Improved reproducibility of metal halide perovskite solar cells via automated gas quenching

Single junction metal halide perovskite solar cells have reached a certified efficiency of 26.1%, approaching the single junction record for crystalline silicon cells.^1,2 Perovskite active layers can be deposited through solution processing and annealed at modest temperatures, permitting the use of inexpensive large scale deposition techniques.³ However, minor changes in the solution fabrication process can change the optoelectronic quality of the resulting perovskite film.^4,5 It is popular among the perovskite community to dispense an antisolvent during film drying to rapidly remove the precursor solvents and form a compact perovskite thin film.⁶ However, there are many variables that can be challenging to sufficiently control when the antisolvent is manually dispensed, such as the timing and location of the pipette tip. Furthermore, as the vapor pressure of the antisolvent builds up in an enclosed environment, such as an N₂-filled glove box, film drying rates can change, which may then result in changing film quality. Despite these problems, manually dispensing antisolvents is still the most commonly used quenching method in the perovskite community for small area devices.^1,7,8

Gas quenching is an alternative quenching method to using an antisolvent and is more widely used for large area devices (>1 cm²).⁹ Utilizing an inert gas flow, such as nitrogen, to force film nucleation instead of an antisolvent has several benefits. First, there are no antisolvent vapors that could alter the fabrication atmosphere for subsequent substrates,^6,9 making gas quenching processes more robust with respect to the processing atmosphere.⁷ The gas quenching process is also simpler to control with automation and has a larger processing window.^10–13 Finally, gas quenching is more compatible with blade or slot-die coating.^3,12,14–17

When we attempted to reproduce gas quenching procedures reported in the literature, we found that in most cases, the authors described their process by reporting the pressure they used.^7,18–20 Since this pressure is measured at the regulator on the nitrogen source, it does not fully describe a flowing system. The pressure at the exit is heavily dependent on the length and diameter of the tubing used in the pneumatics and whether there are sharp angles in them. Therefore, two of the three parameters, velocity, flowrate, and outlet area, must be reported to fully describe the gas flow and account for differences in pressure drop between the nitrogen inlet and the gas quench nozzle outlet.^5,15,21

When studying how processing parameters affect perovskite film formation, it is helpful to monitor the process in situ to have information on the rate at which the perovskite forms. We chose in situ photoluminescence (PL) because it provides valuable information about the evolution of the bandgap and the growth of crystallites of the perovskites and the time scale at which they form.²² Even though in situ PL cannot conclusively identify the composition of intermediate phases during film formation, it is extremely valuable for monitoring a process to ensure that it operates consistently and identifying process parameters that change film formation kinetics.^23–25 While in situ x-ray diffraction is arguably more powerful, it is not feasible to perform it using a spin coater in a glove box. Since the vapor pressure that builds up in a small chamber used in an x-ray beamline is so different from that in a glove box, we prefer to have in situ PL in the glove box that we routinely use for device fabrication. We also prefer using the same spin coater and quenching equipment that we normally use for device fabrication instead of using slightly different equipment found in a synchrotron x-ray facility.

This study shows that the antisolvent quenching process is very sensitive to the solvent atmosphere, while the gas quenching process is decoupled from the solvent atmosphere.²⁶ We report gas quench device efficiencies that match those of antisolvent devices but with a narrower standard deviation and an improved yield. The improvement in device reproducibility is caused by the reduction in human error surrounding the quenching process and consistent film formation kinetics.

The perovskite composition used for this study, Cs_0.3FA_0.6DMA_0.1Pb(I_0.8Br_0.2)₃, where FA is formamidinium and DMA is dimethylammonium, is a 1.73 eV compound developed for all-perovskite tandems.^27,28 A similar composition has reached high efficiency in both single junction and all-perovskite tandem cells by using the gas quenching method.²⁹ The standard device stack used for all devices is a p–i–n structure shown in Fig. S1. Our fabrication methodology for both the antisolvent method and the automated gas quenching method are provided in the supplementary material. Figure S2 shows the photos of two automated gas quenching systems utilized in this study and a piping and instrumentation diagram (P&ID) showing all interconnections and valves used to control the system.

