Light-induced activation of boron doping in hydrogenated amorphous silicon for over 25% efficiency silicon solar cells

Recent achievements in amorphous/crystalline silicon heterojunction (SHJ) solar cells and perovskite/SHJ tandem solar cells place hydrogenated amorphous silicon (a-Si:H) at the forefront of photovoltaics. Due to the extremely low effective doping efficiency of trivalent boron in amorphous tetravalent silicon, light harvesting of aforementioned devices is limited by their fill factors (FFs), a direct metric of the charge carrier transport. It is challenging but crucial to develop highly conductive doped a-Si:H with minimal FF losses. Here we report that light soaking can efficiently boost the dark conductance of boron-doped a-Si:H thin films. Light induces diffusion and hopping of weakly bound hydrogen atoms, which activates boron doping. The effect is reversible and the dark conductivity decreases over time when the solar cell is no longer illuminated. By implementing this effect to SHJ solar cells, we achieved a certified total-area power conversion efficiency of 25.18% with a FF of 85.42% on a 244.63 cm2 wafer. Low effective doping of boron limits the performance of solar cells based on hydrogenated amorphous silicon. Liu et al. show that light induces the diffusion of hydrogen atoms, which activates boron doping, enabling a power conversion efficiency of over 25%.

H ydrogenated amorphous silicon (a-Si:H) is a technologically important semiconductor for transistors, batteries and solar cells [1][2][3][4] . It has a long history of use in photovoltaic applications as it offers a low defect density and tunable conduction type [5][6][7] . These optoelectronic advantages strongly rely on configurations of hydrogen and silicon in the three-dimensional space (described by the radial distribution function 8 ) and thus precise control of its microscopic structure 9-11 is a critical factor towards achieving good devices. As boron is a trivalent element, it is challenging to establish four-coordinated B−Si 4 compounds in the disordered a-Si:H matrix; reported approaches, which are focused on eliminating invalid Si x −B−H y doping configurations ( Supplementary Fig. 1), include optimizing the B 2 H 6 flow rate and post-deposition annealing. However, a lack of understanding of the complicated conduction mechanism of boron-doped a-Si:H (p-a-Si:H) has obstructed the full potential of relevant optoelectronic devices.
We find that light soaking is a fast means to improving dark conductance (σ dark ) of p-a-Si:H thin films. Our results indicate that a portion of hydrogen atoms is captured by tetravalent-coordinated boron atoms in the silicon network to form weak B−H−Si components, which diminishes the efficient B−Si 4 doping. We demonstrate that the key function of light soaking is to promote the diffusion and hopping of these weakly bound hydrogen atoms, so that efficient B−Si 4 doping is activated. As a consequence of the improved field passivation and conductivity of p-a-Si:H, we achieve a high power conversion efficiency (PCE) of 25.18% with an open-circuit voltage (V oc ) and fill factor (FF) of 749 mV and 85.42%, respectively, on a 244.63 cm 2 amorphous/crystalline silicon heterojunction (SHJ) solar cell. Moreover, our 60-cell modules exhibit a robust operating stability that successfully passes IEC 60068-2-78 (damp-heat degradation at 85 °C and 85% relative humidity, DH85) and IEC 61215-2:2016 (thermal cycle degradation between −40 °C and 85 °C with applied current at 100% I mpp (current at the maximum power point) at the rising edge of temperature) even after threefold-time-long aging standards.

