Unless otherwise stated, all chemicals are purchased from Sigma-Aldrich and used as purchased. ITO coated 100 µm thick WG (Fraunhofer FEP) or ITO coated 1 mm thick glass (Kintec) were used as substrates. The 1 mm Glass/ITO substrates were rinsed sequentially in soapy water, acetone, IPA, and DI water and dried with N₂. The WG/ITO substrates were only blown with N₂ without chemical cleaning. Both substrates were UV-ozone treated (Bioforce Nanosciences Procleaner Plus) for 20 min before film deposition. NiO_x precursor solution was made, modified from Liu et al. (2018), by dissolving 0.1 M nickel nitrate hexahydrate and 0.1 M acetylacetone in 2-methoxyethanol and stirring overnight at room temperature. The NiO_x precursor solution was filtered through a 0.2 µm polytetrafluoroethylene (PTFE) filter before spin coating at 3,000 rpm for 30 s followed by drying on a hot plate at 70°C for 3 min. For thermal annealed (TA) NiO_x, the dried precursor films were calcinated on a hot plate at 250°C for 45 min. For photonic cured (PC) NiO_x, the samples were processed per Photonic Curing immediately after drying. The NiO_x coated samples were then loaded into an N₂-purged glovebox to deposit the active layer. MAPbI₃ precursor solution was made, following Ahn et al. (2015), by dissolving lead (II) iodide (TCI America), methylammonium iodide (GreatCell Solar), and dimethyl sulfoxide (equal molar ratio) in N,N-dimethylformamide. The MAPbI₃ precursor solution was filtered through a 0.2 µm PTFE filter before spin coating at 4,000 rpm for 25 s inside an N₂-purged glovebox. After spinning for 10 s, 1 ml of diethyl ether was applied to the sample. For TA MAPbI₃, the samples were immediately annealed on a hot plate in the same glovebox at 100°C for 10 min. For PC MAPbI₃, the samples were removed from the glovebox immediately after spin coating and processed per Photonic Curing before returning to the glovebox to deposit the ETL. PC₆₁BM (Solenne BV) in chlorobenzene (13 mg/ml) was spin coated at 1,200 rpm for 60 s followed by spin coating bathocuproine (BCP) in ethanol (0.5 mg/ml) at 4,000 rpm for 30 s to create the ETL. No annealing step was required for the ETL. The top contact was deposited by thermally evaporating 100 nm of Al through a shadow mask.

Spectroscopic ellipsometry measurements (M-2000DI, J. A. Woolam) were done at 55°, 65°, and 75° incident angles from 192 to 1,690 nm in wavelength to obtain the index of refraction (n) and extinction coefficient (k) for each material. Uncoated WG, WG/ITO, WG/ITO/NiO_x-precursor and 1 mm Glass/ITO were measured with transparent tape adhered to the backside of the samples to eliminate back-side reflections. The NiO_x, MAPbI₃, PC₆₁BM, and Al films were deposited on single side polished Si wafers, following the same deposition procedures as Solar Cell Fabrication, for the measurement. The data were fitted using Complete Ease software to determine the optical constants and thickness of each film. The fitting for WG was performed using a Cauchy model parameterized into a general oscillator. The ITO coated on WG was fitted using the previously determined WG optical constants to obtain the optical constants of the ITO film. The fitting for ITO on WG starts with a general oscillator function that was first fitted over a limited wavelength range then converted into a B-spline model. The B-spline model was then expanded over the full wavelength range and parameterized into a general oscillator. The fitting for NiO_x-precursor on WG/ITO used a B-spline model and the optical constants of the WG/ITO. Fits for NiO_x, MAPbI₃, PC₆₁BM, and Al were obtained using B-spline models and the optical constants of the Si substrates. A B-spline model was also used for the ITO on 1 mm Glass with the optical constants for the glass taken from Rubin (1985). All fits had a mean square error of less than 9.

