Abstract
The dependence of surface recombination of boron diffused and undiffused silicon surfaces passivated with a-Si:H on the net charge density is investigated in detail. The films are deposited by plasma-enhanced chemical vapour deposition using a 2.45 GHz microwave remote plasma system. The surface charge density on the samples is varied by depositing charge using a corona discharge chamber. Excess carrier lifetime, capacitance-voltage, and Kelvin probe measurements are combined to determine the surface recombination velocity and emitter saturation current density as a function of net charge density. Our results show that the application of negative charge causes a substantial reduction in the surface recombination of samples with boron diffused emitters, even for high boron surface concentrations of . The significant difference observed in surface recombination between boron diffused and undiffused sample under accumulation implies that the presence of boron diffusion has results in some degradation of the Si-Si interface. Further, (111) oriented surfaces appear more sensitive to the boron surface concentration than (100) oriented surfaces.
1. Introduction
The incorporation of silicon nitride (SiNx) films into semiconductor devices, and in particularly solar cells, has become an increasingly important topic. This is partly due to their favorable characteristics, which allow them to act as antireflection coatings as well as passivation layers. For the case of PECVD deposited nitride, the film properties can be varied over a broad range, and the films contain a high concentration of hydrogen atoms that can be used-to-passivate bulk defects. Given the growing interest in solar cells based on n-type silicon substrates [1], the passivation of p-type emitters has become more important and, therefore, the application of PECVD SiNx films on this type of emitter has emerged as a topic of interest as well.
PECVD SiNx generally contains a moderate to high density of positive fixed charge located at or near the Si-SiNx interface. This positive charge is beneficial for the passivation of lowly doped or n doped surfaces but is generally undesirable on heavily doped p type surfaces, such as boron diffused emitters, as the positive charge results in a reduction in hole concentration at the Si surface and consequently an increase in surface recombination velocity. While recent studies have demonstrated that PECVD nitride could actually provide excellent passivation of diffused emitter structures when an appropriate postannealing process was carried out [2], the paper did not provide details of the charge density in the SiNx film, or the surface concentration. Hence, the reason for the excellent reported surface passivation has not been clearly established. Moreover, the exact dependence of the surface recombination rate on the combination of charge density in the PECVD nitride film and the p+ diffused emitter profile has not been explored in detail. Neither does much information exist on whether the presence of a diffusion results in significant changes to the Si-SiNx interface region (such as the creation of more interface defects). This paper aims to investigate these issues.
2. Impact of Charge on Surface Recombination
The surface recombination velocity is defined by the following expression: where is the recombination rate at the surface and is the excess surface carrier concentration. Using the Shockley-Read-Hall (SRH) equation, can be expressed as [3]where is the thermal velocity, and are the electron and hole concentration under equilibrium, is the intrinsic carrier concentration, and are the electron and hole concentrations at the surface, is the interface defects energy level, is the energy-dependent interface defects density, and and are the energy-dependent for electrons and holes, respectively. Examination of (2) shows that, while is in general a function of can be considered to be independent of under low level injection conditions at the surface. Furthermore, for high surface majority carrier concentrations (1018 cm-3), is expected to be only weakly dependent on if the interface properties do not change with
Surface recombination can be minimized by reducing the concentration of one type of carrier at the surface, either electrons or holes. This can be achieved by creating either accumulation or strong inversion conditions at the surface. Further examination of the SRH equation shows that, for defects with energy levels that are not close to the band edges and with a capture cross section ratio , the peak in recombination will occur when the carrier concentration at the surface is where and are the surface hole and electron concentrations, respectively [4]. For approximately equal capture cross sections, the peak will occur when that is, when the surface is in depletion. Surface recombination as a function of charge density can be modelled using the extended SRH formalism developed by Girisch et al. [5] and Aberle et al. [6]. An example of this model is given in Figure 1, showing the effect of surface charge upon surface recombination rate for samples with different surface doping concentrations This model predicts that, under strong inversion or accumulation, surface recombination will become independent of the surface doping, and relatively weakly dependent on charge density. This suggests that a comparison of samples under deep inversion or accumulation conditions can be used to directly compare the interface quality (in terms of interface defect density distribution and capture cross sections) of different samples.
Investigations of surface recombination as a function of surface charge density can be performed by applying a gate voltage on a suitable metal-insulator-semiconductor (MIS) structure, or by depositing corona charges on the surfaces of samples. Several such investigations have been carried out for the Si-SiO2 interface [7–10], where generally good qualitative agreement between experimental and modelled results has been obtained. Few publications report on similar studies on the Si-PECVD SiNx system, but the results are similar to those obtained on Si-SiO2 [11]. An important and consistently observed feature in all the experimental results is the saturation (rather than a continued reduction, as predicted by models) of surface recombination for sufficiently large densities of positive and negative charge. For Si-SiO2, saturation occurs at a net charge density of less than 012 cm-2, or a majority carrier concentration of 019 cm-3 [4, 10, 12]. The majority carrier concentration here means the concentration of the majority carrier taking into account the effect of the application of surface charge. For example, for a diffused surface with a very high positive charge density, the majority carriers would be electrons.