Observation of light-induced dark conductivity increase
Since 1977, light soaking of micrometre-thick a-Si:H films has been widely studied in the research field of a-Si:H thin-film solar cells, but only a small number of works pay attention to its effect on 'thin' a-Si:H films, particularly in the research field of SHJ solar cells 12,13 . Although a few researchers report that light soaking improves the FF of SHJ solar cells by a magnitude of ~0.7% abs (ref. 14 ), the fundamental underlying mechanisms are still unclear, which attracts broad interest in the research fields of optoelectronics. We use in situ methods to monitor the time-dependent changes of p-a-Si:H thin films during illuminations. The films are deposited on quartz glasses, followed by evaporating silver strips to form the transfer-length-method structures. The in situ current-voltage data (Fig. 1a) show that σ dark of the p-a-Si:H thin film steadily increases during 1 sun illumination, reaching σ dark /σ dark0 ≈ 4.71 (σ dark0 is dark conductance before light soaking) after 30 min. This phenomenon is strikingly in contrast to the light-induced degradation of σ dark observed in thick intrinsic, p-and n-type a-Si:H films [15][16][17] . It supports the perspective that accumulated stress in thick films plays an important role in the Staebler-Wronski effect 18 , as the maximum stress is roughly proportional to the film thickness. This indicates the effect of light soaking exhibits a scaling behaviour, where the Staebler-Wronski effect gradually transitions to a different effect as the thickness declines. After turning off the illumination, σ dark gradually decays (close) to its initial value after more than 1,000 min ( Fig. 1b). Such a decay behaviour fits well to a combination of the Debye and Williams-Watts models (Fig. 1c), The terms Δσ D , Δσ WW and τ D , τ WW are constant coefficients and characteristic time constants of the Debye and Williams-Watts models, respectively. The term t is the decay time in the dark. Detailed parameters are summarized in Supplementary Table 1. The Debye model with β D = 1 describes free diffusion, whereas the Williams-Watts model with 0 < β WW < 1 describes a continuous-time random walk composed of alternating steps and pauses 19 . Examples of the Williams-Watts model include the spin-correlation in Cu-Mn and  (V)   0  15  30  45  60  75  90  105  Voltage (V)   0 min  33 min  3 min  63 min  6 min  123 min  9 min  213 min  12 min  743 min  15 min  1,143 min  18 min  21 min  24 [20][21][22][23][24][25] ). The good fitting in Fig. 1c suggests an effect that is different to the Staebler-Wronski effect and mediated by two independent mechanisms that control the fast Debye relaxation and the slow Williams-Watts relaxation, respectively.

Mechanism underlying the light-induced changes
To determine the implicit mechanisms of aforementioned Debye and Williams-Watts relaxation, we investigate the hydrogen distributions in p-a-Si:H thin films by time-of-flight secondary ion mass spectrometry (TOF-SIMS). The H − spectra (Fig. 2a) show that 30 min annealing at 180 °C only slightly changes the hydrogen content in intrinsic a-Si:H (i-a-Si:H), by contrast, the same annealing process expels at least ~21.3% of the hydrogen content from p-a-Si:H. As shown in Supplementary Fig. 2, TOF-SIMS spectra also reveal that room-temperature oxidation of an i/p-a-Si:H stack hardly changes the hydrogen content in the i-a-Si:H film, however, the same oxidation process expels ~17.1% of the hydrogen content in p-a-Si:H from the inside to the surface. Based on these findings, we conclude that the boron doping plays a crucial part in the formation of metastable hydrogen configurations in p-a-Si:H.
We next consider the migration barriers of hydrogen atoms to understand the possible binding configurations of aforementioned metastable hydrogen. Structural relaxations observed displacement of four-coordinated silicon atoms by boron atoms shorten the bonds from ~2.35 Å to ~2.07 Å ( Supplementary Fig. 3), well consistent with the results of Pandey and colleagues 26 . Further simulations demonstrate that these B−Si 4 sites have a large probability of attracting hydrogen atoms to form metastable B−H−Si configurations when diffusive hydrogen atoms pass by ( Supplementary Fig. 4), which is in agreement with the nuclear magnetic resonance signal 27 and relevant simulations 26,28 . As a consequence, conductance of p-a-Si:H is expected to decline due to reduction in the quantity of B−Si 4 (ref. 26 ). Transition-state surveys (Fig. 2b,c)  the hydrogen hopping mechanism (or the tunnelling mechanism 31 at temperatures below 60 K) illustrated in Fig. 2b, resulting in improvement of σ dark as has been confirmed in Fig. 1a. The microscopic migrations in Fig. 2b are consistent with the light-induced formation of Si−H−Si configurations 32 .
