Solar Energy Materials & Solar Cells 72 (2002) 49–58
Solar cells from upgraded metallurgical grade (UMG) and plasma-purified UMG multicrystalline silicon substrates S. De Wolf a,*,1, J. Szlufcika, Y. Delannoyb, I. Pe! richaudc, C. H.alerd, R. Einhause a IMEC vzw, Kapeldreef 75, B-3001 Leuven, Belgium EPM ENSHMG, BP 95, F-38402 St-Martin d’H"eres, France c UMR TECSEN, University of Marseille, F-13397 Marseille, France d Bayer AG, Postfach 166, D-47812 Krefeld, Germany e Photowatt International SA, F-38300 Bourgoin-Jallieu, France b
Abstract High impurity concentrations do not allow the direct use of upgraded metallurgical grade (UMG) Si for PV production. A newly developed prototype inductive plasma-purification system and process allowed the significant reduction of the elements B, C, O, P, Al, Ca, Fe and Ti, depending on the duration of the treatment. Based on this type of purification, it is shown that subsequent appropriate low-cost cell-processing yields homogeneously distributed energyconversion efficiencies throughout the cast ingots. Stabilised cell efficiencies of up to 14.7% were already experimentally shown to be attainable on highly B-doped (ro0:1 O cm) 102 cm2 multi-crystalline Si substrates of high purity. On plasma-purified UMG p-type 0.1–0.2 O cm ingots, efficiencies of up to 12.38% are reached, to be compared with about 10.12% on the same material without prior plasma treatment. Some light-induced degradation is present on processed samples, which is most likely linked to the presence of metastable boron–oxygen complexes in the material, and results in stabilised efficiencies of, respectively, 12.19% and 10.00%. r 2002 Elsevier Science B.V. All rights reserved. Keywords: Crystalline silicon; Impurities; Defects; Plasma purification; Low resistivity; Degradation
*Corresponding author. Tel.: +32-1628-1685; fax: +32-1628-1501. E-mail address:
[email protected] (S. De Wolf). 1 E.E. Department, Katholieke Universiteit Leuven, B-3000 Leuven, Belgium. 0927-0248/02/$ - see front matter r 2002 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 7 - 0 2 4 8 ( 0 1 ) 0 0 1 4 9 - 0
50
S. De Wolf et al. / Solar Energy Materials & Solar Cells 72 (2002) 49–58
1. Introduction For the near future, the crystalline silicon using photovoltaic industry is expected to face a severe shortage of its classical silicon feedstock sources. This scenario is mainly caused by two driving forces: The PV industry continues to experience a rapid annual growth, where most commercially operating production lines rely solely on crystalline silicon-based processes. At the same time, the availability of currently used electronic grade silicon rejects from the semiconductor industry becomes more and more limited, which originates from the increasing reclaims by the latter industry and the presence of a shift towards higher doped silicon within micro-electronics. To tackle this shortage problem, novel purification techniques for low-quality Si feedstock material have recently been investigated in the frame of the European ARTIST project, and are meant to open alternative routes towards a solar grade (SoG) silicon feedstock, exclusively destined for the PV market [1]. One of these investigated alternative silicon feedstock sources was the upgraded metallurgical grade (UMG) silicon. However, the high concentration of metallic and non-metallic impurities generally do not allow the direct use of this material for PV production since not all of the present impurities can be removed sufficiently by segregation during directional solidification. For this, a prototype plasma purification unit has been developed within the project, which combines an inductive plasma-torch with an inductive cold crucible. In this manner, silicon can be melted and stirred inside the crucible by electromagnetic forces, while the plasma is able to volatilise impurities present at the constantly renewed liquid-silicon surface. B removal is considered to be a key-issue in alternative purification techniques due to its rather high segregation coefficient kSO : The influence of the B content in the material on the solar-cell output characteristics will determine, to a large extent, the required purification-parameters prior to casting. For this, a feasibility study of solar-cell processing of very low resistivity (VLR; ro0.1 O cm) B-doped wafersF but of high purityFis a prerequisite.
