31%-efficient GaAs/silicon mechanically stacked, multijunction concentrator solar cell', Conf. Record 20th. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 754-758. Hutchby J. A., Markunas R. J., Timmons M. L., Chiang P. K. and Bedair S. M. (1985), 'A review of multijunction concentrator solar cells', Conf. Record 18th. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 20-27.

Super-High Efficiency III-V Tandem and Multijunction Cells

373

Jackson E. D. (1955), 'Areas for improving of the semiconductor solar energy converter', Trans. Conf. on the Use of Solar Energy 5, University of Arizona Press, Tucson (1958), 122-126. Keener D. N., Marvin D. C , Brinker D. J., Curtis B. H. and Price P. M. (1997), 'Progress toward technology transition of GaInP2/GaAs/Ge multijunction solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 787-792. Kurtz S. R., Myers D. and Olson J. M. (1997), 'Projected performance of three-and four-junction device using GaAs and GalnP', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 875-878. Lammasniemi J., Kazantsev A. B., Jaakkola R., Toivonen M., Jalonen M., Aho R. and Pessa M. (1997), 'GalnP/GaAs cascade solar cells grown by molecular beam epitaxy', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 823-826. Lamorte M. F. and Abbott D. H. (1980), 'Computer modeling of a two-junction monolithic cascade solar cell', IEEE Trans. Electron Devices ED-25, 831-840. Loferski J. J. (1976), 'Tandem photovoltaic solar cells and increased energy conversion efficiency', Conf. Record 12th. IEEE Photovoltaic Specialists Conf, Baton Rouge, IEEE Press, Piscataway, 957-961. Ludowise M. J., LaRue R. A., Borden P. G., Gregory P. E. and Dietz W. T. (1982), 'High-efficiency organometallic vapor phase epitaxy AlGaAs/GaAs monolithic cascade solar cell using metal interconnects', Appl. Phys. Lett. 41, 550-552. MacMillan H. F., Chung B.-C, Hamaker H. C , Kaminar N. R., Kuryla M. S, Ladle Ristow M., Liu D. D., Partain L. D., Schultz J. C , Virshup G. F. and Werthen J. G. (1989), 'Recent advances in multijunction III-V solar cell development', Solar Cells 27, 205-217. Mitchell K. W. (1981), 'High efficiency concentrator cells', Conf. Record 15th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 142-146. Matsubara H., Tanabe T., Moto A., Mine Y. and Takagishi S. (1998), 'Over 27% efficiency GaAs/InGaAs mechanically stacked solar cells', Solar Energy Mat. Solar Cells 50, 177-184. Nell M. E. and Barnett A. M. (1987), 'The spectral p-n junction model for tandem solar-cell design, IEEE Trans. Electron Devices ED-34, 257-266. Olson J. M., Kurtz S. R. and Kibbler A. E. (1990), 'A 27.3% efficient Gao.5Ino.5P/GaAs tandem solar cell', Appl. Phys. Lett. 56, 623-625.

374

M. Yamaguchi

Sharps P. R., Timmons M. L., Hills J. S. and Gray J. L. (1997), 'Wafer bonding for use in mechanically stacked multi-band-gap cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 895-898. Sugiura H., Amano C , Yamamoto A. and Yamaguchi M. (1988), 'Double heterostructure GaAs tunnel junction for AlGaAs/GaAs tandem solar cells', Jpn. J. Appl. Phys. 27, 269-272. Takahashi K., Yamada S., Unno T. and Kuma S. (1998), 'Characteristics of GaAs solar cells on Ge substrate with a preliminary grown thin layer AlGaAs', Solar Energy Mat. Solar Cells 50, 169-176. Takamoto T., Ikeda E., Kurita H. and Ohmori M. (1997a), 'Over 30% efficient InGaP/GaAs tandem solar cells', Appl. Phys. Lett. 70, 381-383. Takamoto T., Ikeda E., Agui T., Kurita H., Tanabe T., Tanaka S., Matsubara H., Mine Y., Takagishi S. and Yamaguchi M. (1997b), 'InGaP/GaAs and InGaAs mechanically stacked triple-junction solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 1031-1034. Timmons M. L. and Bedair S. M. (1981), 'AlGaAsSb/GaAsSb cascade solar cells', Conf. Record 15th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 1289-1293. Umeno M., Soga T., Baskar K. and Jimbo T. (1998), 'Heteroepitaxial technologies on Si for high-efficiency solar cells', Solar Energy Mat. Solar Cells 50, 203-212. Vernon S. M., Tobin S. P., Wojtczuk S. J., Keavney C. J., Bajgar C , Sanfacon M. M. Daly J. T. and Dixon T. M. (1989), 'III-V solar cell research at Spire Corporation', Solar Cells 27, 107-120. Wanlass M. W., Emery K. A., Gessert T. A., Horner G. S., Osterwald C. R. and Coutts T. J. (1989), 'Practical considerations in tandem cell modeling', Solar Cells 27, 191-204. Wanlass M. W., Coutts T. J., Ward J. S., Emery K. A., Gessert T. A. and Osterwald C. R. (1991), 'Advance high efficiency concentrator tandem solar cells', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 38^15. Whisnant R. A., Hutchby J. A., Timmons M. I., Venkatasubramanian R. and Hills J. S. (1994), 'Silicon and GaAs/Ge concentrator power plants: a comparison of cost of energy produced', Proc. 1st. World Conf. Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1103-1106. Wolf M. (1960), 'Limitations and possibilities for improvement of photovoltaic solar energy converters', Proc. Inst. Radio Engineers 48, 1246-1263.

Super-High Efficiency III-V Tandem and Multifunction Cells

375

Yamaguchi M. (1991), 'Dislocation density reduction in heteroepitaxial III-V compound film on Si for optical devices', J. Mat. Res. 6, 376-384. Yamaguchi M., Uemura C. and Yamamoto A. (1984), 'Radiation damage in InP single crystals and solar cells', J. Appl. Phys. 55, 1429-1436. Yamaguchi M. and Amano C. (1985), 'Efficiency calculations of thin film GaAs solar cells on Si substrates', /. Appl. Phys. 58, 3601-3606. Yamaguchi M. and Itoh Y. (1986), 'Efficiency considerations for polycrystalline GaAs thin-film solar cells', J. Appl. Phys. 60, 413^117. Yamaguchi M., Warabisako T. and Sugiura H. (1994), 'CBE as a breakthrough technology for PV solar energy applications', J. Crystal Growth 136, 29-36. Yamaguchi M. and Wakamatsu S. (1996), 'Super-high efficiency solar cell R&D program in Japan', Conf. Record 25th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 9-11. Yamaguchi M., Okuda T., Taylor S. J., Takamoto T., Ikeda E. and Kurita H. (1997), 'Superior radiation-resistant properties of InGaP/GaAs tandem solar cells', Appl. Phys. Lett. 70, 1566-1568. Yeh Y. C. M., Chu C. L., Krogen J., Ho F. F., Datum G. C , Billets S., Olson J. M. and Timmons M. L. (1996), 'Production experience with large-area, dual-junction space cells', Conf. Record 25th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 187-190.

CHAPTER 9

ORGANIC PHOTOVOLTAIC DEVICES JONATHAN J. M. HALLS and RICHARD H. FRIEND Cavendish Laboratory, Cambridge, CB3 OHE, UK jjmh I @cam. ac. uk

"I just want to say one word to you —just one word. Plastics. There's a great future in plastics. Think about it. " Mr Maguire to Ben Braddock in The Graduate, 1967.

9.1 Introduction Despite much effort, semiconductor photovoltaic devices made with traditional inorganic semiconductors have remained sufficiently expensive that their uses are confined to a number of niches. Much effort is currently directed towards the use of thin-film semiconductors, in place of silicon wafers, since the direct fabrication of thin devices on substrates offers the prospect of lower manufacturing costs, particularly for larger area applications. The development of amorphous silicon solar cells in 1976 by Wronski and Carlson had the potential of making photovoltaic cells cheaper to produce, and other techniques have been developed to make larger devices possible, including polycrystalline silicon, cadmium telluride, and copper indium diselenide, as described elsewhere in this volume. Despite these advances, the cost of fabricating photovoltaic cells remains prohibitively high for many applications, particularly when large areas are required. One of the factors that keeps system costs relatively high for these technologies is the requirement for high-temperature processing of the semiconductor in a high vacuum environment. This largely restricts fabrication to batch processing onto glass substrates, with associated costs.

9.1.1 Molecular semiconductor devices An alternative approach is the use of organic, molecular semiconductors, which can be processed over large areas at relatively low temperatures, either by vacuum sublimation of molecular materials, or, preferably, by processing from solution of 377

378

J. J. M. Halls and R. II. Friend

film-forming materials such as polymers. If the many issues that we discuss in this chapter concerning the photovoltaic performance and stability can be satisfactorily resolved, then there is the prospect of considerably lower manufacturing costs. The reduction will result in part from the low cost of the small volume of the thin active semiconductor layers, but more importantly, from the lower cost of the other materials used, such as substrates, and the reduced costs that can be realised, for example, by roll-to-roll manufacturing. The challenges in developing organic semiconductors for use in photovoltaic applications arc considerable, requiring new materials, new methods of manufacture, new device architectures and new substrate and encapsulation materials. The most realistic approach is to make use of available know-how that has been developed in related technologies, and it is our view that this is necessary here. Molecular semiconductors are in fact widely used-they are the dominant technology for xerographic copying and laser printing, as we discuss in Section 9.2. More relevant to photovoltaic devices is the development more recently of molecular semiconductor light-emitting diodes (LEDs). These devices are manufactured on a transparent substrate {e.g. glass) as a layer of molecular semiconductor sandwiched between a transparent bottom electrode (e.g. indium tin oxide) and a top metallic electrode. Figure 9.1 shows such a structure.