The mechanistic understanding is also evident in optoelectronic analysis. We prepared symmetric structures of p-a-Si:H/i-a-S i:H/n-c-Si/i-a-Si:H/p-a-Si:H (here n-c-Si represents n-type c-Si) and i-a-Si:H/n-c-Si/i-a-Si:H, whose injection-dependent effective minority carrier lifetimes (τ eff ) were measured before and after 2 h light soaking under 1 sun illumination, as well as that after 15 min annealing at 180 °C. The right graph in Fig. 2d shows that the τ eff of p-a-Si:H/i-a-Si:H/n-c-Si/i-a-Si:H/p-a-Si:H increased substantially after light soaking and then returned to initial values after the annealing. The recombination rate at the a-Si:H/c-Si interface satisfies a closed-form expression in case of high illumination, which can be fitted by the model of Olibet and colleagues 30 ; by modelling the τ eff at injection >1.0 × 10 15 cm −3 , we determined that light soaking increased (decreased) the surface charge density Q s (the interface dangling-bond density N s ) from 3.0 × 10 10 cm −2 (2.1 × 10 9 cm −2 ) to 3.8 × 10 10 cm −2 (1.4 × 10 9 cm −2 ), and then annealing decreased (increased) the Q s (N s ) back to 3.0 × 10 10 cm −2 (2.0 × 10 9 cm −2 ). As a control, the left graph in Fig. 2d shows that the τ eff of i-a-Si:H/n-c-Si/i-a-Si:H almost remained constant after either light soaking or annealing, which demonstrates that the variation in τ eff in the right graph must originate from p-a-Si:H. According to Sinton and co-workers 33 , pseudo FFs (PFFs) of silicon solar cells take into account the effect of chemical passivation. We probed the PFF of the device Ag/IWO/p-a-Si:H/i-a-Si:H/n-c-Si/i-a-Si:H/n-a-Si:H/ IWO/Ag (where IWO is tungsten-doped indium oxide) before and after 2 h light soaking under 1 sun illumination, as well as that after 15 min annealing at 180 °C. The inset of Fig. 2d finds that the PFF maintains a PFF of ~86.4% regardless of light soaking and annealing. This demonstrates that the decrease in N s has a negligible impact on chemical passivation, probably due to the small order of magnitude of N s itself. On the other hand, ultrafast and broadband transient absorption signals (Fig. 2e) indicate that light soaking increases the mobility of photon-generated carriers from 7.10 × 10 −3 cm 2 V −1 s −1 to 1.81 × 10 −2 cm 2 V −1 s −1 in p-a-Si:H. This probably results from less scattering of carrier transport in the p-a-Si:H network, thanks to the global decline of strain-induced gap states from B−H−Si configurations 26 . Consideration of these light-induced enhancements to σ dark , Q s and the carrier mobility leads to the conclusion that the light-induced dark conductivity increase stems from activation of boron doping via hydrogen movements. In this regard, we further ascribe the decay of σ dark in Fig. 1c to the detrimental reconstruction of B−H−Si configurations, as the binding energy of hydrogen in B−H−Si is ~0.46 eV higher than that in Si−H−Si. In accordance, the fast Debye and slow Williams-Watts relaxations (Fig. 1c) are attributed to incorporation of fast diffusive hydrogen and slow hopping hydrogen into the B−Si bonds, respectively, forming invalid boron doping B−H−Si that has negative effects on the σ dark as has been confirmed in Fig. 1b.