3. Experimental
Samples used for lifetime measurements were Float Zoned (FZ), n-type, , 40–60 ohm.cm, 500 um thick -Si wafers. These samples were etched (removing at least 15 m per side) in HNO3 and HF solution to remove any saw damage. After receiving a standard RCA clean, selected samples were subjected to a diffusion using a liquid BBr3 source in two different groups to form a symmetrical structure of for lifetime measurement. The emitter profiles were determined using spreading resistance and are shown in Figure 2. The surface concentrations obtained after deglaze for groups A and B are approximately 019 cm-3 and 018 cm-3 with corresponding sheet resistances of 200 and 150 respectively. All samples received a clean HF dip prior to nitride depositions.
All samples received subsequent silicon nitride depositions on both sides, one side at a time. The film deposition was performed in Roth & Rau high-frequency microwave remote PECVD systems. For most samples, the deposition process was carried out at a temperature of 450°C, a processing pressure of 0.2 mbar, a microwave frequency of 2.45 GHz, a total microwave power of 1000 W, and SiH4 to NH3 gas mixture ratio of 1 : 1. This process gave 70 nm thick films with a refractive index of nearly 2. For some samples, a slightly different set of deposition parameters was used, with a deposition temperature of 400°C, a microwave power of 750 W and a slightly lower SiH4 to NH3 ratio, resulting in 90 nm thick films with an almost identical refractive index.
The samples were characterized by effective lifetime measurements determined from photoconductance measurements in high level injection. The emitter saturation current density was extracted from the slope of the Auger corrected inverse lifetime as a function of injection level [13]. This was followed by surface charge deposition using a corona discharge chamber. The variation in the density of the deposited charge was 20%. Kelvin probe measurements were performed immediately following charge deposition to monitor the surface potential change and thus allow calculation of deposited charge density at the surface. The surface charge density was subsequently calculated from the measured surface potential change, using appropriate values for the dielectric constants determined separately from capacitance-voltage measurements.
Czochralski (Cz), n-type, , 2–9 ohmcm, 500 um thick -Si wafers were used for Capacitance-Voltage (C-V) measurements. Theses samples received a standard saw damage etch, RCA clean, and clean HF dip prior to PECVD nitride deposition. Nitride films with thickness around 100 nm were subsequently deposited on one polished side of the wafer, followed by Al metal dots (with diameter 700 um) evaporated on top of the surface to form a Metal-Insulator-Semiconductor structure. Gallium-indium paste was applied on the rear side to allow electrical contact to Si bulk. Quasistatic (QS) and high-frequency (HF) C-V measurements were carried out.
4. Results and Discussion
4.1. Charge Density
Figure 3 shows typical quasi-static and high frequency C-V curves of an as-deposited nitride film. The curves shown were measured with two different sweep directions. In addition to some hysteresis which is commonly observed, the shape of the curves shown is significantly different for the two sweep directions. A possible explanation for this unusual behavior could be the presence of interface defects with relatively small capture cross sections, and additionally with As the Fermi level is swept from near the valence band to the conduction band edge, filling of the traps with electrons is sufficiently rapid to result in high QS capacitances and a stretch-out of the HF curve. However, as the Fermi level is swept in the other direction, emission from the traps is much slower, resulting in the modified shape of the curve. These traps are mostly removed following an anneal in at 400°C, resulting in the curves shown in Figure 4.
(a)
(b)
The values of the accumulation capacitance (equal to the insulator capacitance ) and the flat band voltage allow determination of the charge density in the as-deposited and annealed films, using: where A is the capacitor area (0.0039 cm2), is the Al-silicon work function difference, is the work function of Al (4.1 eV), is the electron affinity of Si (4.05 eV), is the electronic charge, and is the difference between the conduction band edge and fermi level in the material used (0.24 eV). Given the flat band voltage obtained from C-V data, the fixed charge density within the nitride films used in this paper was calculated using (3). The results are summarized in Table 1. As can be seen, both the as-deposited and annealed films display a relatively low-charge density.
