High-value organic solvent recovery and reuse in perovskite solar cell manufacturing

Abstract

Recycling high-value organic solvents is crucial but challenging in various industries. For example, the perovskite solar cell (PSC), a rising star of photovoltaic industry, calls for proper management of solvents like N,N-dimethylformamide (DMF). Traditional solvent recovery methods are often less effective, costly, and energy-intensive. To address this, we developed a multistage air-gap membrane distillation (MAMD) system that efficiently recovered DMF from waste solutions using industrial waste heat. Our MAMD system achieved a DMF enrichment factor up to 314 (increasing the concentration from 0.3 to 94.2 weight %) and stable operation over 60 hours. The recovered DMF (94.2 weight %) was used in perovskite minimodule fabrication, achieving a certified stabilized power output of 19.97%. The narrow efficiency deviation, state-of-the-art power conversion efficiency, and small hysteresis demonstrated the viability of using the recovered DMF in industrial fabrication. These results demonstrate the potential of our MAMD system to minimize the environmental footprint and promote sustainable PSC manufacturing.

A MAMD system enables efficient DMF recycling for sustainable PSC fabrication, revolutionizing the organic solvent management.

INTRODUCTION

Organic solvents play a vital role in various industrial processes, serving as versatile media for chemical reactions, separations, and material processing. Their significance spans diverse industry sectors such as solar cell manufacture (1–4), lithium battery fabrication (5, 6), pharmaceutical production (7, 8), and fine chemical synthesis (9); and demands for organic solvents will continue to increase rapidly with the remarkable growth of modern industries. The solar cell and lithium battery industries, which are two key pillars of renewable energy industries, taken for example, are estimated to reach a total revenue of 305 billion USD in China alone by 2030 (10, 11). An average estimation is that waste organic solvent generation worldwide will reach 4.15 million metric tons by 2025 with an annual growth rate of ~3.5% (12). The extensive use and continuous growth in demand for organic solvents have inadvertently led to notable challenges in terms of waste treatment and have brought a detrimental impact to environmental sustainability by industries that are meant to fight against climate change (13).

Perovskite solar cells (PSCs) have emerged as a rising star in the photovoltaic industry because of their cost-effectiveness and impressive photoconversion efficiency, exceeding 26% on a laboratory scale (14). This technology is expected to grow remarkably, with newly-added production capacity expected to rise from 0.4 GW in 2023 to more than 160 GW by 2030 in China alone (15). High-value organic solvents are premium-priced and crucial for enhancing product performance in high-value-added industries. For example, N,N-dimethylformamide (DMF), dimethyl sulfoxide (DMSO), and N-methyl-2-pyrrolidone (NMP) are essential for solution-processed halide perovskite fabrication. However, the organic solvent DMF is listed as a substance of very high concern by the European Chemical Agency because of its potential environmental and reproductive toxicity (16). Furthermore, on the basis of calculation from a typical 100-MW pilot line, DMF takes a considerable proportion of the Bill-of-Materials cost in PSCs, and a 50% recycling rate could reduce the Bill-of-Materials cost by at least 2000 USD/MW. Microquanta estimates that every 1 GW of perovskite production capacity consumes 312 tons of DMF solvent. Waste DMF is then captured in vapor form by the exhaustion system and absorbed into water through a spray tower, generating plenty of dilute waste solutions. Proper management of these organic solutions hence becomes an urgent matter.

Recent attempts have been made to minimize the use of toxic organic solvents in PSC fabrication to reduce their environmental impact. While substituting conventional solvents like DMF with green alternatives (e.g., acetonitrile or alcoholic solvents) shows promise at the laboratory scale, it remains challenging to achieve large-area uniform coating with these solutions (17–19). Vacuum-based deposition methods provide great potential by eliminating toxic solvents entirely and demonstrating effectiveness particularly in perovskite-silicon tandem applications, where vacuum deposition facilitates the conformal growth of the perovskite layer on textured silicon surfaces (20–22). Nevertheless, solution-based deposition using toxic solvents remains popular in lab-scale monolithic PSC fabrication owing to high device efficiencies (23–26). It also benefits from the lower initial capital investment for the solution-based coating machinery. As a result, several pilot production lines of large-area single-junction PSCs have put the solution-based deposition into practice (27). Therefore, the reliance of this method on toxic and high-value solvents poses environmental and economic challenges, which makes it crucial to consider appropriate handling of waste organic solvents in solution-based PSC manufacturing.