Photonic curing was done using a NovaCentrix PulseForge^® Invent equipped with one 500 V lamp driver, 1.5 kW power supply, and 20 mm (diameter) × 150 mm (length) xenon flash lamp (75 mm × 150 mm illumination area). All photonic curing was performed in ambient conditions (22–24°C, 30–50% RH). The processing stage height was adjusted to be as close to the lamp as the machine allows. The WG/ITO/NiO_x samples were suspended between two pairs of magnetic steel slats such that there was an air gap below the sample, between the WG substrate and the processing stage, during photonic curing. MAPbI₃ samples were placed directly on the processing stage and held down by two magnetic steel slats for photonic curing.

The radiant energy at each PC processing condition was measured using a calibrated bolometer five separate times to get an average value. The bolometer was also used to determine the surface absorption of the WG/ITO/NiO_x-precursor sample by measuring the radiant energy with and without the sample over the bolometer. The difference in the radiant energy measured with and without the sample, minus 8.8% substrate reflection, was used as the surface absorption of the sample for SimPulse^® simulations. The 8.8% reflection value was determined by performing the same measurement on an uncoated WG sample and assuming that the WG absorbs no light, which is quite accurate (Supplementary Figure S2). The WG/ITO/NiO_x-precursor and uncoated WG samples were pulsed with a lamp voltage of 200 V and a pulse length of 4 ms five separate times and the average radiant energy was used for calculations. The standard deviation in radiant energy for each set of bolometry measurements was less than 1%. This method estimates the surface absorption for WG/ITO/NiO_x-precursor is 18%.

The percentage of WG/ITO/NiO_x-precursor absorption can also be calculated with 1D transfer matrix method (TMM) (Burkhard et al., 2010) using the xenon lamp spectrum (NovaCentrix, 2020) as the source and n and k of WG/ITO/NiO_x-precursor obtained from spectroscopic ellipsometry for the sample. We obtained 16% using the TMM method, which is consistent with the bolometry method described in Photonic Curing. The UV-vis spectrum of a dilute NiO_x-precursor solution is shown in Supplementary Figure S3.

Simulations of film surface temperature vs. time were generated using NovaCentrix SimPulse^® software. The parameters input into the software were the PulseForge^® configuration (used to model the lamp output), the measured light absorption by the sample, the material properties (i.e., thermal conductivity, specific heat, and density) from literature, and the measured sample thickness. Note that SimPulse^® only models the light absorption and thermal transfer processes and does not take into account any photochemical processes that may occur during photonic curing (Daunis et al., 2020). The WG/ITO/NiO_x-precursor sample stack was modeled as 100 µm borosilicate glass and 150 nm ITO. The simulations were done in surface absorption mode using 18% absorption determined from the bolometry measurement as described in Photonic Curing. Using the TMM calculated 16% surface absorption value will only affect the simulated temperature values, but will not change the conclusions drawn from SimPulse^® results. Because the NiO_x-precursor layer is only ∼30 nm thick, it is much thinner than the thermal length and is not included in the SimPulse^® model stack. It is treated as a surface absorbing layer with its contribution measured by bolometry. SimPulse^® assumes that the light pulse is absorbed at the surface of the sample and the heat generated there is either conducted down through the sample stack, raising the overall sample temperature, or removed from the sample via convective heat transfer to air from the top and bottom surfaces. The sample is suspended and has air above and below it. SimPulse^® models this boundary condition using a convective heat transfer coefficient of 15 W/m²K at the top and bottom sample surfaces. The assumption that the heat is generated at the top surface is reasonable because most of the light is absorbed in the NiO_x-precursor and ITO layers, which account for only 0.18% of the entire stack thickness.

Short-circuit current (J _SC) calculations were performed by TMM with an AM 1.5G 100 mW/cm² light source. Two different material stacks were used (Supplementary Figure S4A) to model the different substrates: WG/ITO and 1 mm Glass/ITO. The device layers, consisting of 10 nm NiO_x, 450 nm MAPbI₃, 50 nm PC₆₁BM, and 100 nm Al, were the same for both substrates. The optical constants and thickness of the sample layers were determined from spectroscopic ellipsometry, except for 1 mm Glass, as described in Spectroscopic Ellipsometry. The TMM calculations assume the MAPbI₃ active layer has an internal quantum efficiency of 100%, which was reported in Yang et al. (2015).

