Introduction

Passivated emitter and rear cells (PERCs) have emerged as a promising technology for both high efficiency and competitive cost in recent years. The most difference between the PERC and the traditional full-aluminum back surface field silicon solar cell is rear passivation of wafers. Considerable efforts have been made in order to improve wafer surface passivation. Minority carrier lifetimes of 0.8–8 ms have been reported for p-type floating zone wafers passivated by vacuum [1,2,3,4] or spatial atomic layer deposition (ALD) aluminum oxide (Al2O3) [5,6,7]. The passivation quality for p-type Czochralski wafers is lower, in the range of 0.1–2 ms [8, 9]. Spatial ALD Al2O3 have been extensively studied and applied to the industry in recent years due to their higher deposition rate (0.03–1.2 nm/s) compared to that of a conventional vacuum-type ALD (< 0.03 nm/s) [10, 11]. Trimethylaluminum (TMA) and H2O are the most widely used precursors as they are inexpensively volatile liquid and easy to handle. Some research groups use other precursors such as AlCl3 or O3 [12,13,14]. Al2O3 is currently considered to be the best passivation material due to its field effect and chemical passivation [15]. It is found that the H2O-based ALD process mostly leads to a silicon oxide (SiOx) layer at the Al2O3/Si interface, and this interfacial layer can appear after deposition or annealing [16]. Post annealing for Al2O3 films in either nitrogen or forming gas (FG) has been shown to significantly increase the wafer lifetime [12, 17]. Hydrogen in FG or Al2O3 cause hydrogenation of Si interface during annealing. The annealing temperature is typically below 500 °C, beyond which dehydrogenation occurs. However, other annealing processes for further improving passivation quality are rarely reported.

In this study, Al2O3 films are prepared on Si by spatial ALD with TMA and H2O as precursors. Effects of oxygen (O2) and FG post annealing on passivation of Si wafers are investigated and analyzed. A two-step annealing as a combination of O2 and FG annealing is proposed and demonstrates a higher wafer lifetime compared to the individual gas annealing process. Finally, photovoltaic performance of PERCs fabricated with industry standard, O2, FG, and two-step annealing are presented.

Methods

P-type (100) Czochralski silicon wafers with resistivity of 1 Ω-cm and thickness of 200 μm were used as substrates. The wafers were cleaned using standard RCA process, followed by a 30-s HF dip to remove native oxide on the wafers. The Al2O3 thin films with a thickness of 18 nm were deposited using a spatial ALD system, with H2O and TMA as oxidant and aluminum source, respectively. The gap between gas injection heads and the movable substrate holder was about 1 mm. The detailed deposition parameters are summarized in Table 1. The temperature of the pipes was 70 °C to prevent condensation of precursors. Some of the wafers were passivated with silicon nitride (SiNx, 120 nm)/Al2O3 (18 nm) stack, where the SiNx layer was deposited using a 13.56-MHz inductively coupled plasma vapor deposition at 120 °C with a gas mixture of ammonia (NH3) and tetramethylsilane (TMS). Other parameters for SiNx deposition are listed in Table 2. The oxygen, FG, or two-step annealing process was performed on the samples, and the annealing parameters are listed in Table 3. The lifetime of the samples was measured by Sinton WCT-120. The capacitance-voltage (C-V) measurement was carried out on metal-oxide-semiconductor (MOS) samples by a capacitor meter (HP 4284a) at 1 MHz at room temperature. For MOS fabrication, the wafers were deposited with an 18-nm-thick Al2O3 layer and annealed. Aluminum films with a thickness of 500 nm were evaporated on both sides of the samples as electrodes. The area of the MOS samples was 1 mm2. The cross-sectional images of the samples were obtained using a transmission electron microscope (TEM). For PERC fabrication, a schematic of the devices is shown in Fig. 1, where the ALD passivation is only on the rear side. The wafers were textured using alkaline solution to generate random pyramids. Emitter was formed by POCl3 diffusion in a standard tube thermal furnace with a sheet resistance of 100 ohms/square. A SiNx of 85 nm thickness was deposited on the front side of the wafer as an antireflective layer by inductively coupled plasma vapor deposition (ICPCVD). The back side of the wafer was polished by KOH solution for 3 min at 70 °C. The Al2O3 films of 18 nm in thickness were deposited using spatial ALD. An ICPCVD SiNx of 120 nm in thickness was deposited on Al2O3. The samples were annealed with different annealing processes. The rear local openings with a diameter of 40 μm and a pitch of 260 μm were created by 532-nm laser scribing. Finally, a silver grid was screen printed on the front and aluminum on the rear dielectric, followed by co-firing at a peak temperature of 850 °C. The current density-voltage (J-V) curves were measured by a dual light source-type solar simulator (Wacom Co., Japan) using both xenon lamp and halogen lamp with a calibrated class A AM 1.5G simulated light spectrum.