Current practices for waste organic solvent disposal, such as on-site incineration or deep-well injection, are not only wasteful but also harmful to the environment. With stricter environmental legislation and increasing prices of organic solvents, high-value organic solvent recovery becomes more attractive than simple disposal (28–30). Conventional recovery methods such as distillation, adsorption, extraction, and membrane filtration are often less effective, costly, and energy-intensive, leading to considerable environmental footprint (31). For example, thermal distillation is economically unfeasible for low-concentration organic solutions because of excessive energy demand (32, 33). Liquid-liquid extraction is typically applicable for feed solutions with high water content (e.g., 99 wt %), which necessitates specific extractants and generates large extract streams requiring further separation (33–35). Membrane filtration methods like nanofiltration suffer from low permeability, poor selectivity, and high pressure demands (36). Therefore, researchers and solvent-intensive industries are exploring innovative techniques for efficient, energy-saving, and environmentally friendly high-value organic solvent recovery.

In this work, we developed a multistage air-gap membrane distillation (MAMD) system to efficiently recover high-value organic solvents from waste solutions containing binary volatile components (organic solvent and water) (37, 38), and its feasibility was validated by implementation in DMF recycling of solution-based PSC fabrication. The MAMD system for solvent recovery in this study differs from membrane distillation desalination [where soluble salts can hardly evaporate (39)] by using vapor pressure differences between the organic solvent and water. This system is driven by low-grade industrial waste heat and also maximizes latent heat utilization through multistage vaporization to reduce greenhouse gas emissions. In this study, we systematically investigated the separation mechanism of binary volatile components and impacts of key operation parameters on the DMF enrichment performance of the MAMD system including input powers, feed solution concentrations, and inlet flow rates. These parameters are controllable and critical for optimizing the DMF enrichment and recovery process. We also optimized the hydrophobic membrane pore size to obtain better enrichment performance. After the validation of DMF recovery performance, we further discussed the versatility of the MAMD system to other high-value solvents. In addition, the multiunit MAMD system was designed by integrating multiple units for varying requirements and a scaled-up MAMD system was built up for DMF enrichment and recovery. Last, the recovered DMF was successfully used to fabricate perovskite minimodules, with performance analysis confirming the impact of the DMF concentration on perovskite minimodule quality. Our work established a sustainable recycling loop for high-value organic solvents such as DMF, encompassing their consumption, recovery, and reuse in various industries including solar cell manufacture, semiconductor fabrication, pharmaceutical production, and fine chemical synthesis (Fig. 1A).

Fig. 1. DMF organic solvent recovery using a MAMD system.

(A) Schematic illustration of the MAMD system for DMF organic solvent recycling showcasing its applicability in industrial production processes including solar cell manufacture, semiconductor fabrication, pharmaceutical production, and fine chemical synthesis. Specifically, the production capacity of PSC and the associated DMF solvent consumption and recycling are given as examples. (B) Layout of a MAMD system in the cross-flow mode, with the effluent from the previous stage flowing into the subsequent stage following the direction indicated by the solid red arrow and the distillate from each stage following the direction indicated by the dashed blue arrow. Close-up illustration: The feed solution is gradually enriched with the DMF solvent when flowing through the evaporation layer as the H₂O vapor evaporates and diffuses through a hydrophobic membrane to form the distillate.

RESULTS

Layout and working principle of the MAMD system

In our design, a MAMD system included five stages, each consisting of four primary components: a thermal conduction layer, an evaporation layer, a hydrophobic porous layer, and a vapor condensation layer with an air gap (Fig. 1B). The thermal conduction layer in each stage had a thermal conductivity of 16.3 W m⁻¹ K⁻¹ for efficient heat transfer to the next layer. The evaporation layer beneath the thermal conduction layer was where the DMF feed solution flowed through. The hydrophobic porous layer, separating the evaporation layer and the vapor condensation layer, ensured the vapor molecule permeation. In direct contact with the top thermal conduction layer of the subsequent stage, the vapor condensation layer allowed for the release of latent heat of vaporization to drive evaporation in the next stage. To prevent heat loss into the surrounding environment, the top and side surfaces of the MAMD system were thermally isolated with polyurethane sponges (thermal conductivity of 0.020 to 0.030 W m⁻¹ K⁻¹).

During each stage of the MAMD system, heat was transferred from the top thermal conduction layer to the evaporation layer to drive the heat-up and evaporation processes of the DMF feed solution. The DMF and H₂O vapor molecules generated from the evaporating interface subsequently diffused downward and passed through the micropores of the hydrophobic membrane. These vapor molecules then condensed to form the distillate in the vapor condensation layer. The distillate accumulated in the condensation layer then flowed into the distillate container by the wet hydrophilic nonwoven fabric, offering the potential to serve as a sustainable water source in a closed-loop PSC production process. The air gap formed between the hydrophobic membrane and the condensing interface enabled limited conductive heat loss (air thermal conductivity of ~0.027 W m⁻¹ K⁻¹), facilitating the heat transfer in the form of latent heat along with vapor diffusion (40–42). The latent heat of vapor molecules released from the upper stage condensation layer was recovered and used as the heat source in the subsequent stage to maximize the energy efficiency (43–45).