We next distinguish the weakly bound hydrogen atoms from the normal Si−H bonds in p-a-Si:H to strengthen the mechanism underlying the light-induced dark conductivity increase. Figure  3a illustrates the preparation of p-a-Si:H films for TOF-SIMS, Fourier-transform infrared spectroscopy (FTIR) and current-voltage characterizations. The capping of an IWO layer on the p-a-Si:H is to mimic the structure of SHJ solar cells, which may have an effect on the redistribution dynamics of hydrogen atoms during the annealing process. TOF-SIMS signals (Fig. 3b) find 2 h annealing at 180 °C reduced >20% of the hydrogen content in the p-a-Si:H film, whereas the content of silicon and boron (almost) remained unchanged. By contrast, Fig. 3c finds that all of the wagging, bending and stretching intensities of the normal Si−H bonds (almost) remain unchanged after the same annealing process. A comparison between the TOF-SIMS signals of hydrogen atoms and the FTIR spectra of Si−H bonds unambiguously demonstrates that relatively low-temperature (180 °C) annealing merely expels weakly bound hydrogen atoms from the p-a-Si:H film while the normal Si−H bonds are hardly affected. The light-induced enhancement of dark conductance of the p-a-Si:H film is plotted as a function of the annealing time at 180 °C in Fig. 3d, evidently, the σ dark /σ dark0 gradually declined to ~1 due to the exhaustion of weakly bound hydrogen atoms during the prolonged annealing. This definitely proves the light-induced dark conductivity increase and boron doping activation does stem from weakly bound hydrogen atoms rather than normal Si−H bonds in the p-a-Si:H.
According to Pandey and colleagues 26 , there exist a host of possible configurations of weakly bound hydrogen atoms with respect to boron atoms in the complicated p-a-Si:H network, such as weak hydrogen atoms nearby B−Si 4 doping, boron dimers and boron clusters and so on. By changing the flow rate of B 2 H 6 during the film deposition, we fabricated four p-a-Si:H films on quartz glasses, their current-voltage characteristics are shown in Supplementary Fig. 5a. We find that dark conductance ( Supplementary Fig. 5b) gradually saturates when the flow rate of B 2 H 6 exceeds ~45 sccm, which suggests that a huge amount of boron atoms are invalidly doped into p-a-Si:H, or do not make contribution to hole concentration. Furthermore, light-induced enhancement of dark conductance substantially decreases when the flow rate of B 2 H 6 exceeds ~20 sccm ( Supplementary Fig. 5c). Taking into account the possibility that boron dimers and clusters dominate only in case of high flow rates of B 2 H 6 molecules, we conclude that the light-induced dark conductivity increase and boron doping activation mainly stem from the weak hydrogen atoms nearby the most important B−Si 4 doping sites, rather than those nearby boron-superabundant configurations, such as boron dimers and boron clusters and so on.

Application to high-efficient SHJ solar cells
Encouraged by the enhancement of σ dark by light soaking, we attempt to develop the full potential of SHJ solar cells by this light-induced effect. Figure 4a showcases the device structure (where the thickness of the p-a-Si:H is ~15 nm; Supplementary Fig. 6) whose initial open-circuit voltage (V oc ), short-circuit current density (J sc ), FF and PCE are 744.30 ± 0.68 mV, 38.43 ± 0.07 mA cm -2 , 83.70 ± 0.22% and 23.94 ± 0.04% respectively, based on 316 continuous devices from our daily production line. Under 1 sun illumination, as expected, the FF of these cells undergoes a steady increase (standard cell in Fig. 4b).