\ > organic layer Photovoltaic mode

metal

LED mode

Figure 9.1 Schematic diagram of a molecular semiconductor diode which, according to the selection of electrodes and semiconductor layers, can iunction as a light-emitting diode or as a photovoltaic diode. Fabrication is by successive deposition of bottom transparent electrode (e.g. indium tin oxide) onto the transparent substrate (e.g. glass), the semiconductor layer or layers (by vacuum sublimation and/or solution processing) and top metal electode (by vacuum deposition).

Organic semiconductor LEDs, or OLEDs, have advanced very rapidly over the past five years, and now provide a full range of colour, high efficiency (of order 10% quantum efficiency), and, very importantly, have been engineered to give good shelf life and operational lifetime (10,000 hours operation is a minimum requirement for

Organic Photovoltaic Devices

379

most display applications). Work on sublimed molecular devices originating from the Kodak group of Tang in the mid 1980s (Tang and VanSlyke, 1987) has been developed by a number of companies including Pioneer, Japan, who have developed LED displays for automobile audio systems and for mobile electronics. The use of semiconductor polymers for LEDs was developed in Cambridge, UK (Burroughes et al, 1990). The state-of-the-art polymer LEDs are very efficient and are also being commercialised, for example by Philips, Eindhoven (Friend et al, 1999). Solution processing of polymers is particularly attractive for large-area coating, as will be required for photovoltaic devices. Organic transistors networked to form small integrated circuits are being developed by Philips, and are expected to be used in smart cards and electronic barcode labels within the next few years. Organic photodiodes are further from commercial exploitation than these other applications, although there is considerable interest in using these novel semiconductors to fabricate cheaper photovoltaic panels and photodetector arrays. However, using a combination of new materials and novel device structures the efficiency of organic photovoltaic cells continues to rise, and their comparatively low fabrication cost makes these cells increasingly attractive. The general structure of the diode as used for LEDs, shown in Figure 9.1, is directly transferable to operation in a photovoltaic mode (though the electrodes and semiconductor layers need to be correctly designed). There is therefore scope for the direct transfer of know-how from OLEDs to the manufacture of practical, durable and efficient photovoltaic devices. This know-how includes materials synthesis and purification, electrode manufacturing, semiconductor layer deposition and encapsulation.

9.1.2 Photovoltaic properties of molecular semiconductors Molecular materials show semiconducting properties when constructed so that the carbon atoms present in the molecule or polymer chain are bonded as sp2 + pz hybrid orbitals. The pz orbitals form delocalised n and 71* molecular orbitals, which are conventionally recognised as the alternation of carbon-carbon 'single' and 'double' bonds in the molecule. A range of such materials is shown in Figure 9.2. The semiconducting properties of these materials have been very extensively investigated over many decades; we review in more detail the properties relevant to photovoltaic properties in later sections, and refer the reader to a number of monographs (Borsenberger and Weiss, 1993; Greenham and Friend, 1995; Pope and Swenberg, 1999). However, we can summarise the salient characteristics briefly here.

380

J. J. M. Halls and R. H. Friend

molecules metalloporphyrin

C&H13

polyacetylene

MEH-PPV

Figure 9.2 Chemical structures of a range of organic semiconductors. Perylene derivatives are used extensively as electron acceptors and charge-transport layers for xerography. Porphyrins can be made with a range of metal ions (M) at their centres; magnesium, copper and zinc are common choices.

Organic Photovoltaic

381

Devices

Photoabsorption in these materials creates an excited state which is generally confined to a molecule or a region of a polymer chain. This localised excitation is generally termed an 'exciton'. This can be considered either as a neutral excited state of a molecule, or, using a semiconductor description, as an electron-hole pair, bound together by Coulomb and lattice interactions. The electron-hole binding is usually very strong, of order 0.5 eV or above, so that at room temperature {kT = 25 meV), there is little likelihood of electron-hole separation. These materials are therefore commonly strongly luminescent, with emission resulting from radiative decay of the exciton. Electron-hole separation is clearly required for photovoltaic operation, and can be achieved by a number of extrinsic processes. The most important of these is the use of a heterojunction formed between two molecular semiconductors (which can be deposited one on top of the other). The two semiconductors must be chosen so that one can act as electron acceptor and the other as hole acceptor, as is shown schematically in Figure 9.3. Charge separation (often termed photo-induced charge transfer) requires that the offsets in the energies for hole states (n valence band) and for electron states (n* conduction band) at the heterojunction exceeds the binding energy of the electron-hole pair when present on one or other molecular semiconductor (Halls et ai, 1999). This approach has been developed over several years, and is found to be effective both with molecular structure (Tang, 1986), using perylene/phthalocyanine heterojunctions, and with polymer devices (Halls et ai, 1999), using for example MEH-PPV and CN-PPV (see Fig. 9.2).

vacuum level glass ITO interpenetrating polymer network LUMO

acceptor Exciton

donor (a)

acceptor

HOMO

(b)

JHole

Electron

energy acceptor

donor

Figure 9.3 (a) Schematic diagram of photoinduced charge transfer at the interface between two semiconductors with different ionisation potentials and electron affinities, (b) Schematic showing how a mixture of electron- and hole-accepting polymers can be used to provide heterojunctions distributed throughout the polymer composite layer.

382

J. J. M. Halls andR. H. Friend

This double-layer structure separates charge, which can be collected at the two electrodes without difficulty in the strong internal electric field present for these thin devices (Note that optical absorption depths are less than 1 /an for most molecular semiconductors.) The structure functions well, provided that light is absorbed sufficiently close to the heterojunction that the excitons produced can diffuse to the heterojunction in order to ionise. Unfortunately, typical diffusion ranges are an order of magnitude lower than the thickness required to absorb the incident light, so that only about 10% of incident light can be captured in such device arrangements (Halls etal, 1996). Much of the current interest in organic photovoltaic devices is therefore directed to finding new architectures which allow all absorbed light to produce excitons which do reach heterojunctions. One approach is the use of phase-separated polymer blends, which provide a 'distributed heterojunction' throughout the layer thickness (Halls et ai, 1995a; Yu et ai, 1995). Quantum efficiencies up to nearly 30% are achieved in this way. The structure of such a device is shown schematically in Fig. 9.3b.

9.1.3 Overview of this chapter In Section 9.2 the development of organic photovoltaic cells is put into historical context. In Section 9.3 we shall consider why certain organic molecules and semiconductors behave like semiconductors, and look at some of their characteristic properties. Section 9.4 covers the development of simple molecular and polymeric photovoltaic cells based on metal-semiconductor-metal sandwich structures. In Section 9.5 the physics that underlies the charge separation and charge transport properties is discussed, and in Section 9.6 the photocurrent action spectra and currentvoltage characteristics are interpreted in the context of these phenomena. Section 9.7 introduces techniques to improve the performances of these simple calls, beginning with the fabrication of heterojunctions. In Section 9.8 the use of dispersed heterojunctions is introduced; in these the donor and acceptor materials are scrambled together, using, for example, phase separated polymer blends. In Section 9.9 the use of diffuse interface heterojunctions is considered, in which the surface are is increased by intermixing over a limited part of the semiconductor layer, as may be achieved by lamination. In Section 9.10 we look at the technological benefits and drawbacks of these new devices, and speculate on future uses of what promises to be a low-cost avenue to the production of large area photovoltaic cells. Section 9.11 brings the chapter to a close with some general conclusions.

Organic Photovoltaic Devices

383

9.2 Background—early work on photoresponsive organic semiconductors Molecular semiconductors were incorporated into light-sensitive electronic devices at an early stage in their development; indeed, the solid-state photovoltaic effect was first observed in a cell using selenium. Until the middle of this century, molecular, and primarily organic, materials provided the basis of research for the photovoltaic effect. It was not until 1954, when Pearson, Chapin and Fuller invented the silicon photocell at Bell Laboratories, that inorganic materials were destined to become the material of choice for commercial applications. The first report of solid-state photoconductivity is that by Smith (1873) who observed the phenomenon in selenium. The first detailed study of the subject was carried out in the 1920s by Gudden, Pohl and co-workers, on diamond, ZnS and alkali halide single crystals (Borsenberger and Weiss, 1993). The phenomenon was originally interpreted as a radiation-induced structural effect. It was not until the full understanding of the Hall effect that photoconductivity was attributed to the creation of free electrons by the absorption of light. Anthracene was the first molecular material in which photoconductivity was observed, in work by Pochettino (1906) and Volmer (1913). Covalently bonded solids formed the basis for investigations in the 1940s and 1950s, when research into organic materials was limited by the need for single-crystal samples. Interest in organic photoconductors was renewed by the discovery that common artificial pigments and dyes, such as malachite green and methylene blue, had semiconducting properties (Bube, 1960). The photovoltaic phenomenon was observed in cells containing thin films of these organic pigments and dyes in amorphous, crystalline and microcrystalline phases (Merritt, 1982). At the same time it was realised that many biological compounds, and their synthetic analogues, had photoconductive properties. These included carotenes, chlorophylls and other porphyrins, phthalocyanines, cyanines, merocyanines and porphyrins, many of which are important in biological systems. Most of the understanding of the photovoltaic effect in organic photocells comes from the study of devices fabricated from these molecular materials. In recent years semiconducting polymers have been applied to organic photovoltaic cells. Their electronic properties are, in the main, very similar to those of the smaller molecules described above, but their physical properties tend to make them easier to process. Much of the interest in molecular semiconductors was driven by the search for organic photoconductors for xerographic applications in laser printers and photocopiers. In these applications an image is projected onto a statically charged photoconductive drum, and the drum is discharged wherever the drum is exposed. Toner is picked up by the areas of the drum that remain charged, and is subsequently