Scheme 1 depicts the different combinations of TA or PC processes to form NiO_x HTLs and MAPbI₃ active layers on WG/ITO substrates. Since both layers can undergo TA or PC processes, the resulting PSCs can have four different configurations: TA NiO_x + TA MAPbI₃ (TA/TA), TA NiO_x + PC MAPbI₃ (TA/PC), PC NiO_x + TA MAPbI₃ (PC/TA), and PC NiO_x + PC MAPbI₃ (PC/PC). The PC₆₁BM, BCP, and Al top layer deposition processes are kept constant. During the NiO_x PC process, a xenon flash lamp delivers short, 0.45–10 ms, pulses of light (1–30 pulses) to the sample at repetition rates ranging from 0.1 to 1 Hz and a lamp voltage of 500 or 700 V.

All PC MAPbI₃ samples were processed using a single pulse with a lamp voltage of 300 V and a pulse length of 20 ms. This PC processing condition for MAPbI₃ was adapted from our previous work on 1 mm glass (Xu et al., 2020), with the lamp voltage reduced for the thinner WG substrates. To compare the PC MAPbI₃ conversion on top of TA or PC NiO_x HTL, we first conducted XRD to confirm the crystallinity. Figure 1A shows that the underlying NiO_x HTL processed by TA or PC has little impact on the crystalline phase of the PC MAPbI₃ films. Correspondingly, SEM shows that the PC MAPbI₃ films have similar grain size distribution and film morphology (Figures 1B–D).

Figure 2A shows that PSCs made in the PC/PC configuration exhibit similar champion J-V performance as TA/TA and TA/PC control samples. The PCE values of the champion devices were very similar for all three configurations: 11.5, 12.5, and 11.7%, for TA/TA, TA/PC, and PC/PC, respectively. It is worth noting that the PC/PC devices show less hysteresis than the other two types of devices. Figure 2B shows the average and variation in reverse-scan open-circuit voltage (V _OC), J _SC, fill factor (FF), and PCE for all three configurations are similar. The NiO_x HTLs in the PC/PC devices shown were photonic cured using 500 V lamp voltage, 10 ms pulse length, and 10 pulses at a repetition rate of 0.2 Hz. The inset of Figure 2A shows an image of the flexible WG device. Table 1 contains average reverse-scan PSC device parameters and total NiO_x processing time for each photonic curing condition used to process the NiO_x HTL.

To optimize the process for high-throughput manufacturing, we aim to find the PC processing conditions that can successfully convert NiO_x with the shortest time within the PulseForge^® Invent’s capability. The maximum lamp voltage of 500 V was selected to maximize the amount of power delivered per pulse. For each pulse condition, the maximum repetition rate allowed by the system was selected to minimize the processing time. The maximum repetition rate depends on the lamp voltage and pulse length, and is limited by how fast the capacitor bank can be recharged by the power supply. At a fixed lamp voltage, longer pulse lengths result in lower repetition rates. More energy is expended within a single pulse for longer pulses because the lamp is on for a longer time; thus, more time is needed to recharge the capacitor banks between pulses. Therefore, with a given tool and a fixed 500 V lamp voltage at the maximum repetition rate, the two independent variables are the pulse length and number of pulses.

At a fixed lamp voltage, the peak sample temperature (T_p) for a single-pulse process increases with pulse length up to a saturation value when the pulse length is longer than a threshold value (t₀). For pulses longer than t₀, T_p does not increase further with pulse length. Supplementary Figure S6 shows how T_p changes with pulse length at 500 V lamp voltage. For our sample in our tool at the maximum 500 V lamp voltage, t₀ equals ∼3 ms. T_p increases with pulse length for pulses <3 ms until it reaches the single-pulse saturation value at ∼ 3 ms. T_p is then equal to the single-pulse saturation value, independent of pulse length, for pulses >3 ms. t₀ and the associated single-pulse saturation T_p value depend on the photonic curing tool and parameters used particularly lamp voltage and number of lamp drivers as well as the sample properties and boundary conditions (Xu et al., 2020). Higher lamp voltages will produce higher T_p values for a given pulse length because more power is being delivered to the sample.