To achieve reproducible devices, film formation at the point of crystallization must be kept consistent to produce similar films. The quenching process and local fabrication atmosphere must be regulated to have consistent film formation kinetics.^30,31 As a device fabrication session progresses, the spinning atmosphere becomes saturated with precursor solvents, such as dimethylformamide (DMF), unless the spinning atmosphere is cleaned.³² We studied film formation for both the antisolvent and gas quenching methods using in situ PL with a 405 nm LED excitation source.³³ We can get information on the rate of film formation by analyzing the time scale at which the spectrum evolves.²³ The in situ PL methodology and analysis techniques are detailed in the supplementary material. We observe a rather complex behavior, including the formation of two PL peaks. It is beyond the scope of this study to fully understand the film formation process. Additional techniques, such as in situ XRD, that provide more information about the structure of the film would be needed.

For films quenched with an antisolvent, the film formation kinetics drastically change when DMF is introduced into the atmosphere, as shown in Fig. 1. Films were quenched with methyl acetate in varying solvent atmospheres: a fresh atmosphere, a DMF-rich atmosphere, and a cleaned atmosphere. We consider the atmosphere to be fresh when we spin the first substrate during a fabrication session and there are no solvents present in the atmosphere from previous substrates. In a fresh atmosphere [Figs. 1(a) and 1(b)], two PL peaks emerge ∼20 s after the antisolvent quench. However, when DMF is present in the atmosphere, there are never two PL peaks present during any portion of film formation, as shown in Figs. 1(c) and 1(d). In addition, the final PL peak location differs between the fresh environment (726 nm) and the DMF environment (702 nm). DMF introduction is detailed in the supplementary material. The changing kinetics of film formation have a lasting effect on the structure of the antisolvent quenched perovskite films. XRD was conducted after the films were spun in varying solvent environments before and after annealing. As shown in Fig. S3, the structure of the films prior to annealing changes when DMF is introduced. After annealing, the differences are maintained as the intensity of the PbI₂ peak changes as a function of the solvent atmosphere.

The original in situ PL transients can be recovered once DMF is removed by thoroughly cleaning the spinning atmosphere with nitrogen [Figs. 1(e) and 1(f)]. A cleaned environment is created by introducing DMF, as detailed in the supplementary material, and then removing the solvent vapor by blowing out the spin coater bowl using a nitrogen blowgun. The PL transients of the cleaned and fresh environments show a consistent pattern and end at very similar peak locations, 725 and 726 nm, respectively. A double peak is present starting at 20 s after the quench for both environments, also indicated by the increase in full-width at half maximum (FWHM) values. However, the structural changes do not recover by cleaning the environment. As shown in Fig. S3, the cleaned environment film structure resembles that of the DMF environment. Unless strict care is taken to maintain a clean atmosphere, the film formation kinetics and the resulting film homogeneity of antisolvent films will drift during a fabrication session.

Gas quenching removes the need to manually purge the spinning atmosphere between each substrate because the quench naturally cleans the fabrication atmosphere. When the excess solvent vapors are removed during the quenching process, the film formation process becomes decoupled from the fabrication atmosphere. Based on in situ PL transients, the film formation kinetics of gas quenched perovskite films do not significantly change when DMF is introduced into the atmosphere, as shown in Fig. 2. The PL transients in different environments both show double peaks early in film formation, also indicated by the high FWHM values in the first 10 s. The final PL peak location is also consistent between the fresh (665 nm) and DMF (663 nm) environments. While the film formation kinetics remain consistent, introducing DMF does change the film structure prior to annealing even when utilizing gas quenching, as shown in Fig. S4(a). However, unlike antisolvent quenching, the film structure is consistent after annealing in varying solvent atmospheres, as shown in Fig. S4(b). The combination of consistent in situ PL transients and consistent post-annealing film structure demonstrates the improved reproducibility of gas quenching.