384

J. J. M. Halls and R. H. Friend

transferred to the paper (Borsenberger and Weiss, 1993). A wide range of organic materials has been investigated for this application, many of which have been exploited in organic photovoltaic cells. Present day xerographic devices almost universally use organic molecular photoconductive materials, dispersed in a polymer binder, rather than more 'traditional' selenium-based photoconductors. In a photoconductive material, charge carriers are created by the optical absorption. An externally applied field is required to produce current by extracting these photogenerated charges from the photoconductor before they recombine. The photovoltaic effect is an extension of photoconductivity in which this field is 'builtin' to the system and exists in the dark. In a photovoltaic cell, this field typically arises from the interaction ofp- and w-type semiconductors (such as in silicon, GaAs, CdTe and CuInSe2), or (less commonly in the case of inorganic semiconductor devices) from the interaction of the photoconductor with the metal. Photovoltaic materials are necessarily photoconductors, but the converse is not always true.

9.3 Conjugated molecules: a new class of semiconductors 9.3.1 Introduction Until recently, carbon-based molecules and polymers have been considered to be insulating materials, and as such have been exploited as electrical insulators in numerous applications. Although it was known from the turn of the twentieth century that certain organic materials were photoconductive, it was arguably the extensive development of molecular electronic materials such as anthracene (Fig. 9.2) by Pope and Swenberg (1999) and the subsequent discovery in 1974 that doped polyacetylene, the simplest conjugated polymer, can exhibit metallic levels of conductivities (Chiang et al, 1977), that initiated an exciting and rapidly expanding field of research into these materials. The novel electronic properties of both molecular and polymeric semiconductors arises from their conjugated chemical structure, and on a molecular level the physical processes behind their properties can be dealt with in the same way. In their undoped state, molecular semiconductors are generally medium to wide bandgap semiconductors. Conjugated polymers have the additional processibility advantages of engineering plastics. Carbon has the electronic structure Is2 2s22p2, and forms hybrid orbitals with its four valence electrons (2s12p2). In conjugated materials, which have alternating double and single bonds in their canonical structures, three sp2 hybrid orbitals form covalent bonds: one with each of the carbon atoms either side of it, and the third with

Organic Photovoltaic

Devices

385

a hydrogen atom or other group. The remaining electron occupies a pz orbital. Collectively, thepz orbitals overlap to create delocalised % bonds which, in the case of a polymer, extend along the full length of the backbone. The most stable configuration of a conjugated polymer is planar, since this maximises the overlap of the pz orbitals (Fig. 9.4), and so these materials tend to be rigid, insoluble and intractable. This originally made it difficult to produce samples in a useful form; fabrication techniques were limited to gas-phase polymerisation and electrochemical growth.

Figure 9.4 Schematic diagram of (a) trans-polyacetylene and (b) poly(p-phenylenevinylene) showing the pz orbitals which overlap to provide the extended delocalised n-system.

The 7i-bonds are weaker than the strong covalent bonds formed by the sp2 electrons, and the electrons in the delocalised 71 system therefore have a smaller binding energy. These electrons dominate the electronic and optical properties, whereas the .^-derived bonds maintain the physical structure of the molecule when electrons are excited from the bonding % orbital to the anti-bonding 71* orbital. The development by Wessling of the sulphonium polyelectrolyte precursor route to poly(p-phenylenevinylene) (PPV) made available thin, high-quality conjugated polymer films (Wessling and Zimmerman, 1968; 1972). The precursor does not have a fully conjugated % system, and is therefore soluble in organic solvents. The final polymer is formed by thermal elimination of the alkyl sulphonium leaving group, which conjugates the links between the benzene rings, as illustrated in Fig. 9.2. Spincoating, a technique commonly used to deposit thin layers of photoresist onto silicon wafers, can therefore be used to deposit uniform thin films of PPV onto planar substrates. By substituting alkyl chains into the benzene rings of PPV, polymers can be synthesised which are soluble in common organic solvents, and can therefore be spin-cast directly. Poly(2-methoxy-5-(2'-ethyl-hexyloxy)-/>phenylenevinylene), MEH-PPV, was one of the first derivatives synthesised for this purpose (Wudl et al, 1991). By using electron withdrawing or donating substituents the electronic energy levels of these materials can be adjusted (Bredas and Heeger, 1994).

386

J. J. M. Halls and R. H. Friend

The physical properties of organic semiconductors depend to a large extent on the morphology of the bulk material, and the differences between small molecules and conjugated polymer-based devices are largely a result of this dependence. Much of the early work on small molecules was carried out on single-crystal samples, in which relatively high electronic mobilities can be obtained. Amorphous samples have a lower mobility, as charges must hop between adjacent molecules, a process that involves an activation energy. Despite this, amorphous films of molecular semiconductors have turned out to be advantageous for applications of organic lightemitting devices. Uniform amorphous films can be produced over large areas by vacuum sublimation. Small molecules have a tendency to crystallise, a process associated with premature device failure in organic molecular displays, and preventing this from occurring has been the focus of much recent research. Crystallisation can create pinholes in the organic films, and grain boundaries along which diffusion of impurities may occur. Bulk samples of conjugated polymers tend to be highly amorphous, although the morphology of a particular material depends critically on its chemical structure, and on the method of synthesis and film preparation. PPV provides a relevant example. Electron diffraction studies of PPV by Granier et al. (1986; 1989) revealed the presence of microcrystallites with a monoclinic unit cell containing two monomer units. Masse et al. (1990) demonstrated that the microcrystallites were on a typical length-scale of 50 A. A number of attempts have been made to increase the degree of molecular order, including stretch-alignment of heated polymer films (Briers et al., 1994) and the use of a precursor polymer consisting of rigid conjugated chain segments separated by flexible spacer groups (Halliday et al, 1993; Pichler et al., 1993). Optical measurements by McBranch et al. (1995) suggest that, in the case of thin spin-cast films made from soluble polymers, the molecular chains lie primarily in the plane of the film.

9.3.2 Electronic properties of conjugated molecules The novel electronic properties of conjugated molecules arise from the overlap of the pz orbitals. The interaction between these orbitals on two adjacent carbon atoms causes their degeneracy to split, and a pair of 7i-type molecular orbitals are formed, as illustrated in Fig. 9.5. In a polymer chain, several electrons contribute to the 7i system, and the bonding and anti-bonding orbitals become broad quasi-continuous energy bands, analogous with the conduction and valence bands of inorganic semiconductors. As the overlap between adjacent pz orbitals and the number of electrons participating

Organic Photovoltaic

387

Devices

in the Ji-system increases, the bands become wider, and the energy gap between them decreases. Thus larger molecules and longer polymer chains tend to have smaller band gaps. In the context of molecular semiconductors we shall define the band gap as the energy difference between the top of the valence band (the highest occupied molecular orbital, or HOMO) and the bottom of the conduction band (the lowest unoccupied molecular orbital, or LUMO). The band gap controls the optoelectronic properties of conjugated materials, and its value is typically in the range 1-4 eV. As a caveat, however, we note that the exact energy of the optical transition may differ from the band gap due to excitonic effects; we will discuss this later. anti-bonding

orbital

<

•<

v<

bonding orbital

Degenerate p 7 levels

(a) Non-degenerate molecular orbital levels

LUMO

«—HOMO

(b) Quasi-continuous energy bands resulting from overlap of many pz orbitals

Figure 9.5 Schematic diagram showing the energy levels of electrons in pz hybridised atomic levels, and subsequently in the n bonding and anti-bonding molecular orbitals when (a) two atoms are brought together to form a dimer, and (b) when a large number of atoms in a chain contribute to the delocalised n system.

Exciting an electron from a bonding orbital to an anti-bonding orbital is equivalent, in the band picture, to transferring an electron from the valence to the conduction band by supplying it with an energy greater than the band-gap energy. Electrons in the % system can therefore be electronically and vibrationally excited by absorption of light, or by addition or removal of charges by electric fields or chemical dopants. In a real polymer chain, the conjugated structure is unlikely to extend along its full length, as imperfections, defects, and conformational kinks interrupt the orbital overlap. Instead, there will be a series of chain segments, each of which is characterised by a different number of repeat units and has a different band gap. The longer the conjugated segment, the narrower the gap between the HOMO and LUMO energy levels. Estimates of the conjugation length come from comparisons of the optical and electronic properties of a particular polymer with those of a range of related oligomers with different, known, chain lengths (Woo et ai, 1993).

