Investigating Wafer Quality in Industrial Czochralski‐Grown Gallium‐Doped p‐Type Silicon Ingots with Melt Recharging

Herein, a systematic study of the electronic quality of gallium‐doped p‐type silicon wafers from Czochralski‐grown ingots with melt recharging is presented. It is found that in the as‐grown state, the ingots contain interstitial iron concentrations in the range of 3 × 109–2 × 1010 cm−3, with a trend of slightly higher concentrations toward the tail end of each ingot, and in subsequently grown ingots. However, analysis of the effective lifetimes indicates that iron–gallium pairs are not the dominant recombination centers in the as‐grown state. Moreover, when these wafers are subjected to a tabula rasa step, an increase in the iron concentration is observed in the range of 1 × 1010–6 × 1010 cm−3, with iron–gallium pairs becoming the dominant recombination centers. This is possibly caused by the dissolution of pre‐existing precipitated iron in the wafers. Nevertheless, the negative impact of iron contamination can be dramatically reduced by subjecting the wafers to a phosphorus diffusion gettering step, as is commonly incorporated in the fabrication of p‐type passivated emitter and rear cells. Therefore, it is concluded that the quality of the ingots is not limited by iron contamination, even after multiple ingots are pulled from the recharged melt.


Introduction
The silicon photovoltaic (PV) industry has recently moved away from boron (B) and transitioned to gallium (Ga) as the main dopant element for p-type Czochralski-grown silicon (Cz-Si) wafers, primarily to avoid the well-known boron-oxygen defect. [1,2] Although several articles have reported higher effective minority carrier lifetimes in Ga-doped wafers than in B-doped wafers, [3][4][5][6][7] detailed studies of the material quality of industrial Ga-doped ingots are still rather limited. Recently, Horzel et al. [7] reported effective lifetimes in the range of 3-17 ms along a customgrown Ga-doped Cz ingot with a resistivity range of 5-12 Ω cm. However, studies of effective lifetimes along standard ingots with resistivities below 1 Ω cm, as is typical for industrial Ga-doped passivated emitter and rear cells (PERC) are necessary to assess the quality of ingots used in mass production.
Further, the silicon PV industry has adopted the recharged Czochralski silicon (RCz-Si) process as a mainstream ingot growth process, moving away from the single Cz-Si ingot pulling method. RCz-Si allows for the growth of multiple ingots from a single crucible, with melt replenishment after each ingot pulling. As a result, it offers significant cost reduction opportunities via reduced energy consumption, reduced material handling requirements, reduced labor costs, and faster throughput. [8,9] However, given the possibility of impurities accumulating in the residual melt after each ingot growth, investigating the material quality of the wafers from subsequent ingots grown in the same RCz-Si process is important to identify and quantify possible contamination.
Iron is one of the most ubiquitous and unavoidable metal impurities in silicon technology. Furthermore, it is well established that dissolved interstitial iron (Fe i ) forms iron-acceptor pairs with substitutional acceptors such as B À , Ga À , and In À . [10,11] Iron-gallium (FeGa) pairs introduce recombination centers which ultimately reduce the electronic quality of the Ga-doped p-type silicon. [11][12][13] Post et al. [3] reported that high carrier lifetime Ga-doped silicon wafers can be degraded noticeably even if low concentrations of Fe (<10 9 cm À3 ) are present. Therefore, an investigation to identify the presence of iron in industry-grown RCz-Si Ga-doped wafers from subsequently grown ingots may allow a quantitative assessment of the degree of metal accumulation in the recharged melt.