Table 1 Deposition parameters of the ALD Al2O3 layer
Full size table
Table 2 Deposition parameters of the SiNx layer
Full size table
Table 3 Parameters of O2, FG, and two-step annealing processes
Full size table
Fig. 1
figure 1

Schematic of PERC solar cells with SiNx/ALD Al2O3 rear passivation

Full size image

Results and Discussion

Figure 2a shows the injection-level-dependent minority carrier lifetimes of the Al2O3/Si/Al2O3 samples without and with different annealing processes. Before annealing, the minority carrier lifetime is low as below 100 μs over the whole injection level range. The lifetime greatly improves after the annealing process as a consequence of chemical passivation and field effect passivation brought by annealed Al2O3. However, the lifetime values are different in these three annealing conditions, in which oxygen annealing has the lowest curve, FG annealing has the intermediate, and the two-step annealing has the highest. The lifetime values at the injection level of 3 × 1015 cm−3 are extracted as shown in Fig. 2b. The O2-, FG-, and two-step-annealed samples have lifetimes of 818, 934, and 1098 μs, respectively. Note that the two-step annealing can obtain the highest lifetime only with the annealing sequence of the first step in O2 and the second step in FG. The reverse sequence results in a lifetime similar to that of the sample with O2 annealing alone. This might be because if FG annealing was performed first, the following O2 annealing might cause dehydrogenation. Niwano et al. reported that for a wafer terminated by Si–H or Si–H2 bonds, exposure to oxygen results in the replacement of the hydrogen bonds with the Si–O bonds [18].

Fig. 2
figure 2

a Injection-level-dependent minority carrier lifetime. b Lifetime at an injection level of 3 × 1015 cm−3 for Al2O3/Si/Al2O3 samples with O2, FG, and two-step annealing

Full size image

As overall passivation is governed by field effect and chemical passivation, the C-V measurement is helpful to clarify which passivation dominates in the cases of O2, FG, and two-step annealing. Figure 3a shows the normalized C-V curves for the samples without and with different annealing processes. The slope magnitude of the curves in the depletion region can be used as an indicator of interface defect density (Dit), since the existence of interface traps causes C-V curve stretch-out [19]. The two-step annealing gives the largest slope among the others, and thus the lowest Dit is expected. To gain further information, the values of fixed oxide charge density (Qf) and Dit are extracted from the C-V curves as plotted in Fig. 3b. The Qf is helpful for evaluating the field effect passivation and is calculated by [20]

$$ {Q}_f=\frac{C_{\mathrm{ox}}\left({W}_{\mathrm{ms}}-{V}_{\mathrm{fb}}\right)}{q\ A} $$
(1)
Fig. 3
figure 3

a Normalized C-V curves. b Dit and Qf for samples with O2, FG, and two-step annealing