388

J. J. M. Halls and R. H. Friend

benzenoid form

quinoid form

benzenoid form

LUMO

^H^ u,

•H-

U;

u, u U„ 2

*k

HOMO bipolaron q = ~2

polaron q = -1

U:

± exciton q=0

u2

4 3 polaron

q=+1

u?

3

bipolaron q = +2

Figure 9.6 Schematic representation of a bond alternation defect on a segment of a chain of poly(/>phenylene). Below, the energy level diagrams of neutral excitons and positive and negative polarons and bipolarons are illustrated. The associated optical transitions are marked in red.

In the solid state, the dominance of intramolecular forces over intermolecular effects make these materials pseudo-one-dimensional semiconductors. This allows structural and electronic modification of conjugated segments when excited states are formed, allowing their energy levels to relax within the band gap. In most conjugated polymers the local geometrical rearrangement is towards a reversal of the sense of double/single bond alternation which raises the energy of that section of the chain. Such is the case in poly(p-phenylene) (PPP), in which the benzenoid form is preferred in the ground state (Friend et al, 1987). Excitation causes a rearrangement to the higher energy, quinoid form, as illustrated in Fig. 9.6. This depicts a doubly charged defect termed a bipolaron, which is generally considered to be the product of chemical doping. Also shown are the energy level diagrams of the neutral exciton, the positive and negative polarons and bipolarons, and their associated optical transitions.

9.3.3 Photoexcited states in conjugated materials Excited states in small conjugated molecules are known to have the form of bound electron-hole pairs, termed excitons. However, opinion is divided as to whether optical transitions in conjugated polymers are best described by an exciton or semi-

Organic Photovoltaic Devices

389

conductor band model. The anisotropy of inter- and intramolecular coupling in conjugated polymers indicates that these materials occupy the middle ground between small conjugated molecules and inorganic semiconductors, in which strong intersite coupling of the extended crystal lattice gives rise to uncorrelated electrons and holes. Measurements of photoconductivity (Lochner et al, 1978), electroreflectance (Sebastian and Weiser, 1981) and site-selective fluorescence (Pautmeier et al, 1990; Rughooputh et al, 1988) established the excitonic nature of the optical transitions of polydiacetylene. For PPV, in many ways the prototypical conjugated polymer, models of the primary products of photoexcitation range from weakly-bound excitons that are well delocalised along the polymer chains (Lee et al, 1993a), to moderately strongly bound intrachain excitons (binding energy -0.4 eV) (Bredas et al, 1996; Marks et al., 1994), to models with binding energies above 1 eV (Chandross et al, 1994). The band model was initially employed to explain the lowest %-n* transitions in conjugated polymers such as polyacetylene, poly(p-phenylenevinylene), poly(pphenylene), and polythiophene. In this approach, optical transitions are simulated by the single-chain one-electron Su-Schrieffer-Heeger (SSH) Hamiltonian, which ignores electron-electron interactions (Su et al, 1979). There is substantial evidence that optical transitions in conjugated polymers such as PPV are described by an excitonic model. Primary support comes from the site-selective fluorescence measurements of Rauscher et al (1990). For excitation above the localisation threshold, the emission spectrum is independent of energy, with a fixed Stokes shift from the absorption spectrum. Below this energy threshold, the energy of the peak in the emission spectrum decreases with excitation energy. These observations are consistent with a model in which excitons diffuse to regions of longer conjugation length before recombining radiatively. Excitons generated below the localisation threshold have insufficient energy to migrate to lower energy sites. Moreover, the rise of the photoconductivity in PPV at the absorption edge (Pichler et al, 1993) can be explained by a model in which excitons are subsequently dissociated by defects. (Antoniadis et al, 1994a; Marks et al, 1994). The dissociation of excitons at these sites to yield polarons and bipolarons explains the observations of photoinduced absorption experiments. The quenching of photoluminescence in conjugated polymer films subjected to an electric field has been used to argue that photoexcitation creates excitons, which are dissociated by the electric field (Kersting et al, 1994). The role of interchain interactions in solid films of these polymers has been considered, and Rothberg and co-workers have suggested that only a small fraction of photoexcited carriers in PPV are intrachain singlet excitons, and that approximately 90% are bound polaron pairs on adjacent chains that lead to negligible quantities of luminescence (Yan et al, 1994). However, the synthesis of conjugated polymers with

390

J. J. M. Halls and R. H. Friend

luminescence yields higher than 60% suggests that only a minority of the photoexcited species in these materials are in the form of non-emissive polaron pairs (Hwang etal, 1996). In the excitonic model, the electronic transitions are strongly coupled to the vibrational modes of the molecule, giving rise to vibronic structure in both the absorption and emission spectra. The photoexcitation of triplet excitons from the ground state is dipole-forbidden, and excitation creates vibrationally excited singlet excitons that relax by phonon emission to the lowest vibrational sub-level of the new electronic level within -100 fs. The emission spectra of most conjugated polymers approximately mirror their corresponding absorption spectra. The 0-0 bands do not coincide, but are separated by the Stokes shift, which arises from the structural relaxation of the exciton before it emits radiatively. There is extensive evidence for the diffusion of excitons between adjacent conjugated sites, in addition to that provided by site-selectivefluorescence.Transient photoluminescence experiments show a monotonic red shift of the emission with time for excitation above the localisation threshold (Hayes et al, 1995; Kersting et al, 1993; Samuel et al, 1993). Monte Carlo simulations of such measurements indicate that the migration is three-dimensional in nature (Rauscher et al, 1990). The migration of excitons to lower energy sites has the effect that emission occurs from a narrow distribution of conjugation lengths, whereas absorption probes the full range of lengths. This is consistent with the observation that emission spectra are generally narrower and richer in vibronic structure than absorption spectra. In conclusion, the excitonic model of photoexcitation will therefore be assumed to be the most appropriate in the conjugated polymers and molecules used in organic solar cells. In order for photogenerated charges to be collected in a solar cell these excitons must be dissociated. This is an important consideration in the design and development of efficient polymer photovoltaic cells, as we shall see later.

9.4 Basic organic photovoltaic cells 9.4.1 Introduction Before some specific examples of organic photovoltaic devices are discussed, the basic physics underlying the operation of such cells will be introduced. Most of the basic research into the photovoltaic effect in organic materials is carried out using simple sandwich-geometry cells, in which afilmof a single organic photoconductor is sandwiched between different planar contacts. Clearly, if the cell is to be exposed to

Organic Photovoltaic Devices

391

illumination at lcasl one of the contacts must be (semi-)transparent and this is usually achieved by using glass coated with indium tin oxide (ITO), or a thin metal layer. Such a cell is illustrated in Fig. 9.7. ITO is extensively used in the fabrication of liquid-crystal displays as a transparent conducting electrode. Polymer LEDs, which are structurally the same as simple organic photovoltaic cells, generally employ ITOcoated glass as the hole-injecting contact, through which the emitted light can escape. The transparency and rigidity of ITO-coated glass makes it an ideal substrate for the fabrication of polymer photovoltaic cells.

r\ ^-\

light )

T

/ /

substrate

glass quartz plastic

electrode 1 semitransparent

organic photoconductor

ITO metal

molecular polymer

electrode 2 opaque A;

fcMg

JL

Figure 9.7 Schematic cross-section of a sandwich-type organic photocell. The thickness of the organic layer is typically 0.01-1 fan.

Thin films of small molecules are generally deposited by vacuum sublimation. Conjugated polymer films may be deposited by spin-coating from solution, or alternatively deposited by blade coating. The active layers in most organic devices are a few tens or hundreds of nanometres thick. The top contact in a sandwich cell is generally deposited by thermal evaporation, or, less commonly, electron beam evaporation. In these processes, which are carried out under vacuum, the metal is heated either by a filament or crucible made from a high melting point metal, or by a focussed electron beam, to beyond its melting point. Metal atoms evaporated from the source are deposited onto the polymer surface. The contacts are patterned using a shadow mask placed in the path of the evaporated metal atoms. The different work functions of the two contacts that sandwich the organic photoconductor set up an electric field in the organic layer. This field is responsible for separating the charges and driving them to their appropriate electrodes, and for the open-circuit voltage measured across an illuminated cell. The field profile depends on the conductivity of the organic material. If the material is doped n- or p-type, charge redistribution will occur until the Fermi level is constant throughout the cell. This gives rise to a Schottky-type junction, illustrated in Fig. 9.8a, familiar in many doped inorganic semiconductor systems, and therefore not described in any detail here. Depletion is expected to occur at the lower work function contact if its work function is lower than that of the semiconductor. In contrast, if the material is highly insulating

392

J. J. M. Halls andR. H. Friend

with a very low dopant concentration, the depletion layer thickness exceeds the thickness of the semiconductor layer and the field is constant throughout the cell as is the case in most organic devices (Fig. 9.8b). Such devices are termed tunnel diodes, and are less familiar in conventional semiconductor devices. Under illumination, either excitons or charge carriers are generated. Providing the neutral species can be ionised and the charges separated by the built-in field before recombination occurs, the negative charges can drift to the low work function electrode (such as aluminium) and the positive charges can drift to the higher work function material (such as ITO). Energy

Vacuum level

ITO

Polymer

A

(a) Before contact

(b) High carrier concentration

(c) Low carrier concentration

Figure 9.8 Schematic representation of the energy levels in a polymer photocell (a) before the polymers and the two contacts (in this case ITO and aluminium, with work functions Onx> and A| respectively) are placed in contact, and (b and c) after contact, with no external bias. In case (b) the polymer has a extrinsic carrier concentration, and a Schottky barrier is formed at the polymer/Al interface. In contrast, in (c), the polymer is highly insulating and the depletion layer extends throughout the device.