2. Wafer fabrication and cell processing An uncoated and a Si3N4-coated quartz crucible were, respectively, used for melting and crystallisation of several 30 kg multi-crystalline ingots. Three basic types of Si-based material were cast: VLR-material of high purity and, respectively, unpurified and plasma-purified UMG material. Details of the plasma treatment are discussed elsewhere [2,3]. Wafers (area of 102 cm2 and thickness of about 350 mm) were subsequently sawn from the respective ingots. Samples throughout the obtained columns were then processed using a low-cost screenprinting technology, based on the fired-through PE-CVD SiNX:H approach (Fig. 1). Compared to standard 1 O cm material, some processing steps need optimisation, such as the dopant-dependent acidic isotropic texturing, the emitter diffusion and the electrical contact definition.
S. De Wolf et al. / Solar Energy Materials & Solar Cells 72 (2002) 49–58
51
Saw damage removal and acidic isotropic texturing POCl3 emitter-diffusion Parasitic edge-junction removal by plasma etching PE-CVD SiNx:H deposition Screenprinted metallisation (emitter and base) Co-firing of the contacts
Fig. 1. Generic process flow.
35 Jsc [mA/cm ] 2
30 25 20 15 10 1.00E+16
PC-1D experimental 1.00E+17
1.00E+18
1.00E+19
background concentration [cm-3] Fig. 2. Experimental and simulated dependence of Jsc on background doping, based on different VLR ingots.
3. Results and discussion 3.1. VLR material Solar cells were processed from a number of ingots with average B concentrations varying from 1 1016 to 1 1019 cm3. Simulated data (PC-1D) of the short-circuit current Jsc show an expected decrease as a function of an increase in doping density, due to Auger recombination and a diminishing minority-charge carrier mobility (Fig. 2). Experimental data, as obtained from processed samples from the respective ingots, display, however, a more severe decline in Jsc : This might indicate the presence of an increasing concentration of impurities or crystal defects with increasing background-doping levels. It can be concluded that dopant densities below 6 1017 cm3 are required to avoid serious decrease in solar cell performance due to heavy-doping effects.
52
S. De Wolf et al. / Solar Energy Materials & Solar Cells 72 (2002) 49–58
1.00E+18 -3
concentration [cm ] oxygen boron 1.00E+17 0
20
40
60
80
100
relative position in column (bottom → top) [%] Fig. 3. Interstitial oxygen- and electrically active boron-concentration throughout a VLR ingot.
15
η [%] 14 13 12 0
20
40
60
80
100
relative position in column (bottom → top) [%] Fig. 4. Stabilised efficiencies throughout the VLR ingot (ro0:1 O cm).
Generally, within an ingot, as for most impurities, the B concentration is seen to increase slightly towards its top due to segregation during crystallisation. On the other hand, the oxygen concentration of the ingot is governed by oxygen incorporation into the melt through the quartz crucible and oxygen release from the melt through SiO evaporation, usually resulting in a decrease towards the ingot top. Fig. 3 represents such variations in concentration within one of the cast VLR ingots as a function of the relative column height. The boron concentration rises from 3 1017 to 6 1017 cm3, while the oxygen concentration decreases more significantly from the bottom to the lower top region of the ingot, with values of 3.9 1017–1.1 1017 cm3, respectively. Employing the fired PE-CVD SiNX:H screen-printing approach, solar cells were processed from samples taken throughout the latter VLR ingot. Fig. 4 shows how the energy-conversion efficiency Z varies throughout the ingot, when identical processing conditions were applied for all samples. Due to the presence of some light-induced degradation, the samples were measured after prolonged illumination, from where on the values are considered as being stabilised. The decreasing resistivity trend, as encountered in the ingot, is however not clearly reflected in this curve. The relatively inferior electrical behaviour of the top part of the ingot can
S. De Wolf et al. / Solar Energy Materials & Solar Cells 72 (2002) 49–58
53
Table 1 Illuminated I(V)-parameters of the best VLR cell after degradation Measurements at Fhg-ISE Z (%) Jsc (mA/cm2) Voc (mV) FF (%)
14.7 29.91 624 78.6
0.3 pl. p. UMG
[Ω cm]
UMG 0.2
0.1 0
100 20 40 60 80 relative position in column (bottom → top) [%]
Fig. 5. Resistivities throughout unpurified and plasma-purified UMG ingots.