In this work, we studied wafers from three different Ga-doped ingots grown in the same RCz-Si ingot growth process. First, we investigate their material quality as indicated by effective carrier lifetimes before any high-temperature processing, reflecting the DOI: 10.1002/solr.202300304 Herein, a systematic study of the electronic quality of gallium-doped p-type silicon wafers from Czochralski-grown ingots with melt recharging is presented. It is found that in the as-grown state, the ingots contain interstitial iron concentrations in the range of 3 Â 10 9 -2 Â 10 10 cm À3 , with a trend of slightly higher concentrations toward the tail end of each ingot, and in subsequently grown ingots. However, analysis of the effective lifetimes indicates that iron-gallium pairs are not the dominant recombination centers in the as-grown state. Moreover, when these wafers are subjected to a tabula rasa step, an increase in the iron concentration is observed in the range of 1 Â 10 10 -6 Â 10 10 cm À3 , with iron-gallium pairs becoming the dominant recombination centers. This is possibly caused by the dissolution of pre-existing precipitated iron in the wafers. Nevertheless, the negative impact of iron contamination can be dramatically reduced by subjecting the wafers to a phosphorus diffusion gettering step, as is commonly incorporated in the fabrication of p-type passivated emitter and rear cells. Therefore, it is concluded that the quality of the ingots is not limited by iron contamination, even after multiple ingots are pulled from the recharged melt.
as-grown quality. Then, we aim to identify the defects limiting the bulk quality of the wafers before and after impurity gettering and tabula rasa steps, which are intended to remove dissolved metal impurities and dissolve grown-in oxide particles, [14] respectively. In particular, we track changes in the interstitial Fe concentrations [Fe i ] along the ingots, and before and after the hightemperature processing steps.   Figure 1a by the fit of the Scheil equation, C s ¼ kC 0 ð1 À gÞ kÀ1 , where C s is the concentration of dopant in the ingot, at a solidified fraction g, C 0 is the initial concentration of the dopant in the melt, and k is the effective segregation coefficient. The effective segregation coefficient (k Ga ) of Ga is 0.008. [15] As is often observed, the dopant distribution does not fit Scheil's law very well at higher solidified fractions (%1), as can be seen in the R9 ingot in Figure 1a. This could be at least partly due to uncertainty in the solidified fraction estimation arising from the residual melt in the crucible. Further, Figure 1b shows the distribution of [O i ] in the three RCz-Si ingots, which was measured to be in the range of 5-7 Â 10 17 cm À3 . The [O i ] distribution shows the expected opposite trend to the dopant distribution, decreasing toward the bottom of the ingots, due to the effective segregation coefficient of oxygen being greater than 1 as a result of evaporation from the melt. [16] We note a clear increase in the dissolved oxygen concentration for the subsequently pulled ingots. Figure 2 shows the measured effective lifetimes (τ eff ) of the samples at an excess carrier density (Δn) of 1 Â 10 15 cm 3 as a function of solidified fraction from the three ingots. Based on the surface recombination velocity (S) value of 7.5 cm s À1 and in the absence of defect-related bulk recombination, we calculated the expected effective lifetimes in the three RCz-Si ingots, due to only intrinsic and surface recombination, via