9.4.2 Molecular photovoltaic cells Pochettino (1906) and Volmer (1913) were the first workers to observe photoconductivity in the organic solid anthracene, the structure of which is shown in Fig. 9.1. Anthracene is one of the most widely studied molecular electronic materials. Its crystal structure was determined accurately in the 1950s, and procedures for synthesising high-purity anthracene samples were refined in the following decade. Since the discovery of the effect, it was not until the 1950s that the photoconductivity of anthracene was investigated more fully (Borsenberger and Weiss, 1993). It was demonstrated in these studies that absorption of light leads to the generation of excitons, which migrate to the surface before dissociating to yield free charges at defect or impurity sites, including adsorbed oxygen atoms. Chaiken and Kearns

Organic Photovoltaic Devices

393

(1966) suggested that at high excitation energies (above 4.0 eV) the photocurrent arises from a direct ionisation process (i.e. a band-to-band transition) whereas the dissociation of singlet excitons was responsible for photogeneration at lower energies. As with much of the early research on organic photoconductors, research was directed towards finding materials suitable for reprography, rather than for solar cell applications. However, the photoelectrical requirements for reprographic materials are very similar to those for solar energy conversion, a fact not exploited fully by workers until the 1970s. The first classes of organic materials to be considered seriously as photovoltaic (rather than simply as photoconductive) materials were the porphyrins and phthalocyanines. Chlorophyll, the familiar green pigment responsible for the conversion of solar energy in plants, is a porphyrin, a class closely related in structure to the phthalocyanines. Kearns and Calvin (1958) demonstrated the photovoltaic effect in a cell based on magnesium phthalocyanine (referred to here as MgPh and shown in Fig. 9.2) with transparent conducting glass electrodes, and measured a photovoltage of 200 mV. Using copper phthalocyanine, Delacote et al. (1964) observed rectifying behaviour in asymmetric metal 1/CuPh/metal 2 cells, a necessary requirement for a photovoltaic cell. Federov and Benderskii (1971, 1971) fabricated an Al/MgPh/Ag cell and observed a rectification ratio of 103 at 1.5 V. They attributed the diode-like behaviour to the formation of a p-n junction by the diffusion of Al into the organic layer (the device had been heat-treated) where it complexes with the phthalocyanine, replacing the magnesium. They observed a photovoltaic effect which increased in magnitude as the cell was exposed to oxygen, a phenomenon associated with the dissociation of photogenerated excitons at oxygen sites. Ghosh et al. (1974) used cyclic voltammetry to confirm the formation of a Schottky barrier in an Al/MgPh/Cell, and determined photovoltaic characteristics in agreement with the extracted Schottky barrier parameters. Their results are illustrated in Fig. 9.9a. They obtained a photovoltaic efficiency of 0.01% under illumination at 690 nm, one of the highest values reported up to that time. The exciting prospect of emulating the highly evolved natural process of photosynthesis in an artificial solar cell inspired the use of chlorophyll-a (Chl-a) in organic photovoltaic cells. Early investigations of the photovoltaic effect in Chl-a in the late 1950s were not promising, and the low quantum yields obtained were attributed to trapping of charges. However, in 1970 Meilinov et al. measured a photocurrent quantum yield of 10% in an Al/Chl-a/Al cell. Work by Putseiko et al. suggested that the photoconductivity of Chl-a is enhanced in the presence of water, which was later shown by Katz and co-workers to give rise to the formation of a microcrystalline phase of a Chl-a-H20 adduct (see Tang and Albrecht (1975a) for a

394

J. J. M. Halls andR. H. Friend

I

500

600

700

800

Wavelength (nm)

900

400

500

600

700

800

Wavelength (nm)

Figure 9.9 The left hand graph (a) shows the absorption spectrum (circles) of magnesium phthalocyanine along with the photocurrent action spectrum (triangles) of an Al/MgPh/Ag photodiode. The cell was illuminated through the gold contact. The solid line through the triangles is a simulation of the photocurrent, based on the assumption that excitation of the MgPH close to a barrier region at the aluminium back contact contributes predominantly to the photocurrent. After Ghosh (1974). The right hand graphs show the absorption spectra (lower graph, c) of microcrystalline chlorophyll-a, as prepared (1) and after heat treatment (2). The upper graph (b) shows the photocurrent action spectrum of Al/Chl-a/Hg cells before (1) and after heat treatment (2). The cells were illuminated through the aluminium film. In this case the action spectra closely resemble the optical density of the film. Heat treatment causes a phase transformation to a disordered phase, and the photocurrent decreases by an order of magnitude. After Tang and Albrecht (1975a).

summary of this early work). Tang and Albrecht (1975a; 1975b) published extensive investigations into the photovoltaic effect of Chl-a in which they used a range of different metals in simple sandwich structures. They concluded that a Schottky barrier was formed in thep-type Chl-a in the vicinity of the lower work function contact, and found this region to be the active area for photogeneration. A maximum power conversion efficiency of 0.05% was obtained in a Cr/Chl-a/Hg cell under monochromatic illumination at 745 nm. Figures 9.9b and 9.9c show their results for an Al/Chl-a/Hg cell. Anthracene belongs to a class of materials termed polyacenes, which are made from different numbers of fused benzene rings. The photoconductive properties of a number of other polyacenes have been investigated. Naphthalene, which consists of two fused benzene rings, has a very low photoconductivity, with a quadratic intensity dependence, suggesting that photogeneration arises from exciton-exciton annihilation

Organic Photovoltaic Devices

395

(Braun and Dobbs, 1970). The photoconductivity of tetracene, with four rings, is considered to occur by diffusion of excitons to surface sites where ionisation leads to the trapping of the electron, as in anthracene (Borsenberger and Weiss, 1993). The photoconductivity of pentacene, which has five benzene rings, was investigated by Silinish et al. (1980) and found to be consistent with the electric-field-induced dissociation mechanism proposed by Onsager (1934). Perylene derivatives, which are widely used in xerography as electron acceptors and charge generation layers, have been incorporated in organic photovoltaic cells in both single- and double-layer structures (Tang, 1986; Hiramoto etal, 1992a; 1992b). Hiramoto and co-workers observed photocurrent amplification by a factor of 104 in a gold/perylene/gold sandwich-type cell, and found the phenomenon to be associated with electron injection from one of the electrodes to the perylene film through the depletion layer formed in the organic layer under a high electric field. (Hiramoto et al., 1994). Figure 9.2 shows the structures of some common perylene derivatives. The efficiency of organic molecular photovoltaic cells remained low until the mid 1980s, when various techniques were employed to increase their efficiency. The most significant advance was the discovery by Tang (1986) that the photocurrent in a molecular semiconductor could be increased by orders of magnitude if it was used in a double-layer (heterojunction) structure in tandem with a second, carefully chosen, organic semiconductor with different energy levels. This approach will be discussed in detail later in the chapter; in the meantime, simple conjugated polymer-based photovoltaic cells will be considered.

9.4.3 Polymer photovoltaic cells Conjugated polymers were applied to photocells at an early stage in the development of these organic semiconductors. Initial investigations exploited polyacetylene (Weinberger et al., 1982) and various polythiophenes (Glenis et al, 1986), but the results were not promising, with relatively low open-circuit voltages and efficiencies. Glenis et al. (1986) measured an external quantum yield of 0.17%, an open-circuit voltage of ~0.4 eV and a fill factor of 0.3 in a cell containing poly(3-methylthiophene). This sandwich-type device had electrodes of aluminium and platinum, and was illuminated with polychromatic light. In the same decade, Weinberger (1982) investigated the photovoltaic effect in polyacetylene with an aluminium top contact and a rear contact of graphite. This cell had an open-circuit voltage of 0.3 V and a charge collection efficiency of 0.3% at low light levels. These low open-circuit voltages were attributed to the formation of polarons, which bring the one-electron

396

J. J. M. Halls andR. H. Friend

hole and electron levels deep inside the energy gap, so that the energy difference between these states is considerably smaller than the 7i-7t* gap. Thus photogenerated charges relax energetically so that the potential difference between them is small, and the open-circuit voltage of the cell is limited. The observation of electroluminescence in PPV provided indirect evidence that this family of conjugated polymers might be more suitable for incorporation in polymer photovoltaic cells. Electroluminescence involves the capture of positive and negative polarons to form singlet excitons that decay radiatively. Luminescence in PPV is observed close to the onset of the n-n* absorption (Rauscher et al, 1990), so the free energy of the electron and hole polaron states cannot be much smaller than that of the singlet exciton. Simple sandwich-type photovoltaic cells, identical in structure to polymer LEDs, containing PPV layers were first investigated by Karg et al. (1993) and have since been the subject of extensive research by various groups. Karg fabricated an ITO/PPV/A1 cell by blade-coating the ITO-coated glass substrate with a polymer precursor and heating the resulting film to convert the precursor to fully conjugated PPV. The cell showed rectifying behaviour in the dark, consistent with a Schottky diode structure, and developed an open-circuit voltage of 1V under illumination. The power conversion efficiency of this cell was approximately 0.1% under white-light illumination. Impedance spectroscopy of the cell confirmed the Schottky-type barrier at the aluminium/PPV interface. Antoniadis and co-workers at Xerox (Antoniadis et al, 1994b) fabricated similar ITO/PPV/A1 devices, and reported open-circuit voltages of 1.2 V. The power conversion efficiency was 0.07% under illumination at 460 nm with an intensity of 1 mW cm"2, and at very low intensities the quantum collection efficiency was 5%. Antoniadis also attributed the behaviour of the cell to the formation of a Schottky barrier at the low work function contact, and characterised the depletion width and carrier density using cyclic voltammetry. PPV and many other polymers show p-type behaviour allowing the formation of depletion layers with low work function metals. In many polymers this doping is known to result from the incomplete removal of materials used during their synthesis. In the case of the polythiophenes the doping is considered to result from residual iron chloride, which is used as a catalyst during its synthesis (Einsiedel et al, 1998). In the case of PPV the p-type behaviour is most likely to result from doping of the polymer by atmospheric oxygen. Marks and co-workers (1994) also fabricated ITO/PPV/A1 cells, but considered their cells to be fully depleted, as the dark carrier density was found to be very low in their PPV, leading to a conductivity of less than 10~12 S cm"1 in the dark. The results were interpreted in terms of photoexcitation in a tunnel-diode structure, the rectifying characteristics arising from the tunnelling of carriers through the triangular barriers at