possibly be attributed (next to the increase in boron concentration) to the presence of a relatively larger amount of crystal defects due to casting and segregation of impurities in this region. For the bottom part of the ingot, the lower efficiencies might again be related to the presence of a larger amount of crystal defects due to casting, and also to the presence of certain oxygen-related defects (as oxygen diffuses from the crucible into adjacent parts of the ingot). Some of the latter defects are seen to degrade solar cells under illumination (see also Section 3.3), others can act e.g. as thermal donors [4]. The best cells made from this type of highly doped material, gave stabilised efficiencies of up to 14.7% (15.07% before degradation), as confirmed by Fhg-ISE (Table 1). This result clearly shows that in combination with appropriate cell processing, highly B-doped material may represent an alternative feedstock source for PV production in the future. 3.2. UMG and plasma-purified UMG material In the case of non-purified and plasma-purified UMG Si, the resistivities are seen to increase towards the top of their respective ingots (Fig. 5). This is expected to be attributable to some neutralisation effect of P atoms from the electrical point of view, due to a transformation of these atoms into phosphates during ingot manufacturing. The plasma-treated material is seen to have a slightly higher resistivity, although it is still close to 0.1 O cm. Fig. 6 compares the impurity distribution throughout both the ingots. In Table 2, the characteristics of the best cells of each ingot are given (italics give stabilised values), while in Fig. 7, again,
S. De Wolf et al. / Solar Energy Materials & Solar Cells 72 (2002) 49–58 UMG silicon only
1E+19
Plasma purified UMG silicon
1E+20
P
Cu
P
Cu
Cr
Mn
Cr
Mn
V
Al
V
Al
Ca
Fe
Ca
Fe
-3
-3
concentration (cm )
1E+20
concentration (cm )
54
Ti 1E+18
1E+17
1E+16
1E+19
Ti 1E+18
1E+17
1E+16
0
50
100
150
0
block height (mm)
50
100
150
block height (mm)
Fig. 6. Impurity distribution in directionally solidified UMG silicon without prior purification (left) and after plasma purification (right). In the latter ingot, the concentrations of Cr, V, Ti, Cu, Mn, Al and Mg were below the detection limit (1.2 1016 cm3), even in the segregation region. Table 2 Illuminated I(V)-parameters of the best cells from unpurified and plasma-purified UMG material, results in italics are after degradation r (O cm)
Jsc (mA cm2)
Voc (mV)
FF (%)
Z (%)
nmpp
Unpurified UMG
0.135
21.83 21.68
603.3 601.2
76.84 76.72
10.12 10.00
1.30 1.31
Plasma-purified UMG
0.139
26.43 25.93
612.8 611.5
76.43 76.92
12.38 12.19
1.21 1.25
13
η [%] 11 9
pl. p. UMG UMG
7 0
100 60 80 20 40 relative position in column (bottom → top) [%]
Fig. 7. Stabilised efficiencies throughout unpurified and plasma-purified UMG ingots.
stabilised efficiencies are displayed throughout both the columns, when the same processing scheme is applied for all wafers. It can be seen clearly that an efficiency gain of about 2% absolute is reached, where some regions are even more improved.