In this work, flow channels in the evaporation layer were designed to minimize the short flow phenomenon (46, 47). These channels enabled the establishment of a horizontal DMF concentration gradient along the evaporation layer, effectively maximizing the utilization of the evaporation area (fig. S1). Consequently, the feed solution in the evaporation layer became progressively enriched with DMF as it flowed through the channel, driven by the preferential evaporation of H₂O (Fig. 1B). The concentrate eventually exited from the top of the MAMD system after undergoing a five-stage separation and enrichment process. In addition, the modular design of the MAMD system allowed for the selective assembly of different numbers of units depending on the desired DMF concentration and specific application. The concentrate from the previous MAMD unit served as the feed solution in the next unit, where further enrichment took place.

DMF enrichment and recovery performance

To assess the DMF enrichment performance of the MAMD system, laboratory experiments were conducted under controlled conditions (Fig. 2A). Notably, input powers, feed solution concentrations, and inlet flow rates are critical operation parameters affecting DMF recovery. DMF concentrations in the outlet and distillate were measured (table S1). As the input power rose from 2.8 to 5.0 W, the outlet DMF concentration rose across inlet concentrations of 5, 10, 20, 40, and 60 wt % (Fig. 2B). The DMF enrichment factor (outlet-to-inlet DMF concentration ratio) peaked at 2.76 (2.8-W input power), 3.40 (4.0-W input power), and 3.66 (5.0-W input power) at a 5 wt % inlet concentration, gradually converging to 1 as it increased to 60 wt % (Fig. 2B). The top surface temperature and temperature difference between the top and bottom surfaces of the MAMD system showed a gradual increase as the inlet concentration became higher under the same input power condition (Fig. 2C and fig. S2). Specifically, under the 5.0-W input power condition, the MAMD system achieved ultrahigh outlet concentrations of 18.3, 31.1, 49.4, 65.8, and 77.7 wt % (with corresponding enrichment factors of 3.66, 3.11, 2.47, 1.65, and 1.30) (Fig. 2B). In this case, the top surface temperature of the MAMD system was 71.1°C for a 5 wt % inlet concentration and increased to 77.6°C when the inlet concentration increased to 60 wt % (Fig. 2C). Furthermore, controlling inlet flow rates demonstrated a trade-off between the outlet concentration and the concentrate yield (fig. S3). The mass flow rate ratio of the distillate to the concentrate (Q_dis/Q_con) decreased from 10.72 ± 0.61 to 7.92 ± 2.08 as the inlet flow rate increased from 9.60 to 9.98 g/hour. This resulted in a decrease in the outlet concentration (55.9 ± 1.3 to 51.3 ± 0.2 wt %) but an increase in the concentrate yield (0.82 ± 0.04 to 1.15 ± 0.27 g/hour).

Fig. 2. DMF enrichment and recovery performance evaluation of the MAMD system.

(A) Photo of an experimental setup for DMF enrichment and recovery performance test with a MAMD system. The blue dashed circle indicated the close-up illustration of a MAMD system (7 cm by 7 cm). T and H represent temperature and humidity, respectively. (B) DMF enrichment performance and (C) temperature variations at the steady state of the MAMD system under different operating conditions. DMF solutions with concentrations of 5, 10, 20, 40, and 60 wt % were used for the tests. Solid and dashed black lines represent the top (T_top) and bottom surface temperatures (T_bot) of the MAMD system. (D) Comparison of the double-sided PTFE hydrophobic membranes with different pore sizes of 0.22, 0.45, and 1.0 μm under the same operating condition. (E) Long-term operation reliability test of DMF organic solvent enrichment through the MAMD system. The time interval of each measurement was 12 hours.

We optimized the membrane pore size to achieve better DMF enrichment performance from the MAMD system. Better enrichment performance was observed using the commercial double-sided polytetrafluoroethylene (PTFE) membrane with a larger pore size (Fig. 2D). The DMF outlet concentration improved from 62.2 ± 2.7 wt % (pore size of 0.22 μm) to 65.3 ± 2.4 wt % (pore size of 0.45 μm) and then marginally improved to 66.4 ± 1.3 wt % (pore size of 1.0 μm) under the same operating condition. On the basis of comparable DMF outlet concentrations for the membranes with 0.45- and 1.0-μm pore sizes, we chose 0.45 μm as the optimal pore size because of its lower wetting possibility and higher concentrate yield than those of the larger one. Beyond DMF recovery, the MAMD system demonstrated exceptional performance in enriching other high-value organic solvents, achieving enrichment factors of 3.59 for DMSO and 3.89 for NMP (fig. S4). These results confirmed the system’s versatility and potential for broader application in solvent recovery processes. After the validation of the solvent enrichment performance of our MAMD system, we further evaluated its long-term reliability. During the long-term operation test, the DMF outlet concentration increased from 20 to 56.4 ± 2.1 wt % after the first 12-hour operation and remained at 53.4 ± 4.2 wt % after the 60-hour operation. These results indicated the durability of the membrane and the stable enrichment performance of the MAMD system throughout a long-term test (Fig. 2E and table S2).