Organic Photovoltaic

Devices

397

the polymer/contact interfaces. Open-circuit voltages saturating at 1.2 V and lowintensity external quantum yields of order 1% were reported. Using calcium instead of aluminium as an electrode, open circuit-voltages approaching 1.7 V were obtained, owing to the larger work function. Yu et al. (1994a) invoked the same tunnel-diode model for cells which they fabricated using a soluble PPV derivative, poly(2methoxy-5-(2'-ethyl-hexyloxy)l,4-phenylenevinylene) (MEH-PPV). They reported external quantum yields of order 0.1% in short circuit, and up to 20% under an applied reverse bias of 10 V under monochromatic illumination. Photodiodes fabricated from PPV and its derivatives also function as light-emitting diodes when driven in forward bias, a fact highlighted in Yu's publication; the MEH-PPV cells operated as LEDs with an external quantum efficiency of 1%. The photocurrent action spectrum of an MEH-PPV device is shown in Fig. 9.10 (Halls, 1997a). As will be discussed in Section 9.6. this type of response (one in which the photocurrent peaks at the onset of absorption when the cell is illuminated through the ITO contact, and falls off when the absorption coefficient increases) is typical of organic photovoltaic cells. and indicates thai excitation of the polymer close to the aluminium back contact contributes most to the collected photocurrent. Photoconductivity measurements have been performed on many other conjugated polymers, many of which are based on PPV. These include poly(2,5-diheptyloxy-pphenylcnevinylene) (HO-PPV) (Frankevich et al.. 1996), ladder polymers with a double backbone such as polybenzimidazo-benzoisoquinoline (BBL) (Narayan et al., 1994). and polythiophenes such as poly(3-hexylthiophene) (Tada and Yoshino, 1997a). In these cells the monochromatic current collection efficiency in short-circuit mode rarely exceeds 1%. 0.08

3- o

1.8 o

1.8

2.2

2.6

3.0

3.4

Energy/eV Figure 9.10 Graph showing the photocurrent action spectrum of an ITO/MEH-PPV/A1 device, under illumination through, alternately, the aluminium electrode and the ITO electrode. The absorption spectrum of the polymer is shown for comparison (dashed line). After Halls (1997a).

398

J. J. M. Halls andR. H. Friend

9.5 Photogeneration and charge transport in organic PV cells 9.5.1 Introduction This section considers in more detail the physics underlying the operation of the simple organic photovoltaic cells introduced in the previous section. The production of a photocurrent in an organic cell of this type, in which the built-infieldis provided by the difference in work functions of the electrodes, arises from four distinct processes: 1. Absorption of a photon to create a carrier pair, which may be bound. 2. Dissociation or separation of the carrier pair. 3. Transport of electrons to one contact, and holes to the other contact, driven by the built-in field. 4. Collection of the charges at the electrodes. The absorption spectrum of the organic photoconductor defines the spectral range over which the cell will respond to light. This will depend on the chemical structure of the particular molecule or polymer, a useful feature which enables organic semiconductors to be synthesised with controllable absorption spectra. Most organic dyes and conjugated polymers are strongly absorbing, and a film of only a few hundred nanometres is sufficient to absorb a significant proportion of the light providing it falls within the absorption band of the material.

9.5.2 Photogeneration in organic semiconductors If a semiconductor band model is adopted, the photogeneration mechanism will be similar to that encountered in inorganic semiconductor materials: absorption of a photon creates an uncorrelated electron and hole which move to opposite contacts under the influence of the applied field and are collected, giving rise to a photocurrent. In contrast, if an exciton model is employed, the photogeneration process is complicated by the need for the bound electron-hole pair to be separated. It is almost universally considered that the excitonic model provides a more satisfactory basis for the understanding of the electronic properties of most conjugated polymers, and is certainly more applicable to molecular semiconductors. Although the internalfieldis relatively high (about 107 V m"1) in typical thin-film organic devices, it is unlikely that afield-assistedexciton ionisation, of the type first proposed by Onsager (1934), is the dominant charge generation mechanism in most

Organic Photovoltaic Devices

399

cells. Evidence points to the role of extrinsic effects in the ionisation of excitons in both molecular and polymer semiconductors. In practice, a molecular or polymer film will have a certain concentration of defects and impurities at which one may expect exciton dissociation to occur. We will now consider a number of these possibilities. Molecular oxygen Oxygen is known to act as an electron acceptor and may therefore assist in the ionisation of excitons. Lyons in 1955 (Lyons, 1955), and later Chaiken and Kearns (1966), proposed that the photogeneration process in anthracene involved diffusion of excitons to the crystal's surface, where sites associated with oxygen molecules act as deep electron traps. It later became generally accepted that anthracene was a predominantly extrinsic semiconductor. Harrison (1969) argued that a similar mechanism occurs in metal-free phthalocyanines, and investigations carried out in various controlled ambient conditions have shown that the photoconductivity of many other molecular semiconductors is enhanced in the presence of oxygen. Molecular oxygen has also been shown to play a similar role in conjugated polymer photodiodes. From secondary ion mass spectrometry (SIMS) measurements by Sauer et al. (1995), it is evident that molecular oxygen is one of the primary contaminants incorporated into conjugated polymer films. The role of oxygen in controlling the photoconductivity of a soluble PPV derivative, HO-PPV, was investigated by Frankevich et al. (1996). The photoconductivity of the polymer increased on application of an external magnetic field, and this magnetic field effect (MFE) was enhanced in the presence of oxygen. A photogeneration mechanism was proposed in which positive mobile charge carriers are produced by dissociation of singlet excitons by very fast electron transfer to oxygen molecules. The MFE was shown to be connected with the reaction of singlet and triplet excitons with oxygen molecules, the magnetic field causing spin evolution of various intermediate excited states. The MFE cannot be explained by a simple band-type model in which photogeneration generates uncorrelated electron-hole pairs prior to ultrafast electrontransfer to an acceptor species. Such a mechanism was proposed by Lee (1993a) and Sariciftci (1993a) for the sensitisation of the photoconductivity of conjugated polymers by the addition of fullerenes. In this scheme, holes have the greater mobility and are the main contributors to the photocurrent, whereas electrons are trapped on the acceptor molecules.

400

J. J. M. Halts and R. H. Friend

Carbonyl groups Photo-oxidation of many conjugated molecules causes a reduction of their luminescence efficiencies. Fourier transform infra-red (FTIR) spectroscopy of photooxidised PPV by Rothberg (1996a) suggests that double bonds in the molecular backbone are broken in the presence of light and oxygen, shortening the conjugation length. In addition, the carbonyl (C=0) content is seen to increase substantially. These effects are consistent with the formation of carbonyl groups at the vinylic carbon atom sites. These groups quench luminescence by ionising singlet excitons, through charge transfer of an electron to the electronegative carbonyl oxygen atom. This process is illustrated in Fig. 9.11. Rothberg has shown that one carbonyl group incorporated for every 400 PPV repeat units is sufficient to reduce the photoluminescence (PL) intensity by a factor of two. Harrison and co-workers (1996) found that photo-oxidised PPV samples contain a distribution of quenching centres following the profile of the absorption depth, which causes the PL efficiency to decrease with increasing excitation energy. While other photo-oxidation products may well be formed, the importance of Ihe aldehyde group in the quenching of luminescence has been confirmed by a study of the PL efficiency of model oligomers and their aldehyde counterparts (Rothberg el al.. 1996b).

pristine segment oxidised segment Figure 9.11 Schematic energy level diagram, showing exciton dissociation by charge-transfer of an electron to an oxidised chain segment. The hole remains on the pristine segment. After Rothberg (1996a).