S. De Wolf et al. / Solar Energy Materials & Solar Cells 72 (2002) 49–58
55
Table 3 teff throughout the processing sequence for neighbouring wafers (A–E) from unpurified UMG Si A Saw damage removal and acidic isotropic texturing X POCl3 emitter diffusion POCl3 emitter diffusion temperature profile (no process gases) PE-CVD SiNx:H deposition on the front side PE-CVD SiNx:H deposition on the rear side Screenprinted Al on the rear side Firing step Al removal in HCl and/or CP4 etch X PE-CVD SiNx:H deposition on both sides X teff (injection level: B1014 cm3) (ms) 0.15
B
C
D
E
X
X X
X X
X X
X X
X
X
X X 0.18
X X 0.27
X X X 0.37
X X X X 0.36
The effect that the overall solar cell process has on the bulk quality of the material is shown in Table 3. The effective minority charge carrier lifetime teff was measured after each single processing step on a set of neighbouring wafers from the unpurified UMG ingot, using a quasistatic photoconductance analysis method and at injection levels close to normal solar-cell operation conditions under a 1 sun AM1.5 spectrum (B1014 cm3). Prior to measurement, all samples received a CP4 etch to remove all the present features, followed by the deposition of not-fired PE-CVD SiNX:H layers on both sides for surface passivation. Although care should be taken with interpreting samples as a function of their thermal history (see also Section 3.3), the results suggest the beneficial effect the POCl3 diffusion for junction definition (phosphorous gettering), as well as the fired PE-CVD SiNX:H layers (bulk passivation by hydrogenation) have on teff. 3.3. Light-induced degradation As stated, a Jsc degradation under illumination has been measured for all types of investigated samples. In literature, t instabilities have been observed for already some time in mono-crystalline 1 O cm Cz–Si of high purity [5]. More recently, this was attributed to the reversible formation and annihilation of BOn complexes under, respectively, illumination (or forward bias, thus injection) and annealing. Initially, a model was suggested in which n ¼ 1 [6], although, currently, there seems to exist a consensus that nB5 [7,8]. The boron atom of this complex is supposed to reside on a substitutional lattice site, while the oxygen atoms are presumably in an interstitial configuration [9]. To diminish the encountered degradation, a number of suggestions have been proposed in literature for 1 O cm Cz–Si solar cells, which can basically be divided into three categories: changing the boron content, oxygen reduction or optimised cell processing. For the VLR material of high purity, a detailed analysis of the Jsc degradation under illumination was already given elsewhere [10]. It was concluded that the phenomenon is most likely to be linked to a reversible BOn complex formation
56
S. De Wolf et al. / Solar Energy Materials & Solar Cells 72 (2002) 49–58
4 Relative decrease of Jsc 3 [%] 2 1 0 0
20
40
60
80
100
relative position in column (bottom → top) [%] Fig. 8. Relative Jsc degradation under illumination (AM1.5-1sun) throughout the VLR-column for cells with a fired PE-CVD SiNX:H layer.
within the material, very similar to the degradation behaviour as encountered in 1 O cm Cz–Si. Furthermore, the oxygen content in the material was seen to play a decisive role for the formation of these defects. The relative decrease in Jsc (and thus light-induced degradation) reaches a maximum in the lower middle part of the VLR ingot, while the bottom of the ingot, which is the part with the highest oxygen concentrations (Fig. 3), shows the lowest efficiencies. However, this region does not suffer more from this degradation (Fig. 8). Presumably, competing oxygen-based, but illumination-independent, defects are formed at higher oxygen concentrations, and deplete the oxygen required to form BOn complexes in the same zones in this way. Fired PE-CVD SiNX:H layers yield passivation of BOn complexes by hydrogenation, leading to a diminished degradation action. Finally, a first lowtemperature post-annealing (B2001C for 300 ), after stabilisation of degradation, was able to restore the initial condition of the cell to some extent. Subsequent illumination/annealing cycles showed a completely reversible behaviour only from then on. The resulting difference between the first and subsequent cycles was attributed to a partial de-passivation of the BOn defects by the first post-annealing. For less pure UMG and plasma-purified UMG material, basically the same holds: again, a completely reversible degradation phenomenon is present after some annealing/illumination cycles. This is believed to be the completely reversible BOnrelated degradation that is starting to show its net effect. However, the first annealing treatment is not able to recover completely the initial condition that the cell was in, while the subsequently reached degradation stabilisation level is also lower than that after the first illumination cycle (Fig. 9). This non-reversible behaviour cannot solely be related to the previously proposed de-hydrogenation effect due to annealing, since samples with no fired PE-CVD SiNX:H layer, and thus no hydrogenation, suffer from the same trend (Fig. 10). Possibly, the reaction dynamics of the BOn complex are less straightforward than initially thought. Another theory could be that despite the plasma purification, traces of impurities, such as, e.g. Fe, are still present in the material that can cause the formation of other illumination/annealing sensitive defects. The Fei donor is known to be able to form a stable pair with a Bs acceptor [11]. This FeB pair is about a 10-times-less-effective recombination centre than Fei. However, the pair can be dissolved in its constituents
S. De Wolf et al. / Solar Energy Materials & Solar Cells 72 (2002) 49–58
57
26.5
Jsc [mA/cm ] 2
25.5
24.5 0
5
10
15
20
Illumination time (AM1.5 – 1sun) [min] Fig. 9. Light-induced degradation of Jsc directly after firing, and after cyclic annealing treatments of plasma-purified UMG material, with fired PECVD SiNX:H layer.