In this study, numerical simulations were also carried out to confirm the experimental findings of the DMF enrichment and recovery performance of the MAMD system (Fig. 3; the simulation model and governing equations in fig. S5 and Supplementary Text S1). The temperatures of the MAMD system were monitored using 10 thermocouples inserted in the evaporation layer and vapor condensation layer of each stage and validated by numerical models (Fig. 3, A and B; additional temperature and DMF concentration profiles in figs. S6 and S7). The simulated temperature decreased from 51.6° to 36.3°C as the distance between the top of the MAMD system and the temperature sensing point increased from 1 to 20 mm (under operating conditions of a 2.8-W input power, 40 wt % inlet concentration, and 3.96 g/hour inlet flow rate). This result was in good agreement with our experimental data (Fig. 3C). Furthermore, the experimental and simulated results indicated a consistent trend of increasing outlet concentrations of DMF with higher inlet concentrations under 2.8-, 4.0-, and 5.0-W input power conditions (Fig. 3, D to F, and fig. S8).

Fig. 3. Experimental and numerical simulation results of the DMF enrichment and recovery performance under various conditions.

(A) Photo of the layout of the MAMD system with 10 inserted thermocouples in the evaporation and vapor condensation layers of each stage. (B) Numerical simulation of the temperature profile. The white dots represent positions of 10 thermocouples in experiments. (C) Temperature changes as the distance between the top of the MAMD system and the temperature sensing point in the evaporation and vapor condensation layers of each stage in experimental and simulation analyses. (D to F) Outlet concentrations obtained in experiments and simulations under input power conditions of 2.8 W (D), 4.0 W (E), and 5.0 W (F).

To further increase the outlet concentration of DMF, the enrichment performance of the multiunit MAMD system was evaluated under three different input power conditions (Fig. 4, A and B). To achieve a final outlet concentration of more than 80 wt % from a 5 wt % DMF feed solution, a series assembly of six MAMD units was required under a 2.8-W input power (Fig. 4C). Fewer units were required (five MAMD units) with a higher input power (4.0 W) and further decreased to four MAMD units with a 5.0-W input power (Fig. 4, D and E). Hence, highly efficient DMF enrichment performance was achieved with a multiunit MAMD system working under the 5.0-W input power, surpassing that under 2.8 and 4.0 W (Fig. 4E). In this case, the enrichment factor of an individual unit ranged from 3.66 of unit 1 (the first one) to 1.09 of unit 5 (the fifth one). The DMF concentration increased from 5 to 18.3 wt % after flowing through unit 1; went to 46.0 wt % in the outlet of unit 2, 72.8 wt % after unit 3, and then 86.3 wt % after unit 4; and eventually reached 94.2 wt % at the exit of unit 5. Notably, the main scope of this work is to establish a closed-loop system for DMF recycling, with specific emphasis on the manufacturing of PSCs. During the solvent capture process, dissolved perovskite precursors can hardly evaporate with the DMF vapor and enter organic wastewater. The initial DMF concentration was set at 0.3 wt % following an assessment of the actual wastewater quality metrics (table S3). An outlet concentration of 1.6 wt % was achieved with a 0.3 wt % feed solution, which could subsequently be further enriched to more than 5.0 wt % (table S4). Consequently, the DMF concentration increased to 94.2 wt % from an initial 0.3 wt % dilute waste solution under the 5.0-W input power condition, exhibiting an overall enrichment factor of 314. Compositions of DMF samples in different concentrations (20, 50, and 95 wt %) heated at 50° and 80°C were further analyzed using ultra performance liquid chromatography–mass spectrometry (UPLC-MS). The results showed only one peak ionic intensity corresponding to DMF, indicating that no obvious DMF hydrolysis occurred during the MAMD operation (figs. S9 and S10). Even if hydrolysis were to occur, the formation of dimethylamine may not adversely affect the performance of perovskite films and resulting devices (48).

Fig. 4. DMF enrichment and recovery performance of the multiunit MAMD system.

(A) Schematic illustration of DMF stepwise enrichment using a multiunit MAMD system. (B) Photo of an experimental setup for the multiunit MAMD system. (C to E) DMF enrichment performance during laboratory testing of the multiunit MAMD system under 2.8-W (C), 4.0-W (D), and 5.0-W (E) input power conditions with an initial DMF concentration of 5 wt %.