Effective Lifetimes
W , as shown in Figure 2. Given that the measured τ eff values are significantly  www.advancedsciencenews.com www.solar-rrl.com lower than the τ eff ðintriþsurf Þ values, we conclude that these wafers are at least partly affected by unintended bulk defects. Furthermore, we observed two trends in the measured effective lifetime profiles in these ingots. First, the τ eff within a given ingot shows a decreasing trend from the top to the bottom of the ingot. This is likely due to the segregation of contaminants such as metals, with segregation coefficients well below 1. Second, we also observe a slight decreasing trend in τ eff of the samples for the ingots pulled subsequently in the RCz-Si growth process (τ R1 > τ R4 > τ R9 ). This again suggests that some impurities may be accumulating in the residual melt after each ingot growth, and are then partly incorporated in subsequent ingots. We note that the effective lifetimes reported here are significantly lower than those reported by Horzel et al., [7] for example. This is due to the significantly lower resistivity of the wafers used in this study, which is representative of typical industrial Ga-doped p-type PERC cells. Note that tail section of the last RCz-Si ingot is generally recycled, and not used for solar cell fabrication. Therefore, the tail sections of the R9 ingot with solidified fraction greater than 0.7 were for recycling but were retained for the purpose of this study.

Interstitial Iron Measurements
We then measured the interstitial iron concentrations [Fe i ] present in the Ga-doped samples by using a method similar to the one used in boron-doped p-type silicon wafers. [17] First, the samples were kept in the dark for >48 h to allow the FeGa pairs to fully form, [18] and the effective lifetimes (τ dark ) were measured. Then, a flash was used to repeatedly illuminate the samples until no further change in the lifetime was observed, to fully dissociate the FeGa pairs fully. We then measured the lifetime after illumination ðτ illum ). The τ dark and τ illum values at Δn = 1 Â 10 15 cm 3 for all the samples used in this work are shown in Figure 3. The [Fe i ] is then determined by the expression, where the conversion factor (C) depends on the energy levels and capture cross sections of the Fe i and FeGa states, the doping concentration and the injection level, which was in the range of 6 Â 10 14 -6 Â 10 15 cm À3 in this work. [8] Here, we used Shockley-Read-Hall (SRH) parameters for Fe i ðE Fe i t ¼ E v þ 0.38eV, σ n = 1.3 Â 10 À14 cm 2 , and σ p = 7 Â 10 À17 cm 2 ) reported by Macdonald et al. [19] and for FeGa (E FeGa t ¼ E v þ 0.2 eV, σ n ¼ 2.5 Â 10 À14 cm 2 and σ p ¼ σ n 2 cm 2 ) reported by Schmidt and Macdonald. [12] Note, we estimated [Fe i ] based on the conventional assumption that Fe remains in a 99% FeGa state after dark storage and 99% Fe i immediately after illumination.
To confirm that the lifetime changes after illumination are indeed due to the formation of FeGa pairs, we measured the time constant of the lifetime changes in a 0.83 Ω cm wafer from the R9 ingot. The sample was illuminated for 4 min (to achieve FeGa pair dissociation), and then the effective lifetime was measured as a function of time. Fitting an exponential to this data allows us to determine a time constant of 865 AE 50 s, as shown in Figure 4a. Based on the known diffusivity of Fe i in silicon, we would expect a FeGa pair association time constant τ assoc of 799 s, from the expression, τ assoc ¼ 5.0 Â 10 5 T N A exp 0.66 kT À Á ,   where T is temperature and N A is dopant concentration. [18] The very good agreement between this value and the measured time constant is strong evidence that the changes in the effective lifetimes after light soaking are indeed caused by the formation of FeGa pairs, as opposed to other possible light-induced degradation (LID) such as light and elevated temperature-induced degradation (LeTID) [3,4,6,7] or copper-related LID [20] Figure 4b shows the estimated [Fe i ] as a function of solidified fraction in the three RCz-Si ingots. Note, we used a range of injection levels (6 Â 10 14 -6 Â 10 15 cm À3 ) instead of a single injection level to help estimate the uncertainties in [Fe i ]. The [Fe i ] detection limit for this method is in the range of 1Â 10 9 cm À3 . The measured [Fe i ] values in the range of (3 Â 10 9 -2 Â 10 10 cm À3 ) indicate that the Ga-doped RCz-Si wafers contain a small but clearly detectable amount of Fe contamination. The [Fe i ] is higher toward the tail-end of the ingot. This nonuniform axial distribution of [Fe i ] is expected due to the low effective segregation coefficient (<0.05) of Fe in the Si melt, [15] similar to the trend of the Ga distribution in Figure 1a. Further, and similarly to the dissolved oxygen concentration, the [Fe i ] values show a slight trend of higher concentrations in the subsequently pulled ingots ([Fe i ] R1 < [Fe i ] R4 < [Fe i ] R9 ), confirming increased iron accumulation in the residual melt during subsequent feedstock replenishments. However, this increase is quite small. We note that a full comparison between the ingots is somewhat inhibited by the lack of samples from higher solidified fraction values for the R1 and R4 ingots.