Full size image

where Cox is the accumulation oxide capacitance, Wms is the work function difference between semiconductor and electrode (in this case − 0.9 V), Vfb is the flat band voltage, q is the electron charge, and A is the area of the MOS devices. The Qf is − 3.2 × 10−11 cm−2 for the as-deposited sample. Qf at this level leads to weak field effect passivation [21]. All the annealed samples elevate Qf to the level of 1012 cm−2. It is seen that the O2 annealing gives the highest Qf of 3.9 × 1012 cm−2, the two-step annealing gives the intermediate Qf, and the FG annealing gives the lowest Qf. On the other hand, Dit value estimated by the Terman method [22] is also shown to evaluate chemical passivation. The as-deposited sample has a Dit of more than 1013 eV−1 cm−2. It reduces to 5.4 × 1011 eV−1 cm−2 for O2 annealing, 3.7 × 1011 eV−1 cm−2 for FG annealing, and 3.1 × 1011 eV−1 cm−2 for two-step annealing. Thus, by comparing O2 and FG annealing, it is found that O2 annealing has the better field effect passivation, whereas FG has the better chemical passivation. The former might be linked to the interfacial SiOx growth. Unlike FG annealing which is performed at a relatively low temperature and with lack of oxygen, O2 annealing is expected to have an improved SiOx interfacial layer growth. This could increase the possibility of Al substitution for Si at the Al2O3/SiO2 interface, which is regarded to be one possible origin of negative fixed charges [23]. Considering the two-step annealing, the intermediate Qf is reasonable as a combination of O2 and FG annealing. However, its Dit value is lower than that of the FG annealing. This is explained by the additional contribution by the higher quality of the interfacial oxide layer due to the first-step O2 annealing. Some studies also reported that a denser SiOx results in a better passivation [24]. The lower Dit in two-step annealing sample can also be attributed to the hydrogenation improvement of silicon surface induced by the hydrogen in Al2O3 film.

Figure 4 shows the cross-sectional TEM images of the samples without and with different annealing processes. Before annealing, a SiOx interfacial layer between Si and Al2O3 is observed although the interface is not clear. This might be because H2O was used in the first-half ALD cycle. For O2 annealing, the interfacial layer thickness increases to 5.6 nm, due to annealing at a high temperature (600 °C) and in oxygen ambient. It has been reported that oxygen has a very small diffusion coefficient in Al2O3 (~ 10−38 cm−1 at 600 °C) [25], and thus, it is unlikely for oxygen to diffuse through the Al2O3 layer to reach the Si interface. Instead, ambient oxygen interchanges with the oxygen in Al2O3, creating a mobile oxygen that can repeat the interchange process in the deeper Al2O3 region until the oxygen reaches the Si interface [26]. The sample annealed in FG shows a clearer interface with a very thin SiOx interfacial layer of 1.4 nm, which is similar to other research groups performing the annealing process in N2 or FG [16]. This evidences that FG annealing limits the interfacial layer growth. The two-step annealing shows an intermediate SiOx interfacial layer thickness of about 4 nm, as a consequence of the reduced time of the O2 annealing.

Fig. 4
figure 4

Cross-sectional TEM images for samples a without annealing and with b O2, c FG, and d two-step annealing

Full size image

Figure 5a shows the injection level-dependent minority carrier lifetime of the SiNx/Al2O3-passivated wafers without and with different annealing processes. The lifetimes at the injection level of 3 × 1015 cm−3 are 1569, 1579, and 2072 μs for O2, FG, and two-step annealing, respectively. The improvements are related to that the plasma chemical vapor-deposited SiNx films may contain certain amounts of hydrogen depending on the deposition process parameters. During the annealing process, some of the hydrogen would move towards the Si interface, and this enhances the Si interface hydrogenation [27]. As reported in literature [6, 28,29,30], the lifetime of SiNx/Al2O3-passivated p-type CZ wafers is in the range of 0.1–2 ms. The optimal temperature of post-deposition annealing either in nitrogen or in FG is around 400–500 °C. In this work, the SiNx/Al2O3-passivated CZ wafer annealed in FG shows a lifetime of 1579 μs and an optimal annealing temperature of 450 °C, which are in accordance with the reported values. However, this optimal temperature is limited by the hydrogenation of the silicon interface. From the viewpoint of the silicon oxide interfacial layer, this layer might have different optimal temperature as high temperatures generally improve qualities of silicon oxide films. Thus, the two-step annealing could optimize both of the interfacial oxide quality and silicon interface hydrogenation, and leads to a higher lifetime of 2072 μs compared to the case of forming gas single-step annealing. To investigate the reproducibility, 50 samples with two-step annealing were prepared, and their minority carrier lifetime is shown in Fig. 5b. The samples have lifetime values ranging between 1939 and 2224 μs. The average value is 2075 μs, and the error is within ± 7%. The intrinsic lifetime limit of the wafer used in this study is about 2300 μs, calculated by using the Richter parameterization [31]. Thus, the two-step annealing yields a lifetime close to the lifetime limitation and demonstrates excellent interface passivation. For other ALD, a silicon oxide interfacial layer between Al2O3/Si is also found, and the two-step annealing should be able to improve the passivation quality of Si wafers. AlOx/SiNx is necessary as the silicon nitride not only enhances passivation but also increases rear reflectance and protects AlOx from a high-temperature cofiring process for PERC fabrication.