Industrial waste heat utilization in a MAMD system for DMF enrichment and recovery

The emission of low-grade heat energy is a major component of energy loss in industrial production, which has enormous potential for further utilization (49). Various forms of industrial waste heat, such as exhaust gases exiting recovery devices (70° to 230°C), cooling water (60° to 90°C), and hot processed liquids (30° to 230°C) (50), could provide cost-effective and sustainable energy sources for the MAMD system (51, 52). In this work, a scaled-up MAMD system (23.5 cm by 23.5 cm) was developed as a proof of concept and used for DMF enrichment and recovery, driven by real industrial exhaust gases or circulated hot water (Fig. 5, A and B). For an inlet concentration of 5 wt %, the specific mass flux of the distillate in the scaled-up configuration was calculated to be 0.49 g/(hour·cm²) at a top surface temperature of 70.7°C (fig. S11). It had 1.44-fold enhancement compared to that of the smaller configuration of 0.34 g/(hour·cm²) under a similar heating condition (at a top surface temperature of 71.1°C). The Q_dis/Q_con of the scaled-up configuration was controlled to be 39.60 at a 5 wt % inlet concentration, resulting in a corresponding outlet concentration of 26.6 wt % and a concentrate yield of 5.24 g/hour (Fig. 5C). In comparison, a higher concentrate yield of 17.16 g/hour was obtained when the Q_dis/Q_con dropped to 9.80 at a 25 wt % inlet concentration, resulting in an outlet concentration of 64.1 wt % (Fig. 5, C and D).

Fig. 5. Scaled-up MAMD system for DMF enrichment and recovery driven by industrial waste heat.

(A and B) Photos of the experimental setup of a scaled-up MAMD system using low-grade waste heat from industry such as industrial exhaust gases (A) and waste hot water (B). The orange solid box indicates the close-up illustration of the industrial exhaust gas connection pipe (A) and the scaled-up MAMD system with a 23.5-cm width (B). (C) DMF enrichment performance of the scaled-up MAMD system. (D) Temperature evolutions of the top (T_top) and bottom (T_bot) surfaces of the scaled-up MAMD system, circulated water outlet (T_out), and ambient environment (T_A) during the simulated industrial waste heat utilization test.

Recovered DMF solvents for reuse in the fabrication of perovskite minimodules

The subsequent reuse of the recovered DMF solvents in industrial production constitutes the last phase of the organic solvent cycle. Herein, DMF in different concentrations recovered from each unit of the MAMD system was used in the preparation of perovskite precursors (fig. S12). Apparently, only DMF with high concentrations, i.e., 86.3 and 94.2 wt %, could completely dissolve the perovskite solute alongside the 99.5 wt % commercial DMF solvent. Therefore, the recovered 86.3 and 94.2 wt % DMF solvents were used for subsequent fabrication of perovskite minimodules (Fig. 6A and fig. S12). The blade coating method was implemented to deposit the perovskite layer to mimic the slot-die coating process in industrial fabrication. It is worth mentioning that the recipe we use (a full coating process in an atmospheric environment) may not achieve the highest laboratory efficiency but offers scalability for large-scale industrial production. This aligns well with our objective of this work to develop an environmentally friendly DMF recycling technique for direct industrial application.

Fig. 6. Photovoltaic performance and stability of the as-prepared perovskite minimodules.

(A) Front-side view of the minimodule covered on the edge with a metal mask. The orange solid square indicated the zoom-in view of the P1-P2-P3 laser cut line and the width of the dead zone. (B) Summary of efficiency distribution, including maximum, minimum, and average values under forward bias (FS) and reverse bias (RS), of minimodules fabricated with three different DMF concentrations. Bars indicate interquartile ranges (IQRs) (25% to 75%) with median value line, and whiskers show maximum and minimum values up to 1.5 times the IQR. (C) Certified J-V characteristic curves with inserted parameters of the best perovskite minimodule fabricated using the 94.2 wt % recovered DMF solvent. (D) Certified maximum power point tracking data of the best perovskite minimodule fabricated using the 94.2 wt % recovered DMF solvent. (E) UV aging test results of the encapsulated perovskite minimodules under continuous AM 1.5G solar irradiation for more than 800 hours. Each error bar represents mean value ± SD of 12 measurement results.

The performance of the as-prepared minimodules was then assessed under simulated one-sun illumination. The average power conversion efficiency (PCE) of perovskite minimodules fabricated using the recovered DMF (86.3 wt %) was 19.17%, increasing to 19.56% when DMF was further enriched to 94.2 wt % (Fig. 6B). In addition, both the average and highest PCEs achieved from the recovered DMF (94.2 wt %) came close to those of minimodules using the commercial DMF (99.5 wt %) (Fig. 6B; detailed parameters in tables S5 and S6). Although water has been confirmed to induce degradation in PSCs, our findings indicated that the presence of trace water impurity in DMF during the fabrication process did not pose a direct adverse impact on initial efficiencies, although slightly inferior hysteresis performance was observed in minimodules fabricated from the recovered DMF. This counterintuitive observation also implies the necessity for further investigations into water’s roles in PSCs, where water quantities, forms, and specific stages at which it is introduced must be discriminatively analyzed. The best perovskite minimodules fabricated with the recovered DMF (94.2 wt %) and the commercial DMF (99.5 wt %) were also sent to a third-party test center (National PV Industry Measurement and Testing Center) for efficiency certification and area correction (figs. S13 and S14). The certified performance data of these minimodules (Fig. 6, C and D, and fig. S15) were consistent with our in-house measurements (Fig. 6B), confirming the accuracy of our results. The narrow efficiency deviation among minimodules, PCEs meeting state-of-the-art commercial products, and small hysteresis demonstrated the viability of using the recovered DMF from our MAMD system in industrial fabrication.