4.2. The Effect of a Boron Diffused Surface
Figure 5 shows curves of Auger corrected inverse lifetime as a function of excess carrier density for a 40–60 Ωcm, planar, n-type undiffused sample following PECVD nitride deposition, as well as after subsequent corona charging. Following the deposition of high negative or positive charge densities, the surface is expected to be in low level injection. The net charge density under these conditions is estimated to be 4.012 and +5.012 cm-2. These net charge densities were obtained from the sum of (from C-V measurements) and the applied surface charge density obtained from the change in surface potential following corona charging as measured using a Kelvin probe. However, the actual value of net charge density may be overestimated by this method, particularly for large This is because the application of large charge densities on the surface of the nitride can induce the injection of the opposite type of carrier into the nitride film. As a result of such injection, it is observed in Figure 5 that the removal of negative surface charge by rinsing in isopropyl alcohol (IPA) results in a lifetime curve which approaches that following positive corona charging (due to the injection and storage of positive charge in the nitride) while the removal of positive surface charge results in a lifetime curve which approaches that following negative corona charging (due to the injection and storage of negative charge in the nitride).
The emitter saturation current density can be determined from the slope of the Auger-corrected inverse lifetime curves. As shown in Figure 5, both after negative and positive charging, the Auger corrected inverse lifetime plots yield reasonably straight lines at high injection levels, as would be expected for a sample in which the surface is in low-level injection while the bulk is in high-level injection. Nevertheless, following negative corona charging, an unusual feature was observed. While extracted from the slope of inverse lifetime is very low, (in fact, following negative corona charge application is significantly lower than following positive charge application) the effective lifetime is significantly lower as well. This unusual behaviour cannot be explained by recombination in the depletion region created by the surface inversion, since, under conditions of high level injection, no depletion region exists in the samples.
In order to analyse the results from minority carrier lifetime measurements, two methods were used to extract over the range of applied charge densities of the same sample as in Figure 5. The first is to determine from the measurement of : where is the excess carrier density in the sample bulk and is the background doping concentration. was determined at an injection level 015 cm-3 and thus, is small compared to The determination of from is not expected to be accurate over the entire range of applied charge densities. The condition for valid measurements—that the surface must be in low level injection—will not be achieved for low-net charge densities, (less than about ±012 cm-2). However, for large densities of either negative or positive charge, the presence of an accumulation or inversion layer should result in a meaningful extraction of
The second method is to use the relationship between bulk lifetime, effective (measured) lifetime, and surface recombination [14] which is given by: where is expressed by By combining (5) and (6), can be calculated by Here, is the ambipolar diffusion constant (cm2/s), is wafer thickness, and is the bulk lifetime. This equation assumed that all recombination can be attributed to either the bulk or the surface region, and that no other recombination source (e.g., depletion recombination) is present. Equation (7) requires determination of We estimate for each sample using (8) in which is calculated with (4) using the highest which was obtained under accumulation conditions, with positive applied charge.
Equation (7) is used to calculate as a function of rather than simpler and more commonly used expression in (8). This is done since, for several of the samples analysed, the relatively high-surface recombination velocities lead to a significant (>10%) underestimation of if (8) is used.
As shown in Figure 6, it is clear that the application of negative charge leads to significant discrepancy between the two calculations due to the opposing trends of decreasing and decreasing , which cannot be reconciled with a simple model of a sample with a uniform bulk and surfaces in low level injection. Similar results have been observed in a range of PECVD nitride passivated, undiffused samples deposited using different PECVD systems and on different substrates, including both lightly and moderately doped, type and n type substrates. It is confined to undiffused substrates with a PECVD nitride layer, following the deposition of negative charge. The reason for this phenomenon is currently not well understood and will require further investigation. In this paper, we will not concern ourselves further with measurements on undiffused substrates following the application of negative charge. In the subsequent discussions, (6) will be used for extraction, since this method does not require the surface to be in low level injection.
The variation of with net charge density for a diffused sample with a high surface concentration is depicted in Figure 7. Negative charges were initially deposited, followed by charge removal in IPA solution and deposition of positive charge (in according with the sequence numbers given on the curve). The starting point (point 1), prior to any charging, is given by (+5.011 cm-2). The results of Figure 6 show a reduction and eventual saturation of as the surface is pushed into accumulation. This trend is consistent with other reports in the literature [8–12]. We applied the extended Girisch model [5, 6] to simulate the behaviour. Figure 7 shows that a reasonably good fit to the data can be obtained when a suitable set of modelling parameters is chosen. The distribution of electrons and holes capture cross sections and the density of dopant-like defects over the bandgap energy that were used in this model are given in Figure 8. However, it is not possible to make definitive statements concerning the variation of defect density or capture cross sections from the available data. As mentioned before, the Girisch model predicts a continuing decrease in as the negative charge density is increased, while in practice, saturation of is observed.