The slope of the current-voltage curve near the low-internal-field region (V oc condition) serves as an indication of charge collection efficiency 35 , as found in Supplementary Fig. 7, the light soaking continuously increases the slope near this low-internal-field region, indicating more efficient charge extraction due to enhancement of the net field across the depletion region. This strongly supports our perspective that the light soaking activates better boron doping. By contrast, we observe a noticeable drop in the gain of FF for devices annealed for 2 h at 180 °C (180 °C in Fig. 4b), attributed to its less metastable hydrogen configurations (inferred from Figs. 2a and 3b), which leads to less hydrogen movements in Fig. 2b. Intriguingly, we   observe that when 13 A current is applied to the cell (Fig. 4b), the FF exhibits a quite similar behaviour to that under 1 sun illumination. This implies the photon energy from light soaking is not the exclusive cause responsible for the dark conductivity increase and boron doping activation, electron-hole recombination caused by current-injected carriers probably also take effect 31 . Further increasing the light intensity from 1 to 11, 48 and 60 sun boosted the FF by 0.32 ± 0.18% abs , 0.39 ± 0.14% abs , 1.40 ± 0.26% abs   and 1.50 ± 0.37% abs , respectively (Fig. 4c). Here the improvement in FF under 60 sun illumination is close to the ΔFF ≈ 1.8 ± 0.4% abs reported via a multifunctional process 36 . We also notice that increasing either the light intensity or the forward bias can improve the magnitude of ΔFF (Fig. 4c and Supplementary Fig. 8). This highlights that intensive light soaking or high forward bias activates more efficient boron doping by pumping more metastable hydrogen from B−H−Si to other configurations, in this consideration, we naturally regard SHJ solar cells as the premium choice for concentrator photovoltaic systems. At the mass-production level, 60-sun illumination obtains state-of-the-art industrial FF and PCE of 85.19 ± 0.18% and 24.46 ± 0.05%, respectively (Fig. 4d,e), together   with improved V oc by ~2.6 mV ( Supplementary Fig. 9), thanks to improvement of the build-in field in c-Si absorber. Numerical investigation on these improvements is based on a traditional drift-diffusion model of the SHJ solar cell. Procedures and simulated parameters are provided in Supplementary Tables 2  and 3 Table 4 reveals the decline of N s from 9.0 × 10 8 cm −2 to 4.3 × 10 8 cm −2 only slightly increases the FF by 0.04% abs , much lower than the experimental 1.50 ± 0.37% abs , but in good consistency with the PFF in Fig. 2d. Samples A and C show that the FF increases by 0.66% abs when the efficient doping concentration of boron (N a ) increases from 2.0 × 10 18 cm −3 to 1.0 × 10 19 cm −3 . Samples A and D show the FF boosts by 0.77% abs when the series resistance (R s ) declines from 0.4 Ω cm 2 to 0.25 Ω cm 2 (Supplementary Table 4). The collective refinements to N s , N a and R s improves the FF and PCE from 83.79% and 24.0% to 85.25% and 24.5%, respectively, in good agreement with the experimental results from 83.70 ± 0.22% and 23.9 ± 0% to 85.19 ± 0.18% and 24.5 ± 0.1% (samples A and E in Supplementary Table 4). Together with that, the simulated increase of V oc from 744.3 mV to 746.9 mV is also identical to the experimental results from 744.3 ± 0.7 mV to 746.9 ± 0.5 mV. Such excellent consistencies between simulations and experiments confirm the improvements in SHJ solar cells do stem from light-induced efficient doping of boron atoms.

. Comparison of samples A and B in Supplementary
After capping an 80 nm SiO x antireflection layer onto a high-efficiency cell, we submitted it to an independent testing centre and achieved a certificated PCE of 25.18% with a FF of 85.42% on a 244.63 cm 2 wafer (Fig. 4f and Supplementary Fig. 10). They are among the highest certificated PCE and FF for total-area two-side contacted silicon solar cells 34 (Fig. 4g and Supplementary Fig. 11). The FF reaches 98.30% of its Shockley-Queisser limit, ~86.9% (ref. 37 ). We also submitted another SHJ solar cell capped with a 110 nm MgF 2 antireflection layer to ISFH CalTeC; they reported total-(244.81 ± 0.91 cm 2 ) and designated-area (226.71 ± 0.91 cm 2 ) PCEs of 25.10 ± 0.38% and 25.45 ± 0.38%, respectively ( Fig. 4f and Supplementary Fig. 12). The a bit lower FFs of 84.28% ± 0.93% and 84.63 ± 0.93% than that certificated from NPVM probably stem from the degradation between the 70 sun light soaking and the certification.