Prior to process optimization, processing variables that affect film formation dynamics must be identified. We have already determined that the precursor solvent atmosphere does not significantly change film formation dynamics during gas quenching. However, the effect of varying the gas quenching velocity on film formation dynamics needs to be explored. In situ PL can be used to explore the differences in film formation as a function of gas quenching velocity, which can be manipulated by changing the flowrate or the outlet area. To determine the effect of flowrate, we varied the flowrate from 2.0 to 4.0 scfm and monitored the changing peak positions of the PL transients. In situ PL transients are shown at 2.0 and 4.0 scfm in Figs. S5(a) and S5(b), respectively. Changing in situ PL transients suggests that film formation/nucleation dynamics are dependent on the gas quenching flowrate. At low flowrates, two peaks are present in the first 10 s of film nucleation during the quench. However, as the flowrate increases, the two peaks become less distinct and are present for less time.

The gas quenching velocity can be increased by increasing the nitrogen flowrate or by narrowing the outlet diameter. By using different outlet nozzles, the drying pattern is also changed along with the velocity. supplementary material note 1 and Fig. S6 detail experiments testing manipulation of the drying pattern and gas quenching velocity based on different nozzles. Overall, similar changes to in situ PL transients occur if gas quenching velocity is changed by the flowrate or the outlet diameter. Since the gas quenching velocity can be independently manipulated by either the flowrate or the outlet diameter, researchers should report both the gas flowrate and details of the exit orifice (shape and size) to maximize lab-to-lab reproducibility. Unless noted otherwise, all experiments were done using a circular straw nozzle with an outlet area of 0.057 in².

The only experimental details regarding gas quenching in most studies are the pressure on the nitrogen bottle and, sometimes, the distance between the nozzle and the substrate.^7,20,29 Since the flowrate at the substrate also depends on the pressure drop in the nitrogen line along with the size and shape of the nozzle, it is helpful to reoptimize the nitrogen flowrate and nozzle height in each laboratory. With the ideal gas quenching parameters, devices can be reproducibly made by different researchers across different batches.

In short, three main steps were followed to optimize the performance and reproducibility of the automated gas quenching system. First, a general sweep of flowrate and straw height was performed to identify preliminary conditions that produced high-performing devices, as shown in Fig. S7. Next, the solution concentration was varied so that the thickness of the gas quenched films matched that of the antisolvent derived films. The gas quenching solution concentration had to be reduced such that the thickness matched that of the antisolvent films (Fig. S8). Solution concentration is the main variable that is manipulated to optimize film thickness as the gas quench flowrate and nozzle height do not significantly change film thickness, as shown in Fig. S9.³⁴ 

The final step to identify the ideal set(s) of parameters is to compare the top performing combinations using the ideal solution concentration. The ideal set(s) of parameters are identified by calculating the yield and the number of devices within one standard deviation of the mean. There may potentially be multiple sets of parameters that produce similar results due to the wide processing range of gas quenching. Yield is defined as the ratio of all working devices to all devices made. We consider a device to be working if the fill factor (FF) is >40% and the short circuit current density J_SC is >2 mA/cm². The mean value of all working devices is calculated, and all devices within one standard deviation of the mean are identified. The data analysis that we utilize to identify the ideal parameters is provided in more detail in the supplementary material. The ideal gas quenching parameters are identified based on performance and standard deviation, as shown in Fig. 3(a). The ideal parameter maintains the highest average performance and lowest standard deviation. Due to the automation of gas quenching, this process can be easily transferred between researchers and still achieve a high performance. The current density–voltage (JV) curves for devices made by different researchers show similar mean values for all performance metrics, as shown in Figs. 3(e) and 3(f). The device performance metrics for the each JV curve shown are tabulated in Table S1.