Furthermore, we also investigated the light stability of these perovskite minimodules by conducting aging tests under continuous ultraviolet (UV) irradiation at an elevated temperature in line with the IEC61215 standard. To mitigate the fluctuations resulting from the fabrication process, 12 minimodules prepared with each DMF concentration were measured (table S7). The as-prepared minimodules (94.2 wt % DMF recovered from MAMD) and control samples (99.5 wt % commercial DMF) exhibited similar PCE degradation rates (Fig. 6E), each maintaining 0.67 and 0.65 of its initial PCE after 60 kWh UV irradiation at 60°C. This UV dose, equivalent to 857-hour continuous AM 1.5G solar irradiation at 60°C, was accelerated to 288 hours in the aging test, representing a threefold increase in UV intensity compared to that of AM 1.5G solar irradiation.

DISCUSSION

In this study, we have demonstrated the highly efficient DMF enrichment performance of a rationally designed MAMD system, which has a distinct advantage of recovering DMF from dilute waste solutions compared to other solvent recovery technologies. In addition, it is applicable to wastewater across a wide range of concentrations and uses stepwise enrichment to enable the reuse of DMF from dilute waste solutions back into industrial processes. Operating with top surface temperatures below 80°C, our five-unit MAMD system achieved remarkable performance of enriching DMF concentration from 0.3 to 94.2 wt %, corresponding to an overall enrichment factor of 314. Different from most membrane distillation studies on desalination, where salts remain nonvolatile, our MAMD system primarily focuses on the enrichment and recovery of the DMF solvent from wastewater, representing a distinct separation mechanism that operates with binary volatile components. To obtain better enrichment performance, we systematically investigated the impact mechanism of various parameters including operating temperatures, feed solution concentrations, and inlet flow rates. In general, temperatures and DMF concentrations in the evaporation layer determined the relative proportion of DMF and H₂O in the generated vapor and distillate, thus affecting the effectiveness of DMF and H₂O separation in the binary component system.

The MAMD system working at a higher steady-state temperature exhibited better enrichment performance because of a considerable difference in vapor pressure between DMF and H₂O. The difference in vapor pressure increased from 1.06 kPa at 10°C (0.17 kPa for DMF and 1.23 kPa for H₂O) to 46.39 kPa at 85°C (11.41 kPa for DMF and 57.81 kPa for H₂O), facilitating the separation of volatile substances (fig. S16). For a MAMD unit, DMF enrichment factors from 3.66 to 1.09 were obtained, covering wide range of inlet concentrations from 5 to 86.3 wt %. The DMF enrichment factor reached its maximum of 3.66 at an inlet concentration of 5 wt % and decreased with increasing inlet concentration, gradually approaching 1. This trend likely resulted from a higher proportion of DMF in vapor at elevated concentrations according to Henry’s law and Raoult’s law. Moreover, an increase in DMF concentration led to a decrease in the total amount of vapor so that more heat was accumulated at the top of the MAMD system rather than transferred by generated vapor molecules, causing a gradual increase in the top surface temperature.

The inlet flow rate was also a key parameter in the MAMD system as it determined the Q_dis/Q_con. A lower inlet flow rate or higher Q_dis/Q_con typically enhanced DMF separation, resulting in a lower concentrate yield but a higher outlet concentration, despite potentially exacerbating concentration polarization (53, 54). Oppositely, a higher flow rate increased the concentrate yield and mitigated potential negative impacts of concentration polarization (55) but led to a decrease in DMF outlet concentration because of the lower Q_dis/Q_con. Therefore, a trade-off between the concentrate yield and the outlet concentration should be considered in the MAMD system. Besides, the versatility of our MAMD system to other organic solvents such as DMSO and NMP underscores its potential to offer a more sustainable and environmentally friendly approach for high-value solvent management across diverse industries. However, it may require more efforts to optimize membrane materials for specific organic solvents to improve both selectivity and durability. Meanwhile, the wastewater matrix can vary substantially across various application scenarios. The investigations on pretreatment countermeasures and impacts of these variations on the solvent recovery performance of our MAMD system deserve further investigation in the future.