We used SimPulse^® to calculate the thermal response of the WG/ITO/NiO_x-precursor sample under different photonic curing conditions. Figures 3A,B show the simulated thermal response of the sample when processed with a single, 500 V pulse at pulse lengths of 1 ms (<t₀) and 5 ms (>t₀), respectively. The yellow shaded region indicates the time that the lamp is on. As soon as the lamp turns on, the sample temperature rapidly increases. The lamp intensity is highest when the lamp initially turns on and decreases as the charges in the capacitor bank deplete; therefore, the sample heating rate will decrease throughout the pulse duration. As the sample temperature increases, the cooling rate increases due to convection and conduction of heat away from the sample. When the lamp turns off, the heating rate goes to zero and sample cooling takes over, as evidenced by a steep decrease in temperature past the yellow shaded region. While the cooling rate is initially very fast when the sample surface is at high temperature, the rate decreases as the sample temperature decreases resulting in the temperature returning to ambient over a time period on the order of seconds, which is not shown in these plots. For pulse length shorter than t₀, the lamp is turned off during the initial rapid increase in sample temperature, reaching the peak temperature T_p right at the time when the lamp is turned off, as seen in Figure 3A. This T_p value (200°C for 1 ms pulse length) is lower than the single-pulse saturation T_p value (234°C) shown in Supplementary Figure S6 because the sample did not have time to absorb enough radiative energy from the lamp. When pulse length is longer than t₀, the cooling rate and heating rate balance each other out at t₀ and the single-pulse saturation T_p value is attained. While the lamp is still on past t₀, the cooling rate becomes greater than the heating rate, resulting in the temperature declining. This is evident from the slow decrease in temperature between 3 and 5 ms in Figure 3B.

Because our tool is limited to a 500 V capacitor bank, multiple pulses are required to convert the nickel nitrate precursor films to NiO_x. Figures 3C,D show the temperature profiles for 10 pulses at 1 ms (1 Hz) and 5 ms (0.27 Hz) pulse lengths, respectively. As previously discussed, the repetition rate is lower for longer pulse length conditions. After the first pulse is finished, and before the sample has time to cool down to ambient temperature, a second pulse is applied. Because the baseline temperature of the sample is now higher than ambient, the second pulse drives T_p to a higher value. As this process is repeated, T_p increases until a multi-pulse saturation value is reached. For a given lamp voltage, multi-pulse saturation T_p is equal to or higher than the single-pulse saturation T_p depending on whether the sample cools down to the ambient temperature in between pulses. The multi-pulse saturation temperature and the number of pulses required to reach it depend on several factors, including pulse length, repetition rate (determined by power supply output), lamp voltage, and number of lamp drivers. For example, Figure 3D shows a multi-pulse saturation T_p of 306°C after four pulses for the 5 ms pulse length condition. In contrast, T_p does not saturate even after 10 pulses for the 1 ms pulse length condition (Figure 3C).

To ensure the maximum possible T_p is achieved, we first chose a pulse length of 3.5 ms and varied the number of pulses to fabricate the NiO_x HTL. Figure 4A shows the average (dashed red curve, closed symbols) and champion device (solid red curve, open symbols) reverse-scan PCE results as a function of number of pulses for 3.5 ms pulse length. The PCE values increase with the number of pulses and reach a saturation value of ∼9% above five pulses. Therefore, for fast processing speed while ensuring NiO_x is converted, the number of pulses can be reduced to five. Next, we fixed the number of pulses to five and varied the pulse lengths (Figure 4B). A similar trend is observed where PCE values increase significantly at short pulse lengths and saturate above ∼3.5 ms. The black curve and grey shaded area in Figures 4A,B represent the average reverse-scan PCE and one standard deviation, respectively, of devices with TA NiO_x HTL, i.e., TA/PC devices, made during the same time frame. Note that all PSCs in Figure 4 are made with PC MAPbI₃ active layers. Thus, we demonstrate that NiO_x HTLs photonic cured using five 3.5 ms light pulses delivered by 500 V lamp voltage at a repetition rate of 0.35 Hz, translating to 11.4 s of total processing time, produce PSC device performance comparable to NiO_x HTLs thermally annealed for 45 min.