Jsc [mA/cm2] 20
19 0
5
10
15
Illumination time (AM 1.5 – 1 sun) [min] Fig. 10. Light-induced degradation of Jsc directly after firing, and after cyclic annealing treatments of plasma-purified UMG material, without fired PECVD SiNX:H layer.
by a short annealing at conditions that are similar to those used for the annihilation of the BOn complexes. In this way, the (initial) annealing treatment synchronically improves and destroys contributing parts to teff by, respectively, annihilation of BOn and FeB complexes. Further work is needed to discriminate between the suggested phenomena.
4. Summary and conclusions Multi-crystalline B-doped Si-substrates with resistivities o0.1 O cm, as well as, respectively, unpurified and plasma-purified UMG Si wafers have been studied in this paper for a possible application in solar cell manufacturing. Solar cells were successfully processed from these materials with a low-cost screenprinting technology and efficiencies of up to 15.07%, 12.38% and 10.12% have been achieved on the respective materials. Although a non-uniform Jsc degradation under illumination has been measured throughout all the ingots yielding stabilised efficiencies of, respectively, 14.7%, 12.19% and 10.00%; these results prove the
S. De Wolf et al. / Solar Energy Materials & Solar Cells 72 (2002) 49–58
58
potential of VLR material as solar cell substrates, as well as the capability of the applied plasma purification for UMG Si to remove electrically harmful impurities. The degradation is thought to be an indication of the presence of metastable BOn complexes, although dissolving FeB complexes might contribute as well to the encountered phenomenon.
Acknowledgements The funding of parts of this work by the EC in the frame of the Joule-III-ARTIST Project (No. JOR3-CT98-0228) is gratefully acknowledged.
References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
R. Einhaus, et al., Proceedings of the 28th IEEE PVSC, Anchorage, AK, 2000, pp. 221. C. Alemany, et al., E-MRS 2001 Spring Meeting, Strasbourg, France, 2001. Y. Delannoy, et al., E-MRS 2001 Spring Meeting, Strasbourg, France, 2001. C. H.aler, et al., Proceedings of the 14th EPVSEC, Barcelona, Spain, 1997, pp. 720. H. Fischer, W. Pschunder, Proceedings of the 10th IEEE PVSC, Palo Alto, CA, 1973, pp. 404. J. Schmidt, et al., Proceedings of the 26th IEEE PVSC, Anaheim, CA, 1997, pp. 13. S.W. Glunz, et al., Proceedings of the Second WCPSEC, Vienna, Austria, 1998, pp. 1343. J. Schmidt, A. Cuevas, J. Appl. Phys. 86 (1999) 3175. J.C. Bourgoin, et al., Proceedings of the 16th EPVSEC, Glasgow, UK, 2000, pp. 1356. S. De Wolf, et al., Proceedings of the 28th IEEE PVSC, Anchorage, AK, 2000, pp. 53. J. Lagowski, et al., Semicond. Sci. Technol. 7 (1992) 185.