Fig. 5
figure 5

a Injection-level-dependent minority carrier lifetime of SiNx/Al2O3-passivated samples with O2, FG, and two-step annealing. b Lifetime at an injection level of 3 × 1015 cm−3 for 50 samples with two-step annealing

Full size image

Figure 6 shows the implied open-circuit voltage (Voc) for the SiNx/Al2O3-passivated samples with different annealing processes. For p-type wafers and long diffusion lengths, the implied Voc can be written as

$$ \mathrm{implied}\ {V}_{\mathrm{oc}}=\frac{kT}{q}\ln \left(\frac{\Delta n\ \left({N}_A+\Delta n\right)}{{n_i}^2}\right) $$
(2)
Fig. 6
figure 6

Implied Voc of the SiNx/Al2O3-passivated samples with O2, FG, and two-step annealing

Full size image

where k is the Boltzmann constant, T is the absolute temperature, ni is the intrinsic carrier concentration, NA is the acceptor concentration, and ∆n is the excess carrier concentration measured at one-sun light intensity by the WCT-120 Sinton lifetime tester. It can be seen that the O2- and FG-annealed samples have similar implied Voc values, which are 696 and 697 mV, respectively. The two-step annealing has an implied Voc of 706 mV.

Figure 7 shows the J-V characteristics and photovoltaic parameters such as Voc, short-circuit current density (Jsc), fill factor (FF), and conversion efficiency (η) of the fabricated PERCs with different annealing processes. The performance of an industrial PERC is also shown for the purpose of comparison. The industry PERC was fabricated under identical conditions but no additional annealing process was used, since the Al2O3 layer was annealed during the SiNx deposition at 400 °C. Note that in this study, during the annealing processes, the front side was placed downward and made contact to a wafer holder. The front SiNx layer was not exposed to the annealing gases, and thus, the influence of the front SiNx layer might be insignificant. The industry PERC shows the lowest Voc of 665.4 mV among the others. This could be attributed to its lower wafer lifetime of 797 μs at the injection level of 3 × 1015 cm−3. The Voc value improves to 671.3 mV for O2 annealing and 672.3 mV for FG annealing. The two-step annealing further increases Voc to 675.5 mV, which is an improvement by about 0.6% compared to one-step annealing, or by 1.5% compared to the industry one. There is no much difference in Jsc and FF between the PERCs. The two-step annealing exhibits the best conversion efficiencies of 21.97%, which is 0.36%abs higher than industry PERC. Finally, five PERCs were fabricated for each annealing process. The mean value and distribution range of Voc and FF are shown in Fig. 8a and b, respectively. The PERCs with the two-step annealing show Voc of 675–677.5 mV with a mean value of 676 mV, and FF of 0.813–0.819 with a mean value of 0.816.

Fig. 7
figure 7

Current density-voltage curves and photovoltaic performance of PERCs with industry standard fabrication, O2 annealing, FG annealing, and two-step annealing

Full size image
Fig. 8
figure 8

Mean value and distribution range of a Voc and b FF for PERCs with different annealing processes

Full size image

Conclusion

The Al2O3 films are prepared using atomic layer deposition, followed by O2, FG, or two-step annealing. Comparing O2 annealing with FG annealing, the former yields a thicker SiOx interfacial layer and the higher Qf density of − 3.9 × 1012 cm−2, indicating a superior field effect passivation. The FG annealing shows the lower Dit of 3.7 × 1011 eV−1 cm−2 resulting from the hydrogenation of the Si interface. The two-step annealing combines the advantages of these two annealing processes and has an intermediate Qf and the lowest Dit of 3.1 × 1011 eV−1 cm2. The SiNx/Al2O3-passivated samples with the two-step annealing demonstrate a minority carrier lifetime of 2072 μs, close to the intrinsic lifetime limit. For the PERC fabricated with the two-step annealing, Voc of 675.5 mV and conversion efficiency of 21.97% can be obtained, which respectively have increases of 10 mV and 0.36%abs as compared to those of the industry PERC.