To examine the chemical changes in the NiO_x film during photonic curing, XPS measurements were performed on NiO_x films after photonic curing with 500 V lamp voltage, 3.5 ms pulse length, 0.35 Hz repetition rate, and varying numbers of pulses, as with the devices shown in Figure 4A. The precursor film and thermally annealed film were also measured for comparison. The precursor film has a large N 1s peak at 408 eV due to the nitrate in the precursor solution. The top plot in Figure 5 shows that the area under this peak decreases significantly with an increasing number of photonic curing pulses, and is no longer present after five pulses. The thermally annealed NiO film also shows no N 1s signal. The Ni 2p_3/2 measurements and fitted curves are shown in Supplementary Figure S7. The data for each measurement was fitted to seven peaks. The four lower energy peaks are the primary peaks, and the three higher energy peaks are satellite peaks. The energy of the lowest energy peak (peak 1), centered at ∼854.1 eV (not present in the precursor or one pulse films), is attributed to the NiO chemical state (Mansour, 1994b). Peak 2 at ∼855 eV contains contributions from several possible Ni hydroxide species including α and ß-Ni(OH)₂, and NiOOH with Ni in the 2⁺ and 3⁺ oxidation states (Mansour, 1994a; Mansour and Melendres, 1994; Biesinger et al., 2009). Peak 3 (∼856.5 eV) contains contributions from both the hydroxide species as well as NiO due to multiplet splitting of the primary NiO peak (Mansour, 1994b; Boyd et al., 2020). Peak 4 at highest binding energy (∼858 eV) is attributed to Ni in the 3⁺ and 4⁺ oxidation states (Grosvenor et al., 2006). Because only NiO contributes to peak 1 and peak 2 arises from the various hydroxide species with no contributions from NiO, the relative areas of these two peaks, expressed as a percentage of the total area under the four primary peaks, are indicative of chemical transformations from the precursor film to various degree of conversion during photonic curing or after thermal annealing. The results are summarized in the bottom plot in Figure 5. The percent contribution of peak 1 is shown in blue (hashed bars) and that of peak 2 is shown in orange (solid bars). While the sum of the two peaks does not equal 100%, due to contributions from both species in the third and fourth peaks not represented in Figure 5, the percent contributions by two peaks are correlated with the concentration of NiO (blue) and hydroxides (orange) and are independent. Initially, there is no NiO present in the precursor film. After one pulse no NiO has formed (peak 1 remains at 0%), but the contribution of peak 2 to the total primary peak area decreases slightly. During this initial pulse, the nickel nitrate precursor is partially converted to nickel hydroxide, and the total concentration of all hydroxide species increases with their XPS signal shifted from peak 2 to peaks 3 and 4 (Supplementary Figure S7). After two pulses, there is a contribution from peak 1 and a large decrease in the area of peak 2 as the hydroxide species begin to convert to NiO. Additional pulses cause further conversion of hydroxides to NiO, and after 10 pulses the relative contributions from the two peaks are nearly identical to those in the TA sample. The XPS results provide an explanation for the device results in Figure 4A, which show that the PCE for the NiO_x HTL processed with five pulses is slightly lower compared to PCE values for NiO_x HTLs processed with 10, 20, and 30 pulses. Figure 5 shows that while the N 1s peak is not observed after five pulses, peaks 1 and 2 do not resemble the thermally annealed reference sample until 10 pulses. Therefore, the small difference in PCE values for NiO_x HTLs processed with 5 and 10 pulses likely arises from the lower conversion of NiO_x at five pulses, as evidenced by the lower peak 1 and higher peak 2 areas.