Organic Photovoltaic Devices

401

Associated with this quenching of luminescence, various groups have observed an enhancement in the charge photogeneration efficiency, consistent with the dissociation of excitons by oxidised chain segments. The quenching of luminescence by activated exciton dissociation gives indirect evidence for the effect of carbonyl groups on the photoconductivity. In a study by Antoniadis et al. (Antoniadis et al, 1994a) thin PPVfilmswere exposed to white light in the presence of air, and the photocarrier generation efficiency was found to be increased by a factor of 40. Measurements of the dark conductivity ruled out an increase in mobility being responsible for the enhancement in the photocurrent. The photoconductivity enhancement was not, however, quantitatively commensurate with the quenching of the PL. Antoniadis therefore ruled out the dissociation of relaxed emissive excitons as the dominant mechanism for photogeneration. Instead, he suggested that an intrinsic photocurrent arises from the dissociation of the initially created hot excitons, and efficient extrinsic photogeneration occurs when these hot excitons are excited near photo-oxidatively introduced defects where dissociation is highly probable. Papadimitrakopoulos (1994) suggests that carbonyl groups may also be formed in PPV during the conversion process. The thermal conversion of PPV in a reducing environment was shown to decrease the concentration of carbonyl defects in the polymer film. Interchain effects Rothberg (1996a) has proposed that the primary product of photogeneration in PPV is not the singlet exciton, but instead a bound polaron pair, consisting of oppositely charged polarons on different conjugated segments, bound by the coulomb interaction. Rothberg considered that these polarons are formed from vibrationally hot intra-chain excitons, and argued that approximately 80-90 % of absorbed photons generate these polaron pairs, which are non-emissive, the remainder generating luminescent singlet excitons. This is consistent with the luminescence efficiency of the PPV used in Rothberg's study, which was -10%. Frankevich et al. (1992) proposed that polaron pairs play an important role in the photogeneration of charged carriers. Since the coulombic binding and wavefunction overlap of polaron pairs is smaller than that of singlet excitons, it is relatively easy for one or more of the oppositely charged polarons to slide away under the influence of the internal field. Rothberg concluded from photoinduced absorption measurements of pristine and oxidised PPV that interchain excitons (polaron pairs) are not quenched by carbonyl defects, perhaps because they are immobile. An intuitive model for a polaron pair is that of an electron trapped on a carbonyl group, bound by the coulomb interaction to an otherwise mobile hole on an adjacent chain or conjugated segment.

402

J. J. M. Halls andR. H. Friend

Mizes and Conwell (1994) have undertaken a theoretical study of polaron pairs. It is proposed by both these authors and by Rofhberg et al. (1996a) that the photoinduced absorption features measured in PPV are in fact due to the generation of polaron pairs, rather than the creation of triplet excitons (Brown et al, 1993) or polarons (Antoniadis et al, 1994a) which are common assignments of the excitedstate absorption transitions. Rothberg deduced from transient photoinduced absorption measurements that polaron pairs are formed in MEH-PPV films but not in solution, confirming the inter-chain nature of this species. The chemical tailoring of luminescent polymers has enabled synthesis of materials with luminescence quantum yields exceeding 60%, suggesting that, in these polymers at least, the yield of non-emissive polaron pairs must be less than 40%. Modelling of the photoluminescence excitation spectra of PPV by Harrison et al. (1996) indicates that the branching ratio for the production of singlet excitons from absorbed photons is approximately unity, ruling out the production of a significant number of nonemissive intra-chain species. It should be noted though that the polymer samples used by research groups around the world are likely to be very different in actual composition, morphology and contamination level, and these factors may well influence the ratio of singlet excitons to polaron pairs generated in a particular sample. Other mechanisms In an ideal polymer film, the polymer chains would be straight, since the bonding in these materials makes the chains rigid, and a linear shape is the most stable conformation. In practice, however, there are likely to be kinks in the polymer chains, distorting the conjugation and causing localised changes in the electronic energy levels of the frontier states. Charge transfer may occur to lower energy segments, giving rise to exciton dissociation. In a polycrystalline molecular photoconductor, grain boundaries and other surface and bulk defect sites may act as exciton dissociation sites. Finally, if the photoexcitation density in the photoconductor is high, exciton-exciton annihilation may be expected to occur, giving rise to charged polarons, the equivalent in organic semiconductors of free electrons and holes.

9.5.3 Charge transport in organic semiconductors There is extensive evidence from temperature- and field-dependent conductivity measurements that charge transport in both conjugated polymers and small molecules occurs by a hopping mechanism. Two mechanisms have been proposed:

403

Organic Photovoltaic Devices

1. It is generally held that 'free' charges on a molecule relax to form polaronic states. The transport of polarons involves delocalised movement within a conjugated domain, and intra- and inter-chain hopping to spatially separated sites. For a polaron to hop to a new site, an activation energy is required to overcome the binding energy associated with the lattice distortion (Schein et al, 1990). The mobility u depends on temperature according to u ocexp(-To/T)

(9.1)

2. In a disordered material there is a broad distribution of both the spatial overlap and energy difference between atomic sites which is best modelled by a variable-range hopping mechanism (Pautmeier et al, 1990). The theory associated with this model was developed to explain the conductivity of molecularly doped polymers, materials which are comprised of active molecules dispersed in an inert matrix, as are encountered in xerography. In this formalism, the mobility varies as

««exp[-(7;/7-)']

<9-2>

In each case, T0 is a constant related to the activation energy. In practice, it is difficult to measure the mobility over a sufficient temperature range to determine which of these mechanisms is the more appropriate. Neither formalism fully simulates the observed variation of mobility withfieldand temperature in conjugated materials. Time-of-flight (TOF) measurements can reveal valuable information related to charge transport in organic semiconductors. In a TOF experiment, a thin (-10 um) organic film is sandwiched between metal electrodes. Carriers are photogenerated by a short, strongly absorbed, laser pulse through one of the electrodes, and are separated by an externally applied electric field. One of the charged species recombines at the illuminated electrode, and the other traverses the polymer layer, giving rise to a timedependent photocurrent (Meyer et al., 1995). In PPV, the TOF current transients are broadened anomalously and the charge transport is said to be dispersive (Antoniadis et al, 1994c). The hopping rate decreases with time as the charge carriers relax into thermal equilibrium. This suggests the presence of disorder in the system, indicating that the disorder model may be more appropriate of the two models described above. The dispersive nature of transport in PPV has been attributed to trapping and release of charges at grain boundaries, defects and impurities (Meyer et al, 1995). As a consequence, the transit time is comparable to the time taken for carriers to equilibrate within the density of states. Release times of charges from traps in PPV may exceed 0.1 s at room temperature.

404

J. J. M. Halls andR. H. Friend

Meyer et al. (1995) observed that the transport of holes in DPOP-PPV, a soluble PPV derivative, was considerably better than in PPV itself, and concluded that this arose from the absence of grain boundaries in the material. Using TOF measurements, Meyer argued that charge transport in DPOP-PPV was best described by a model in which holes hop between monomer units which act as traps. The estimated trap depth corresponded well with the measured activation energy of the mobility, and the mean distance between trapping events (~4 A) was attributed to the distance between adjacent polymer chains. An intermolecular spacing of 6-8 A has been reported in stretch-aligned PPV (Granier et al, 1986). The mobilities of electrons and holes in PPV are very different, despite the fact that the conduction and valence bands have similar widths (Bredas et al., 1982). Hole mobilities of -10"4 cm2V"] s"1 have been measured in PPV (Obrzut et al., 1989; Takiguchi et al., 1987). Electron transients were not observed in these TOF experiments, indicating that the electron mobility is several orders of magnitude less than that of holes. Antoniadis et al. (Antoniadis et al, 1994c) reported mobilitylifetime products, ur, of 10~9 and 10 12 cm2 V"1 for holes and electrons respectively, corresponding to a mean range of 1 jum for holes and 1 nm for electrons. These measurements were made at a field of 105 Vm"1, typical of the electric field in a polymer sandwich photovoltaic cell. The lower electron mobility must arise from the presence of traps, which may be oxygen atoms incorporated in the polymer film. Since both positive and negative charges are present in the polymer layer of a photovoltaic cell following photoexcitation, it is inevitable that many will recombine as they traverse this layer, and so will not contribute to the photocurrent. The electron and hole arising from the dissociation of a singlet exciton may recombine, a process termed bimolecular geminate recombination. Alternatively, electrons and holes from different ionised excitons may recombine (non-geminate bimolecular recombination). The outcome of a recombination event may produce another singlet exciton, which may decay radiatively or non-radiatively, possibly giving rise to charged carriers again. The probability of recombination depends on the population of charges in the polymer layer. Bimolecular recombination is more likely at high light intensities. Trapped electrons are likely to act as recombination centres for holes, and the slow electron release rate indicates that there will be a high equilibrium concentration of these (Meyer et al., 1995). However, in the experiments by Frankevich (1996) on the magnetic field dependence of the photoconductivity it was proposed that the recombination of holes with 02~ ions, which were taken to be thefilledelectron traps, was very slow, and this was manifested by a quasi-persistent photoconductivity.