Previously, we observed that some researchers obtain a much smaller standard deviation in solar cell efficiency than others when they manually dispense an antisolvent. We hypothesized that automated gas quenching would enable different researchers to obtain similar results. To test this hypothesis, an expert in the antisolvent and gas quenching method trained a novice researcher. The results from all device makers are shown in Fig. 4. For the antisolvent method, the expert device maker had a higher performance in four all major metrics (PCE, FF, V_OC, and J_SC) by approximately one standard deviation. In addition, the expert device maker had 5% absolute higher yield in working devices compared to the novice device maker. The discrepancy in performance between the expert and novice device makers is presumably due to the small processing window of the antisolvent method and less learned control of the novice device maker. However, for the automated gas quenching method, the device performance and yield were very similar for the expert and novice device maker. In addition, the yield of working devices was higher for the automated gas quenching devices compared to antisolvent devices (96% compared to 72%). The consistent device performance and higher yield across different experience levels are due to the wide processing range of gas quenched films, consistent film formation substrate to substrate, and the automated nature of the quenching system that reduces human error.

Finally, batches of devices made by expert device makers for both quenching methods, respectively, were made on different days to compare the batch-to-batch reproducibility. Figure 5 shows the device performance across the different batches for both quenching methods. While there is some level of batch-to-batch performance variation for each quenching method, the gas quenched devices are more consistent than the antisolvent devices. The separate gas quenching batches show consistent yield, standard deviation, and percentage of devices within one standard deviation of the mean. However, the yield and standard deviation for antisolvent devices show significant variation between the different batches. In addition, utilizing gas quenching instead of an antisolvent improves the yield of working devices and percentage of devices within one standard deviation of the mean. Across two batches of devices made with the antisolvent method, the average yield is 88% and the average number of devices within one standard deviation of the mean is 83%. However, across two batches of devices made using the gas quenching method, the average yield and devices within one standard deviation of the mean increased to 98% and 96%, respectively. Overall, the automated gas quenching method produces more consistent high performing devices compared to the antisolvent method.

In summary, we have shown that by implementing an automated gas quenching system, high performing devices matching antisolvent devices can be reproducibly fabricated with a high yield and high percentage of devices within one standard deviation of the mean. Gas quenching improves reproducibility between substrates because the film formation kinetics remain consistent even in the presence of precursor solvents. By monitoring the film formation with in situ PL, the gas quench flowrate was identified as an important parameter that can change the film formation of the perovskite. The film formation changes in a predictable manner if the velocity is manipulated by changing either the flowrate or the outlet area. Perovskite devices can be reproducibly fabricated across different batches of devices using the automated gas quenching method because human error is minimized, and the film formation kinetics are less sensitive to the surrounding atmosphere.

The supplementary material includes a detailed description of the automated gas quenching setup, materials and fabrication methods used, film thickness measurements, additional device performance figures (PDF), and videos of films drying with different nozzles (video).

This material is based upon the work supported by the U.S. Department of Energy’s Office of Energy Efficiency and Renewable Energy (EERE) under Solar Energy Technologies Office (SETO) Agreement No. DE-EE0009513 and the National Renewable Energy Laboratory (NREL) operated by the Alliance for Sustainable Energy, LLC, for the U.S. Department of Energy (DOE) under Contract No. DE-AC36-08GO28308. The authors acknowledge the financial support from the U.S. Department of Energy’s Basic Energy Sciences (BES) Award No. DE-SC0022305. This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.

McGehee is an advisor to Swift Solar.

Samantha C. Kaczaral: Conceptualization (equal); Data curation (equal); Methodology (equal); Writing – original draft (equal). Daniel A. Morales, Jr.: Conceptualization (supporting); Formal analysis (supporting); Methodology (supporting); Writing – review & editing (equal). Samuel W. Schreiber: Conceptualization (supporting); Formal analysis (supporting); Methodology (supporting). Daniel Martinez: Data curation (supporting). Ashley M. Conley: Data curation (supporting). Randi Herath: Data curation (supporting). Giles E. Eperon: Data curation (supporting); Writing – review & editing (supporting). Joshua J. Choi: Supervision (supporting); Writing – review & editing (supporting). Michael D. McGehee: Conceptualization (equal); Funding acquisition (lead); Supervision (equal); Writing – review & editing (equal). David T. Moore: Conceptualization (equal); Data curation (equal); Formal analysis (lead); Methodology (equal); Supervision (equal); Validation (lead); Writing – review & editing (equal).

The data that support the findings of this study are available within the article and its supplementary material.