With regard to stability, FF and PCE of devices retain 98.70% and 97.59% of their initial values after 1,000 h DH85 impact (Fig. 4h), without any encapsulations. At the module level, Figs. 4i and 4j show that the FF and PCE retain 98.1% (96.8%) and 95.5% (95.4%), respectively, after 3,000 h DH85 impact (600 thermal cycles between −40 °C and 85 °C), demonstrating their high stability against extreme climate degradation factors. The DH85 (thermal-cycle) degradation of the module is threefold longer than the IEC 60068-2-78 (IEC 61215-2:2016) standard. These tests exclude the high-density (~10 21 cm −3 ) weakly bound hydrogen atoms in the p-a-Si:H film as the key factor that dominates the damp-heat (thermal-cycle) degradation 4 .
In addition to the stabilities in DH85 and thermal cycle environments, we finally explored the reversible behaviour of SHJ solar cells caused by the light-induced dark conductivity increase and boron doping activation. As found in Fig. 5a, we alternated between measuring the cells' FFs under 1 sun illumination for 180 min and the dark for 720 min. Evidently, the FF decays ~0.3−0.35% abs during each sleeping in the dark. From Supplementary Fig. 13, we find that the FF rapidly declined by ~0.15% abs in the first ~20 min, followed by a slow decay in the next ~745 min. This fast decay time of ~20 min is consistent with the characteristic time constant τ D ≈ 15.84 ± 1.55 min of the Debye relaxation (Supplementary  Table 1), confirming the enhancement of FF does stem from the improvement of conductance of doped a-Si:H film. Figure 5a also reveals that the FF rapidly climbs up after turning on the light soaking; thus, the output of power plants comprising SHJ solar cells undergoes a rapid increase after sunrise on sunny days, which challenges the present IEC testing standards, as the in-house certification underestimates their performance in real operations. The following provides a feasible pathway to freezing the dark decay. We took 198 solar cells from the same batch and divided them into 11 groups. First, the devices in each group undergo a 70 s light soaking under 60 sun illumination, followed by a 25 min sleeping in the dark to finish the fast Debye relaxation. Their FFs were then measured before and after 10 min annealing at different temperatures, as shown in Fig. 5b, the decay magnitude of FF (from Williams-Watts relaxation) dramatically drops when the temperature is decreased from 200 °C to 60 °C, which suggests that the low temperature arrests the unfavourable formation of the B−H−Si configurations. This observation agrees with the perspective that annealing can accelerate annihilation of Si−H−Si configurations 38 . Using the average ΔFF (Fig. 5b), we derived the temperature-dependent characteristic time constant τ WW by the Williams-Watts model, According to Kakalios and co-workers 39 , the β WW of a-Si:H is 0.00165T (in Kelvin), independent of the doping type; the τ WW , on the other hand, obeys an Arrhenius relationship, where R is molar gas constant. Figure 5c shows the fitting of equation (3) to the τ WW (blue circles) calculated from equation (2), interestingly, the theoretical τ WW (red circle) from Supplementary Table 1 is close to the extrapolation of the fitting line, confirming the validation of equation (3). The derived activation energy E a ≈ 0.399 eV is well agreement with the prediction of migration barrier ~0.417 eV and the 0.385 ± 0.143 eV inferred from the reported data of doped a-Si:H (ref. 40 ). Figure 5d finds E a of doped a-Si:H is noticeably smaller than that of the intrinsic counterpart [40][41][42][43][44] , most likely owing to existence of the exclusive metastable hydrogen configurations in doped materials (as inferred from Figs. 2a and 3b). We notice phosphorus-doped a-Si:H also has smaller E a , thus it is expected to make similar contributions to the light-induced effect. This speculation is evident from the light soaking behaviour of 'half ' cells with structure of Ag/IWO/n-a-Si:H/i-a-Si:H/n-c-Si/IWO/Ag, where the p-a-Si:H is totally removed (Supplementary Fig. 14). It is interesting to find the doped a-Si:H thin films exhibit an opposite light-induced behaviour in comparison to p-type c-Si when oxygen atoms exist in the form of B s −O 2i complexes inside the crystalline matrix 45 . Given doped a-Si:H has small E a but great τ WW at low temperatures, we conclude that the cold climates can effectively prevent the decay of metastable FF.