SRH Lifetimes Due to FeGa Pairs
Next, we analyzed the components of the measured effective lifetimes using SRH statistics to determine the recombination activity of FeGa pairs in these samples. The effective lifetime in the associated state is given as 1 where τ SRHðFeGaÞ is the SRH lifetime due to FeGa pairs (using the SRH parameters from Schmidt and Macdonald [12] ) and τ SRHðotherÞ is the SRH lifetime caused by other bulk defects present in our samples, for example, oxygen precipitates and other transition metal impurities. Figure 5 shows a comparison of the estimated τ SRHðFeGaÞ and τ SRHðotherÞ values of the as-grown samples, with the measured effective lifetimes taken after FeGa association ðτ dark ) at an excess carrier density of 1 Â 10 15 cm À3 . The τ SRHðFeGaÞ values are in all cases higher than the estimated τ SRHðotherÞ . Please refer supplementary Figure S2a, Supporting Information for an example of the injection-dependent lifetime of a sample from the ingot R9 (resistivity 0.83 Ω cm) and its fitting with the τ SRHðFeGaÞ . This demonstrates that FeGa pairs are not the dominant lifetimelimiting recombination centers in the samples. This may be due to the presence of other metallic impurities, or the formation of grown-in oxygen precipitates (OP) nuclei during the Cz-Si ingot process. It is known that OPs can act as internal gettering sites for metallic impurities, effectively increasing the precipitation of transition metal impurities (such as Fe). [21,22] Such impurity-decorated OPs and transition metal impurities (both in the dissolved and precipitated states) also act as effective recombination centers in p-type silicon. [23,24] Figure 6a shows a comparison of the measured effective lifetimes of the as-grown and TR-treated samples. The effective lifetimes in all the ingots decreased after the TR step. Note, this reduction is not due to process contamination, as the control boron-doped float-zone silicon wafer (1 Ω cm) processed in the same batch retained a very high lifetime, as shown in Figure 6a. Further, PL imaging (not shown here) of the samples reveals that the reduction is not due to the formation of oxygenrelated ring defects. [25,26] Before measuring the [Fe i ] in the TR-treated samples, we once again measured the FeGa repairing time constant τ assoc ðTRÞ in one of the TR-treated samples, as discussed above. The fitted τ assoc ðTRÞ (678 AE 60 s) again fits well with the calculated τ assoc (765 s), confirming that the response of TR-treated samples to light-soaking is due to the breaking and reforming of FeGa pairs. Refer to Figure S3, Supporting Information, for the relevant plot. Figure 6b shows the calculated [Fe i ] in all of the TR-treated samples. The [Fe i ] in the TR-treated float-zone control sample is still below the detection limit, confirming that no significant process contamination occurred. The [Fe i ] values in the TR-treated samples are significantly higher than in their as-grown counterparts. This clearly indicates that the additional Fe in the TR-treated samples comes from within the wafers, rather than from external sources such as process contamination. It is probable that the additional Fe originates from the dissolution of Fe precipitates, which were once bound to oxygen precipitates (OPs) [27] or other impurities [28] and were released during the high-temperature TR step. As a result of higher [Fe i ], the τ dark is reduced in the TRtreated samples in comparison to their as-grown counterparts. Note that FTIR measurements did not show a measurable increase in [O i ], as might be expected from the dissolution of OPs during the TR step.

Tabula Rasa
Furthermore, Figure 6c shows the comparison of the estimated τ SRH ðFeGaÞ , τ SRHðotherÞ , and the measured τ dark of Figure 5. Comparison of the measured τ dark at an excess carrier density of 1 Â 10 15 cm À3 with calculated τ SRHðFeGaÞ and τ SRHðotherÞ as a function of solidified fraction in the as-grown samples from the three RCz-Si ingots.
www.advancedsciencenews.com www.solar-rrl.com the TR samples. Contrary to the as-grown samples, τ SRHðeGaÞ is dominant in the TR samples, especially in ingots R1 and R4, and the upper section of the R9 ingot. It is also reflected in the better fitting of injection dependent lifetime of TR-treated sample with its τ SRH ðFeGaÞ , as shown in Figure S3b, Supporting Information. This demonstrates that FeGa pairs are the dominant lifetimelimiting recombination centers in the TR samples, most likely due to the dissolution of Fe that was previously bound to internal sites such as OPs. Figure 7 shows the measured τ eff (at Δn = 1 Â 10 15 cm À3 ) and iV oc (1 sun) of the samples, including the impact of gettering. The PDG step had a positive impact on the samples, as reflected in the improvement of τ eff and iV oc . However, the highest τ eff and iV oc are achieved in the TR þ PDG samples. The PDG step was performed at an intermediate temperature of 880°C, at which the possibility of OP dissolution is minimal, as the dissolution of OP requires much higher temperatures (>1000°C). [29] Further, PL imaging confirms that no ring defects were formed on any samples after the PDG step. Furthermore, we measured the τ dark and τ illum of the PDG and TR þ PDG samples but did not observe any significant difference (<10%), suggesting that the residual [Fe i ] in the gettered samples is less than the detection limit (1 Â 10 9 cm À3 ) of the method. Refer to Figure S4, Supporting Information for τ dark and τ illum . This further suggests that although the [Fe i ] showed a slight trend of a higher concentration toward the subsequently pulled ingots ( R9 ), Fe can be gettered very effectively during the PDG step. Fortunately, PDG is naturally incorporated during the n þ phosphorus diffusion on the front surface of p-type PERC cells. This obviously mitigates the detrimental impact of Fe and improves the material quality of the Ga-doped RCz-Si wafers when they are used to make solar cells. However, the presence of this Fe may cause the as-grown lifetimes to fall below specified lifetime limits at the ingot level, effectively limiting the number of saleable ingots that can be grown. In any case, other impurities and defects further limit  www.advancedsciencenews.com www.solar-rrl.com the lifetimes at the tail ends of the subsequent ingots, even after the removal of Fe by gettering.