To explore whether the NiO_x HTL can be converted in a single pulse, which will dramatically reduce the total processing time, we employed a PulseForge^® 1300 with four 950 V lamp drivers, a 4.75 kW power supply, and a 24 mm diameter lamp. The WG/ITO/NiO_x samples were held down by two magnetic steel slats and photonic cured directly on the processing stage. Figures 6A,B show champion device J-V curves and average reverse-scan J-V parameters for PSC devices made in the TA/TA and PC700V/TA configurations. The “700V” sample label denotes that the NiO_x HTLs were photonic cured with a single pulse at 700V lamp voltage and 450 µs pulse length. The NiO_x samples photonic cured in 450 µs give device performance comparable to the NiO_x thermally annealed for 45 min at 250°C. Figure 6C shows a six-order-of-magnitude reduction in processing time when the NiO_x HTL was photonic cured with a single pulse.

When transitioning from 1 mm Glass/ITO to flexible WG/ITO substrates, we observed a drop in performance for both champion and average device performance. Upon further examination of the J-V data, there is a ∼13% decrease in J _SC when going to the WG/ITO substrate. Figure 7A shows the forward and reverse-scan J-V curves for PSCs made with TA NiO_x on the two substrates with the highest J _SC value we have measured on each substrate. To understand the discrepancy in J _SC, we compared UV-vis transmission spectra, surface roughness, and sheet resistances of both substrates in Figure 7B. The AFM images are shown in Supplementary Figure S8. While the surface roughness and sheet resistance are both higher for WG/ITO compared to 1 mm Glass/ITO, they cannot explain the large J _SC discrepancy. Because the NiO_x is thermally annealed on a hot plate, a quasi-equilibrium heating process, any performance differences arising from heat transfer or substrate thickness effects during the NiO_x annealing are negligible. UV-vis transmission spectra show a significant difference between the two substrates, with WG/ITO having a lower transmission, especially between 300 and 500 nm wavelengths. To determine whether the difference in optical transmission between the substrates is the reason for the decrease in J _SC, we determined the ITO optical constants using ellipsometry and then used TMM to calculate the maximum J _SC achievable on each substrate (Figure 7C).

Because both WG and 1 mm Glass are optically transparent and can be well fitted with a Cauchy model, the difference in the transmission spectra in Figure 7B must arise from the ITO layers. When fitting the spectroscopic ellipsometry data, the ITO films must be modeled as a graded layer where the optical constants, n and k, vary throughout the thickness of the ITO film. Fits for the ITO film on WG were achieved by modeling the ITO film as five, equally spaced, distinct layers (Supplementary Figure S4A). In contrast, the ITO film on 1 mm Glass only needed two layers in the grading model. The graded models showed good agreement with the measured spectroscopic ellipsometry parameters and optical transmission spectra. The optical constants for the ITO films on each substrate are shown in Supplementary Figures S4B,C. The extinction coefficient for the top-most ITO layer on WG is significantly higher throughout the 300–500 nm wavelength range, which contributes to the lower UV-vis transmission curve for WG/ITO (magenta) in Figure 7B. The n and k for the two layers in 1 mm Glass/ITO are similar, indicating a much more uniform film. The ITO layers on the two substrates are not deposited by the same manufacturer or the same method. The difference in the two materials likely originates from the ITO deposition processes. The WG/ITO was likely deposited using a R2R process where multiple layers of ITO were coated onto the WG sequentially (FraunhoferFEP, 2020).

Finally, we used the optical properties of each substrate to calculate the maximum J _SC possible using TMM. The WG/ITO substrate material stack was modeled as 100 µm WG and five 30 nm ITO layers. The 1 mm Glass/ITO substrate material stack was modeled as 1 mm soda lime glass and two 75 nm ITO layers. Both substrate materials stacks were modeled with the same top layers, as described in Simulations. Figure 7C shows an excellent agreement between our highest measured J _SC values (hashed bars) and TMM calculations (solid bars) using the models described above and shown in Supplementary Figure S4A. These results show that the optical properties of the ITO film on WG that is commercially available limit the achievable J _SC in PSCs. Higher J _SC and PCE may be achieved with a different ITO deposition process (Dou et al., 2017), but that is beyond this study.