Conclusion
We observed light-soaking-induced enhancement of dark conductance of boron-doped a-Si:H thin films, which is appealing for realizing outstanding optoelectronic devices. We show that light soaking promotes the diffusion and hopping of the weakly bound hydrogen atoms, which allow the activation of B−Si 4 doping. The light-soaking effect noticeably improves the charge carrier transport in SHJ solar cells, yielding an excellent FF of 85.42% (84.63%) and a PCE of 25.18% (25.45%) on a 244.63 cm 2 (226.71 cm 2 ) total-area (designated-area) wafer. during which the chamber pressure, primary ion source and current are 1.0 × 10 −9 mbar, 30 keV Bi + and 1.0 pA, respectively, the depth profiles were acquired using a 500 eV Cs + sputter beam. Cross-sectional images of p-a-Si:H were probed by high-resolution transmission electron microscope (FEI Titan 80-300ST), operated at 200 kV. Injection-dependent τ eff and PFF were measured by the Sinton WCT-120 and Suns-Voc, respectively. The Q s and N s are fitted from a surface recombination model by Olibet and colleagues, they discussed details about the model and also provided the fitting codes in the appendix A of Olibet's thesis 47 .

Methods
Ultrafast and broadband transient absorption spectra were measured using a homebuilt pump-probe set-up. The output of a titanium sapphire amplifier (Coherent LEGEND DUO, 4.5 mJ, 3 kHz, 100 fs) splits into three beams (2.0 mJ, 1.0 mJ and 1.5 mJ), two of which separately pump two optical parametric amplifiers (OPA; Light Conversion TOPAS Prime). TOPAS-1 provides tunable pump pulses and TOPAS-2 generates the probe pulses. A 1,300 nm pulse from TOPAS-2 is sent through a CaF 2 crystal mounted on a continuously moving stage. This generates a white-light supercontinuum pulses from 350 nm to 1,100 nm. The pump pathway length is varied between 5.12 m and 2.60 m with a broadband retroreflector mounted on an automated mechanical delay stage (Newport linear stage IMS600CCHA controlled by a Newport XPS motion controller), thereby generates delays between pump and probe from −400 ps to 8 ns. Pump and probe beams are overlapped on surface of the p-a-Si:H. By a beam viewer (Coherent, LaserCam-HR II) we regulate the size of pump beam about three times larger than the probe beam. The probe beam is guided to a custom-made prism spectrograph (Entwicklungsbüro Stresing) where it is dispersed by a prism onto a 512 pixel complementary metal-oxide semiconductor linear image sensor (Hamamatsu G11608−512DA). The probe pulse repetition rate is 3 kHz, whereas the excitation pulses are mechanically chopped to 1.5 kHz (100 fs to 8 ns delays), and the detector array is read out at 3 kHz. These characterizations are also summarized in Supplementary Table 5.
The transient absorption signals are fitted by the one-dimension recombination and diffusion model: ) .  N(x, t) is the carrier density, which is a function of the time t and the position x in the film, D is the diffusion coefficient, k 1 , k 2 , k 3 are the first-, second-and third-order bulk recombination constants, α is the absorption coefficient, and S f and S b are the front and back interface/surface recombination velocities. The front surface/interface is exposed to the pump laser beam. The general rate equation consists of the diffusion equation and includes the different recombination rates present in the bulk. N(x, t) is proportional to ΔT/T, N(x, t) = β ΔT T with a fitted prefactor β. About 80 nm IWO was grown on both sides of the devices by the reactive plasma deposition (RPD, DS1-12080-SN-D13; Shenzhen S.C.) at 150 °C, whose target material is 1.0% tungsten-doped in indium oxide. Nine silver busbars and fingers were screen printed on the two faces of the devices using low-temperature paste, followed by annealing at 150 °C for 5 min and 185 °C for 30 min. For the certificate cell, an 80 nm SiO x layer was capped onto the sun-side surface in a 13.56 MHz radio-frequency PECVD (ULVAC CME-400).