Organic Photovoltaic Devices

405

9.6 The characteristics of organic photovoltaic cells 9. 6.1 Photocurrent action spectra The dependence of the photocurrent in an organic photocell on the excitation wavelength provides important information about the photogeneration process in the device. It also dictates which applications the photocell may be suitable for. Many of the steps involved in charge generation in an organic photodiode are wavelengthdependent. The first of these is the depth profile of absorption in the semiconductor. At wavelengths where the absorption coefficient is high, light is absorbed close to the transparent contact; when the absorption coefficient is smaller, more light penetrates deeper into the film, closer to the back metal contact. Second, a certain fraction (the branching ratio) of absorbed photons creates singlet excitons, and it is assumed that the photocurrent is dominated by their ionisation. In some molecular systems, this fraction is found to be energy-dependent, as in the case of sexithienyl (Dippel et al., 1993), although a recent investigation by Harrison et al. (1996) suggests that in pristine PPV the branching ratio is broadly independent of excitation energy. Third, in a simple model of photoexcitation, there is no difference between the excitons created at different photon energies. 'Hot' excitons created by photons with energies high above the band gap thermalise to form 'cold' excitons on a timescale of ~100 fs, considerably faster than the processes of ionisation and radiative decay. However, a recent theoretical study by Bredas and co-workers indicates that excitons created in higher energy bands have a more delocalised wavefunction than those created closer to the band edge, and have a lower binding energy (Kohler et al., 1998). This phenomenon is thought to contribute to the sharp rise in quantum efficiency with increasing photon energy in some polymer photocells. A common phenomenon encountered in organic photocells is that illumination through one of the electrodes yields a photocurrent action spectrum which peaks at wavelengths just below the absorption edge and falls to a minimum where the absorption peaks. Such a response is exhibited in Figs. 9.9a and 9.10. An antibatic response of this type indicates that the active area for photogeneration is situated deep in the device, in the vicinity of the back electrode. The bulk of the film acts as an optical filter, so that when the absorption in the film is strong the quantity of light penetrating to the active layer is small. This scenario is illustrated in Fig. 9.12. We now consider mechanisms that would give rise to such a response.

406

J. J. M. Halls andR. H. Friend

organic layer

this region acts as an optical filter

absorption in this region gives rise to a collected current

Figure 9.12 Schematic cross-section through an ITO/semiconductor/metal sandwich cell, showing how the optical filter effect can modify the photocurrent action spectrum so that the photocurrent peaks soon after the onset of absorption, and reaches a minimum at the absorption maximum.

Exciton diffusion to the polymer/metal interface In this model the device performance is controlled by the diffusion of neutral photoexcited species to the back electrode. Excitons in molecular semiconductors are known to be quenched at the interface with a metal (Persson and Lang, 1982). There are several non-radiative decay mechanisms that contribute to this quenching process, not all of which give rise to charge separation. However, if dissociation does not occur efficiently in the bulk of the film, ionisation at the interface may be the dominating photogencration process. Lyons (1955) and Kepler (1960) both reported an antibatic relationship between the photocurrent and absorption spectrum of anthracene. The response was considered to arise from the diffusion of excitons to surface sites occupied by oxygen molecules, where they dissociate. This model was also used to explain the photocurrent action spectra of the Al/merocyanine dye/Ag cells characterised by Ghosh and Feng (1978). The results fitted a model in which only those excitons that diffused to the merocyanine/Al interface produced free carriers, by injection of a hole into the merocyanine and an image charge into the metal. The active region in each of these cases is the zone in which excited excitons can diffuse to the back metal contact, and is equal in width to the exciton diffusion range.

Organic Photovoltaic Devices

407

Electron diffusion model In this model, the properties of the photovoltaic cell are controlled by the diffusion characteristics of the charges with the lowest mobility. The probability of exciton dissociation is assumed to be constant throughout the organic layer. As we have seen, in most conjugated polymers electrons have a shorter mean range than holes. In a simple ITO/polymer/Al photodiode, electrons are driven by the internal field to the aluminium electrode, and holes drift to the ITO contact. The photocurrent is therefore expected to be higher when charges are generated close to the metal contact so that the electrons have a shorter distance to travel to escape the device. For illumination through the ITO contact, this condition is satisfied when the absorption coefficient of the material is low and light can penetrate to the aluminium contact. When the absorption coefficient is high and the penetration depth of light is small the probability of an electron generated close to the ITO contact travelling to the opposite contact is small. The presence of trapped electrons modifies the electric fields in the device, reducing the electric field where the trapped charge density is high, and therefore suppressing further charge separation. Trapped electrons also act as recombination centres for holes, further reducing the collected photocurrent. In this model, then, the width of the active layer is linked to the mean electron range. This model has been used to understand the response of some PPV-derivative based photovoltaic cells, such as that shown in Fig. 9.9 (Marks et al, 1994). Schottky barrier model Under conditions of high extrinsic charge density, a Schottky barrier may form at the polymer/aluminium interface. Excitons created in the high-field depletion region, or within diffusion range of it, are likely to be ionised and give rise to a photocurrent. The depletion region is therefore an active region for photogeneration. The photovoltaic response recorded by Tang and Albrecht (1975a) in metal/chlorophyll-a/metal cells, shown in Fig. 9.7a, was attributed to photogeneration at the blocking contact, at which a Schottky barrier is formed. The photocurrent action spectra suggested an active area width of approximately 250 A, but the diffusion length of excitons in chlorophyll-a was found to be only 130 A. This indicated that exciton diffusion to the metal contact followed by dissociation at the electrode was not the predominant mechanism for photogeneration, unlike the case in merocyanine cells (Ghosh and Feng, 1978). Instead, a combination of diffusion of excitons to the depletion region, and ionisation in that region was proposed. A similar photogeneration mechanism was invoked to explain the operation of tetracene photocells (Ghosh and Feng, 1973).

408

J. J. M. Halls andR. H. Friend

Refined model A more refined model for the action spectra of ITO/polymer/metal cells has been developed by Harrison (1997). The sharp rise in photoconductivity at the onset of absorption was interpreted in terms of the excitation of longer, more planar, chain segments at low photon energies. Harrison suggests that exciton ionisation is efficient in these regions, as a result of an extended exciton lifetime near the localisation edge, or enhanced interchain separation within small crystallites. The model modifies the traditional diffusion-based models of Ghosh and others, which is still required to explain the action spectrum over the full spectral range. All of the above models give very similar simulated action spectra and further information is required to determine which model best describes the photogeneration process in a particular organic photodiode. Furthermore, the actual mechanism of photogeneration is likely to involve a combination of several of the above processes. For example, Ghosh et al. (1974) proposed that photogeneration in an Al/MgPh/Ag cell occurred by dissociation of excitons at impurity sites in the bulk of the polymer film, followed by diffusion of the majority charge carriers to a Schottky barrier at the Al/organic-layer interface where charge collection is efficient.

9.6.2 Current-voltage characteristics Figure 9.13 shows typical I-V characteristics of an organic photovoltaic cell (in this example, one containing PPV). In the dark, the cell shows rectifying behaviour, with a forward-bias turn-on at around 1 V, whereas under illumination (shown with filled circles) a photocurrent is developed. As we noted earlier, it is convenient to model most organic devices as tunnel diodes, in which the depletion region extends throughout the organic layer. The insets in Fig. 9.13 illustrate the field distributions in a tunnel diode under various bias conditions. The HOMO and LUMO lines indicate the top of the valence band and bottom of the conduction band respectively. Inset a shows the diode in short-circuit mode (i.e. the potential difference between the two contacts is constrained to be zero). The internal field, equal in magnitude to the work function difference divided by the organic layer thickness, drives negative charges to the aluminium contact and holes to the ITO. By applying a reverse bias (with the ITO contact held negative relative to the aluminium, as illustrated in inset b) the internalfieldincreases in magnitude and in many cases this enhances the quantum efficiency of photoconductivity, allowing the cell to be used as an effective photo-

Organic Photovoltaic

Devices

409

diode. If the load has a high impedance the flow of electrons and holes to opposite contacts allows a potential difference to develop across the cell, which approaches the work function difference as the light intensity increases. This is the open-circuit voltage, and under this condition the bands are approximately flat, as shown in inset c. Finally, if a positive bias is applied to the cell which is greater in magnitude than Vx then the bands are tilted in tf i opposite direction, allowing tunnelling of charges into the polymer layer, as shown in inset d. In many luminescent organic semiconductors these charges can recombine and give rise to luminescence, a phenomenon exploited in organic emissive displays.

-1.5

-1

-0.5

0 0.5 External bias / V

1

1.5

2

Figure 9.13 The current-voltage characteristics of an ITOPI'V Al photovoltaic device, measured in the dark (open circles) and in the light (filled circles). Hie shape of the l—V curves is interpreted in terms of a tunnel-diode model in which the current is controlled by tunnelling through the triangular barriers at the polymer/contact interfaces, as illustrated schematically in insets a-d. These show the direction and magnitude of the electric field under a variety of biasing conditions: (a) under short-circuit conditions; (b) in reverse bias, corresponding to a negative voltage on the ITO: (c) under the fiat-band condition at a small forward bias equal to the difference in the work functions of the two contacts; (d) under a forward bias greater than the work function difference, at which injection of carriers followed by recombination and light emission may occur. The red arrows represent absorption and emission of light. After Halls (1997a).

Extensive investigations have been carried into the l-V characteristics of sandwich-type polymer diodes, largely under forward bias (the regime in which devices of this type exhibit electroluminescence). Parker (1994) proposed that current flow in MEH-PPV sandwich devices is controlled by tunnelling of charges through the triangular barriers formed at the polymer -metal interlaces. By analysing the

410

J. J. M. Halls and R. H. Friend

characteristics of devices fabricated from MEH-PPV, with electrodes chosen to make either electron or hole injection alone the dominant process, Parker demonstrated that the form of the I-V curves could be modelled closely by Fowler-Nordheim fieldemission theory. In this model, the current varies according to the relationship I x£2 exp

with K = — \ £ J

3 h

-

(9.3)

1

where