Conclusion
We investigated the electronic quality of the wafers from three different Ga-doped ingots from the same RCz-Si ingot growth process. In the as-grown state, the [Fe i ] was determined to be in the range of 3 Â 10 9 cm À3 -2 Â 10 10 cm À3 , increasing from the seed to tail parts of the ingot due to metal accumulation in the melt. A trend of slightly higher [Fe i ] in the subsequent ingots was observed, but the [Fe i ] in the last (ninth) ingot was still well below 1 Â 10 11 cm À3 . In the as-grown state, we found that SRH recombination centers introduced by FeGa pairs are probably not the dominant recombination channel, which we hypothesized to be the presence of recombination active oxygen precipitates (OP) or other transition metal impurities. Further, after subjecting these wafers to a TR step, we observed a higher [Fe i ] (1 Â 10 10 -6 Â 10 10 cm À3 ), which could be due to the dissolution of Fe-decorated OPs. With the increased [Fe i ] in the TR-treated samples, the bulk lifetime is primarily dominated by FeGa pairs. After subjecting the samples to a phosphorus diffusion gettering steps, the negative impact of Fe contamination is reduced dramatically. Nevertheless, the effective lifetimes of the subsequently grown ingots still decrease, limiting the number of ingots that can be grown in RCz-Si growth process. We concluded that during the fabrication of PERC solar cells based on Ga-doped wafers grown by the RCz-Si process, Fe contamination is not likely to be limiting cell performance. However, other recombination centers, possibly related to OPs and other transition metals become increasingly limiting in subsequent ingots.

Experimental Section
The Ga-doped RCz-Si wafers used in this work were grown for the fabrication of conventional PERC solar cells and supplied by Longi Silicon. In the RCz-Si growth process, nine ingots of different heights were pulled. In this work, we used only three ingots that were pulled first (R1), fourth (R4), and ninth (R9) in the same RCz-Si ingot growth process.
We cut lifetime test samples from the center of M10 (182 Â 182 mm 2 ) pseudo-square wafers. All samples were saw-damage etched with tetramethylammonium hydroxide (TMAH) solution to remove 10-12 μm from each side, achieving a sample thickness of 150-160 μm, before standard RCA cleaning steps prior to further processing. Some samples were kept without further high-temperature processing and are referred to as "asgrown" samples. Other samples were subjected to a high-temperature tabula rasa (TR) annealing step. The TR step was performed at 1050°C for 30 min in oxygen and is intended to dissolve any grown-in oxygen precipitate (OP) nuclei formed during the ingot growth and cooling process. [14,30] These samples are referred to as TR samples. Next, some of the as-grown and TR-treated samples were subjected to a phosphorus diffusion gettering (PDG) step with phosphorus oxychloride (POCl 3 ) vapor diffused at 880°C for 40 min, resulting in a sheet resistance of 30-40 Ω □ À1 . These samples are referred to as "PDG" and "TR þ PDG" samples, respectively. Please refer the Figure S1, Supporting Information, for the temperature profiles of the TR and PDG steps. The diffused layers were removed by a short hydrofluoric (HF) acid dip and TMAH etch before any subsequent treatments.
The interstitial oxygen concentration [O i ] was measured on some samples by Fourier-transform infrared spectroscopy (FTIR), using a Bruker Vertex 80 tool and calibrated using the ASTM F121-83 standard with a calibration coefficient of 2.45 Â 10 17 cm À2 at room temperature. Any thermal oxide layers formed during annealing steps were removed by HF before the FTIR measurements.
For effective carrier lifetime measurements, both surfaces were passivated with 20 nm aluminum oxide (AlO x ) layers via a thermal atomic layer deposition (ALD) (Beneq). After the deposition of AlO x , the passivation was activated with a 30 min forming gas anneal (FGA) at 450°C. Effective carrier lifetimes, implied open circuit voltages (iV oc ) at 1 sun light intensity, and wafer resistivities were measured using the quasi-steady state photoconductance and transient photoconductance decay techniques with a WCT-120 tool from Sinton Instruments. [31] PL images were captured using an LIS-R1 PL imaging tool from BT imaging. [32] The surface recombination velocity (S) attributable to the AlO x passivation layers was calculated based on a high-quality p-type (boron-doped) float-zone sample (resistivity of 1 Ω cm, thickness of 180 μm, and effective lifetime of 960 μs at an excess carrier density (Δn) of 1 Â 10 15 cm À3 ), via the expression, 1 where τ intrinsic is the intrinsic lifetime determined using Niewelt's model, [33] and W is the sample thickness. In this way, the S value was estimated to be 7.5 cm s À1 .

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.