Light-Induced Lifetime Degradation in Multicrystalline Silicon

The correlation between interstitial oxygen content and light-induced lifetime degradation in cast multicrystalline silicon is complex. On a wafer-averaged scale, there is a strong positive correlation, which has been parameterized in this paper to model the impact of this degradation on cell voltage for a typical industrial process. However, on a local, grain to grain scale within a given wafer, the degradation can vary by an order of magnitude, while the interstitial oxygen content remains almost unchanged. This supports recent suggestions that Oi is not directly involved in the chemical composition of the defect. Light-induced degradation is a well-known phenomenon in boron-doped Czochralski (Cz) silicon [1], reducing the carrier lifetime and hence the cell efficiency. It has been linked to the presence of both boron and interstitial oxygen Oi, the latter being relatively abundant in standard Cz material. Degradation occurring under illumination of around 1 sun intensity takes approximately 24 hours to reach saturation. Since carriers injected through biasing also cause this degradation, the effect has become more properly known as “carrier-induced degradation” (CID) [2], as it is referred to in this paper from this point on. The initial non-degraded state can be recovered by annealing the sample above 200°C. The extent of CID is characterized by the relative recombination center density Nt, which is defined as the difference of the inverse lifetimes before (τb) and after (τa) degradation [1]: b a t N τ τ 1 1 − = ∗ In Cz-Si, this parameter has a clear linear dependence on the boron concentration NA, implicating the presence of B atoms in the defect composition. It also shows a dependence on the value of [O i], but usually with much more scatter. This indicates that Oi itself may not to be directly involved in the defect, but occurs in association with it. Recently, the oxygen dimer [3] and intrinsic point defects [4] have been suggested as possible alternatives. Similar degradation has been reported in cast multicrystalline silicon (mc-Si) [5,6]. The effect in mc-Si has been shown to be reversible upon annealing, as it is with Cz-Si, and the rate at which the degradation occurs is similar in both materials. These facts suggest that the defect is indeed the same. In this paper, we examine the relationship between the degradation in mc-Si and the interstitial oxygen content, on both a macroscopic level (averaged across a wafer), and from grain to grain. The results allow some insights into the possible composition of the defects. The determined dependence of the average degradation on [Oi] also allows cell performance to be modeled, and criteria for acceptable [Oi] levels in ingots to be established. Fig 1 shows the interstitial oxygen profiles measured by Fourier-Transform Infrared spectroscopy (FTIR) for 3 mc-Si ingots from different manufacturers. The concentration at the bottom of ingot 3 is high even by Cz-Si standards. Fig 2 shows the effect of illumination on the effective lifetime for wafers from these ingots, as well as a Cz wafer ([O i]=10ppma) and a Float-zone (FZ) wafer ([Oi]<1ppma). The lifetimes represent area-averaged values of several grains, measured with the QSSPC technique at an excess carrier density of 3×10 cm. The FZ wafer showed no degradation, due to its low interstitial oxygen content, and confirms that the drop in lifetime observed in the other samples is not caused by changes in the surface passivating film (PECVD SiN), or by sample heating during illumination. After 150 minutes illumination the samples were annealed at 250°C for 15 minutes, which caused the lifetimes to recover. Note also that the rate of lifetime decay is similar for the mc-Si and Cz samples, confirming that the defects are most likely the same. The two mc-Si wafers with the lowest lifetimes were sister wafers from the same ingot. One was measured without processing (other than etching, cleaning and surface passivating PECVD SiN deposition), and the other after emitter diffusion and hydrogenation by spike firing a SiN film, in similar fashion to industrial processing. The extent of CID is almost identical in these two wafers, suggesting that neither gettering nor bulk hydrogenation significantly reduce CID in this material. A further point of interest in Fig 2 is the sudden increase in lifetime of the mc-Si wafers within the first few minutes. This effect is almost completely absent in the FZ and Cz wafers. By analyzing the decay behavior of this initial change we have confirmed that it is caused by the splitting of FeB pairs [7]. At the injection level measured, FeB pairs are more strongly recombining than interstitial Fe, and so the lifetime increases after breaking the FeB pairs. After degradation for 150 minutes, the sample was allowed to relax in the dark for 24 hours, long enough for the pairs to re-form, as shown in Fig 3. The lifetime could then be recovered again after 5 minutes further illumination. Fortunately, the very different time constants of the FeB splitting and CID allow them to be studied independently. By the use of lifetime mapping techniques, such as Carrier Density Imaging (CDI) [8] and Modulated Free-Carrier Absorption (MFCA), it is possible to study the degradation on a local scale, as shown in Fig 4. The first two maps show CDI images of the carrier lifetime before and after degradation for 16 hours. The third map shows the local Nt calculated using the expression above. The CDI images took approximately 3 min to capture, meaning that the FeB pairs were fully dissociated for almost the entire measurement time (dissociation is complete within tens of seconds in most cases). In contrast to the QSSPC technique, both CDI and MFCA operate under fixed generation (equivalent to 1 sun in these measurements). This means that wafers or grains with higher lifetimes will be in higher injection, which will produce a greater apparent value for Nt, as has been shown by Rein et al. [9]. We estimate that these effects will produce an overestimation of Nt in the high lifetime regions of up to 50% for the conditions in this study. The overestimation would be less for lower lifetime grains and wafers. Unfortunately, this distortion is difficult to avoid when using mapping techniques. The image clearly shows a large variation in Nt from grain to grain. Some high lifetime grains remained almost unchanged, while others degraded strongly. This is indicated on the left of Fig 5, which reveals little correlation between Nt and the initial lifetime. After degradation (right of Fig 5), all points lie below the limit imposed by the CID in parallel with a lifetime cap of 100μs due to surface recombination. Many points lie well away from this curve, meaning other bulk recombination centers, such as Fe, are also important in these regions. 0 20 40 60 80 100 0 5 10 15 20 25