The scaled-up MAMD system established in this work, while not yet at industrial size, represented an important advancement from our lab-scale setup, aiming to demonstrate the system’s DMF enrichment and recovery performance at a larger scale for industrial applications. An ideal outlet concentration could be achieved in the scaled-up MAMD system on the basis of the following considerations during the industrial waste heat utilization experiments: (i) The polyurethane sponges with a low thermal conductivity of 0.020 to 0.030 W m⁻¹ K⁻¹ covered around the side surface of the MAMD system effectively reduced the conductive heat loss to the ambient environment; (ii) the effective utilization area (defined as the practical evaporation area of the hydrophobic membrane, 2101.3 cm²) was much larger than the side surface area (216.2 cm²) in the scaled-up MAMD system, guaranteeing a higher heat utilization efficiency of per unit area and a higher distillate yield than those of the smaller configuration. Taking these advantages of the scaled-up MAMD system, a high Q_dis/Q_con could be controlled while also ensuring an ideal concentrate yield. Consequently, the high Q_dis/Q_con contributed to the high outlet concentration and made it possible to reduce the number of units required to achieve the desired outlet concentration. Overall, the scaled-up MAMD system had a large industrial wastewater treatment capacity and the potential to achieve high outlet concentrations and concentrate yields simultaneously. The full-size MAMD system would be further developed in the future to give insights to operate the MAMD system at the industrial scale.

The developed MAMD system had an improved heat utilization efficiency (defined as the distillate yield per unit area of input heat) by reusing latent heat multiple times compared to a single-stage configuration (56–58). The latent heat of vapor released from the previous stage drove the heat-up and evaporation processes of the subsequent stage with a multistage design. However, there would be minimal enhancement in the DMF outlet concentration with additional numbers of stage. For a five-stage MAMD system operating under a 5.0-W input power, the top surface temperature ranged from 71.1° to 77.6°C and the bottom surface temperature varied from 43.1° to 46.7°C across different inlet concentrations from 5 to 60 wt %. Meanwhile, an analysis of a single-stage system operating at a top surface temperature of 40.4°C revealed inferior DMF enrichment performance, with an enrichment factor of only 1.08 (approaching 1). This experiment of a single-stage system suggested that further addition of stages to our designed five-stage MAMD system may not enhance DMF enrichment performance because of the decreasing vapor pressure difference in DMF and H₂O. Therefore, a high top surface temperature could indicate a better separation effect in this MAMD system.

Wetting phenomena were observed on the surface of the membrane at specific temperatures (65°C for an 80 wt % DMF solution and 55°C for a 90 wt % DMF solution). In general, the contact angles of 80 and 90 wt % DMF solutions on PTFE hydrophobic membranes (0.45 μm) decreased with an increase in the solution temperature (figs. S17 and S18). Moreover, analysis of liquid entry pressure revealed an increasing possibility of membrane wetting from the inlet to the outlet as the temperature and concentration of the feed solution increased (Supplementary Text S2), which decreased liquid surface tension and contact angles on hydrophobic membranes, thereby causing lower liquid entry pressure and easier membrane wetting (tables S8 and S9) (59–61). Different from wetting concerns that can be addressed by pretreatment to remove contaminants like surfactants, wetting is induced by the target organic solvent in our MAMD system. Previous efforts have focused on omniphobic membranes to resist wetting (62). However, low contact angles (<90°) of ethanol or decane on these omniphobic membranes also suggest continued challenges in achieving wetting resistance to DMF solutions (63). While composite membranes incorporating a dense hydrophilic layer have shown promise in mitigating oil fouling (64, 65), this approach may not be effective in resisting wetting as DMF molecules can easily penetrate the hydrophilic surface (51). More recently, a membraneless system has been reported to mitigate the wetting concern (66). However, this method may sacrifice the large evaporation area and pose challenges in controlling feed flow rates to obtain desired outlet concentrations and yields. Therefore, it is essential to further develop innovative antiwetting membrane materials for practical applications. Their durability and effectiveness should be evaluated during the long-term operation, and scalable fabrication methods are required for industrial implementation.

In terms of waste high-value organic solvent reuse, the recovered DMF was applied in the manufacturing of PSCs following a procedure with easy scalability for the large-scale production at an industrial level. With 94.2 wt % DMF concentrates recovered from our MAMD system, an excellent certified stabilized power output of 19.97% and PCE of 19.93% from forward/backward J-V measurements were achieved using the industry-friendly method. Furthermore, stability is a key factor toward successful commercialization of PSCs (67–69). In this work, the fabricated minimodules and control samples showed comparable PCE decreasing rates evaluated in line with the IEC61215 standard, confirming the effectiveness of DMF enrichment and recovery using our MAMD system. Our results also indicated that a small trace of water impurity in DMF for the fabrication of PSCs, likely going through a reversible hydration process (70, 71), did not inherently impair their initial efficiencies or compromise UV stability.