It is interesting to note from the experimental results shown in Figure 7 that, even for a very heavily doped surface, the application of charge can still result in a significant further reduction in surface recombination. Previous studies [10, 12] on undiffused surfaces have found that surface recombination tends to saturate for applied surface charge densities of 012 cm-2 or less, with corresponding surface majority carrier concentrations of 019 cm-3 or less. For the above sample, is observed to saturate for charge densities of 5 1012 cm2, but the corresponding surface majority carrier concentrations are significantly higher than 019 cm-3. The main conclusion that can be drawn from Figure 7 is that even heavily doped surfaces are sensitive to the presence of charge.
The best of the diffused sample shown in Figure 7 following negative charge application is approximately 41 fA/cm2. This can be compared to a of ~6 fA/cm2 for an identically prepared, undiffused sample under accumulation. The difference in values is partly due to the contribution to from the emitter region so this factor should be accounted. Results on oxide passivated diffused emitters with similar sheet resistances have shown that this kind of emitter contributes less than 6 fA/cm2 [15]. The remainder of the difference shown in this experiment (~29 fA/cm2) is likely to indicate that the diffused interface is not as good as that of undiffused (i.e., has a higher interface defect density, or defects with greater capture cross sections), as has also been observed for samples passivated with a thermal oxide [16].
Figure 9 shows the extractions for a diffused sample with nitride film deposited at 450°C before and after anneals at a temperature of 400°C for 30 minutes. These anneals were performed in . This was followed by an annealing in FGA on the same sample. Annealing in was able to improve the interface quality of both diffused and undiffused surfaces, as indicated by lower regardless of the charge density on the surface. This improvement is in good agreement with the results shown in Reference [2]. Furthermore, it is also shown by the results given in Figure 9 that the introduction of H molecules did not contribute to the passivation, since there is no further improvement shown following second annealing in forming gas. Therefore, it is suggested that the improved interface quality is more likely due to rearrangements and the removal of some defects at the interface at elevated temperature, as well as possibly the passivation of interface defects with hydrogen supplied by the SiNx film.
4.3. The Effect of Surface Doping and Surface Orientation
The effect of surface doping level on different surface orientations as a function of net charge density is plotted in Figure 10. These samples received anneals in at 400oC.
At low charge density where = 0, for both and samples is dependent on surface doping concentration, with a lower surface concentration resulting in a higher due to a lower hole concentration (and hence a higher minority carrier electron concentration), which results in higher recombination. For the case of , the values for a high applied density of negative charge are almost identical; suggesting that the difference in surface concentration has not significantly affected the interface defect properties, such as their density, distribution, or capture cross sections. Given that the presence of a diffusion clearly degrades the interface, it may be that the extent of this degradation is more closely related to the total concentration rather than concentration at the surface. However, further work will be required to verify this.
For the samples, the dependence of on surface concentration under no charge application is much less pronounced. In contrast to the surface, the final accumulated values of samples are found to be dependent on surface concentration, with a higher surface concentration resulting in higher saturated values. This result suggests that the interface properties of the surface are more susceptible to surface concentration than the surface. For low surface concentrations, there appears to be no significant difference in the passivation quality of and surfaces, suggesting that, for lowly and moderately doped surfaces, the PECVD SiNx passivation is rather independent of surface orientation.
5. Conclusion
In conclusion, we have demonstrated that the application of positive surface charge results in a significant increase in the surface recombination of PECVD SiNx passivated diffused emitters, even for high surface concentrations. On the other hand, negative charge causes a substantial reduction in surface recombination. Under accumulation, the significant difference in between diffused and undiffused sample, as much as ~29 fA/cm2, implies that the diffusion has resulted in some degradation of the Si-PECVD SiNx interface. The very similar passivation quality of samples with different concentrations at the surface suggests that the degradation is not strongly related to the surface concentration. However, the interface properties of surfaces tend to be more susceptible to the concentration at the surface rather than .
There are several possible implications for solar cell design. If PECVD SiNx films were to be applied to the passivation of diffused emitters, then minimising the positive charge density in the SiNx film will be very important in order to obtain good surface passivation, even for a high surface concentration. The results suggest that the surface concentration should be rather high, even though—for some surfaces at least—a higher concentration will lead to a poorer interface. It is likely that textures that avoid the formation of surfaces will perform better than the random pyramid texture.
Acknowledgments
The authors wish to thank members of the PV group at the University of New South Wales for carrying out some of the PECVD depositions; Dr. Jason Tan and Mr. Chun Zhang for help and discussion on PECVD nitride depositions; and Mr. Simeon Baker-Finch and Dr. Keith McIntosh for discussions on emitter recombination. Financial support by the Australian Agency for International Development is gratefully acknowledged.