The initial condition is
Device characterization. Current-voltage characteristics of all solar cells without SiO x antireflection were tested under standard conditions (25 °C, 100 mW cm −2 ) using a solar simulator (Halm IV, ceitsPV-CTL2). The light intensity was calibrated using a certified National Renewable Energy Laboratory (NREL) reference cell. The submitted cell with SiO x antireflection was independently tested by the NPVM in the Fujian Province, China, one of the designated test centres for the solar cell efficiency tables. The device area was captured by an automatic image test system. Before certification, it was light soaked for 30 min under 1 sun illumination, followed by cooling down to room temperature. The cell with MgF 2 antireflection layer was independently tested by ISFH CalTeC, which experienced a 1 sun illumination before the measurement. The conveyor during light soaking was pre-heated to ~200 °C, and the light intensity was adjusted from 1 to 60 sun (ASIA NEO TECH INDUSTRIL Company, NLIDR-S60; red light). These cells were quickly cooled down by cold-air blowing after the light soaking. All devices were measured under standard conditions. These characterizations are also summarized in Supplementary Table 6.
Damp-heat degradation. The devices underwent 1,000 h damp-heat impact at DH85 in the dark, during which they were in open-circuit condition. These devices are six-inch SHJ solar cells without any encapsulations. The 60-cell module underwent 3,000 h damp-heat impact at DH85 in the dark according to the IEC 60068-2-78, during which it was in open-circuit condition. All measurements were conducted under standard conditions (25 °C, 100 mW cm −2 ), out of the climate test chamber.
Thermal cycle degradation. The thermal cycles were conducted in accordance with the IEC 61215-2:2016. The cycle temperature was between −40 °C and 85 °C. The applied current was 100% I mpp at the rising edge of temperature.

Simulation procedures and parameters of SHJ solar cells.
The rear-junction SHJ solar cell was modelled using the traditional drift-diffusion models on the AFORS-HET device-simulation platform previously developed for heterojunction solar cells 50 . General parameters are listed in Supplementary Table 2. The effects of N s , N a and R s on the device performances are different. N s represents chemical passivation, which is dominated by defect density at the a-Si:H/c-Si interfaces. Q s represents the surface charge density at the depletion region of the p-n junction, which cannot be directly used in the simulation. As an alternative option, we use N a to represent the field passivation. R s represents transport series resistance in the device, here its variation mainly stems from the bulk resistance of p-a-Si:H as has been observed in Fig. 1a. Both N a and R s are dominated by the doping efficiency of boron in the p-a-Si:H film. N s is derived from modelling the τ eff of IWO/p-a-S i:H/i-a-Si:H/n-c-Si/i-a-Si:H/n-a-Si:H/IWO, measured before and after 70 s light soaking under 60-sun illumination. R s was measured from the solar simulator under the standard conditions. N a is the only optimized parameter to matching the experimental V oc , J sc , FF and PCE of SHJ solar cells before and after the 70 s 60-sun illumination. Samples A and E in Supplementary Tables 3 and 4 are control samples, whose performances are very close to those of the as-prepared and light-soaked (60 sun) SHJ solar cells respectively. To distinguish the effects of N s , N a and R s on the performance of SHJ solar cells, samples B-D in Supplementary  Tables 3 and 4 change N s , N a and R s one by one. By this means, we can determine the key factors that dominate the device-level light-induced changes.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.