In summary, our MAMD system enables highly efficient DMF solvent recovery from waste solutions and reuse for perovskite minimodule fabrication, which offers a promising approach to enhance the sustainability of solution-based PSC manufacturing. This approach not only addresses the urgent need for reducing the environmental impact of waste solvents but also shows great potential to revolutionize the management of high-value organic solvents, promoting broad sustainability practices across various application scenarios.

MATERIALS AND METHODS

Chemical and materials

The double-sided PTFE hydrophobic membranes supported by polypropylene in the middle (pore sizes of 0.22, 0.45, and 1.0 μm; thickness of 0.18 to 0.26 mm; and porosity of 0.70 to 0.78) were obtained from Lianzhong Filter Technology Co., Ltd. DMF (99.5%) was an analytical reagent and was purchased from Shanghai Hushi Laboratorial Equipment Co., Ltd. The alumina ceramic electric heater with a size of 5 by 5 by 0.2 cm was purchased from Beijing Xinlian century Tech Co., Ltd. All DMF aqueous solutions were prepared using deionized water (resistivity of 18.2 megohms·cm).

Assembly of the MAMD system

The five-stage MAMD system with a size of 7 cm by 7 cm was assembled as shown in fig. S19. The thermal conduction layer was made of a stainless-steel plate with a thermal conductivity of 16.3 W m⁻¹ K⁻¹. We enhanced the MAMD configuration by optimizing the design of the flow channel within each evaporation layer and vapor condensation layer to mitigate dead zones and minimize the short flow phenomenon (fig. S1). The evaporation layer had the same PTFE material as the spacers in the vapor condensation layer, which had a hydrophilic nonwoven fabric inside. The hydrophobic porous layer was composed of double-sided PTFE materials. Using commercial membranes and materials enhances the appeal and feasibility of our MAMD system for large-scale industrial production, making it more attractive and easier to implement. An aluminum heat exchanger with the flow channel was strategically integrated into the system’s configuration, which enables the harnessing of industrial waste heat (Fig. 5, A and B).

Characterizations

The initial concentration of the DMF solution was determined on the basis of the water quality metrics of wastewater samples collected from Microquanta on 11 March 2024 (table S3). The wastewater was generated from the spray towers used for DMF vapor collection with the DMF concentration in the wastewater calculated to be 2455 to 3289 mg/liter on the basis of the COD analysis. The refractive indexes of the collected distillate and DMF concentrate samples were determined by a handheld refractometer (WYA-2WAJ, Lichen Instruments; accuracy of ±0.5%). DMF standard solutions with concentrations of 0, 5, 10, 20, 30, 40, 50, and 60 wt % were prepared for drawing the standard curve (fig. S20). The DMF concentration was then calculated from the refractive index using the standard curve with a slope of 0.0012 and an R² of 0.9997. Similarly, standard curves for DMSO and NMP were plotted and their solution concentrations were calculated accordingly (figs. S21 and S22). Two K-type thermocouples connected to the temperature module were used to monitor the real-time temperature of the top and bottom surfaces of the MAMD system. Sample compositions of the distillate and concentrate were analyzed using UPLC-MS (Waters ACQUITY UPLC H-Class PLUS Core System Ultra) equipped with an SQ Detector 2 single quadrupole mass spectrometer. The contact angles of DMF solutions on the double-sided PTFE hydrophobic membranes (0.45 μm) were measured by the optical contact angle measuring instrument (JC-FAT125, Jiangcheng Precision Instruments).

DMF enrichment and recovery investigation

A lab-scale setup was built to evaluate the DMF enrichment and recovery performance of the MAMD system under different operating conditions. In our laboratory experiments, the DMF solution was transported from the feed solution container (in a 22°C water bath) and controlled using a peristaltic pump with a 1-mm-inner-diameter silicon tube for inlet flow rate regulation. To provide heat in the laboratory tests, a square ceramic electric heater connected to a DC power supply was placed on the top surface of the MAMD system. A square aluminum heat sink with a length of 4 cm and a height of 1.1 cm was placed at the bottom surface of the system to dissipate latent heat released in the last stage. The mass change of the distillate container was recorded using a precision electronic balance (ME104E, Mettler Toledo) to calculate the distillate yield. The specific mass flux of the distillate was defined as the distillate yield per unit heat-conducted area of the MAMD system. The DMF concentrates were collected and measured at a steady state. Before conducting tests under different conditions, the remaining solution in the evaporation layer of the MAMD system was replaced with a fresh DMF feed solution. To assess the long-term operation reliability of the MAMD system, we performed a 60-hour continuous enrichment test with a 20 wt % DMF feed solution under a 5.0-W input power condition, and the concentration of the outlet was determined every 12 hours.

MAMD system driven by waste heat

The performance of the MAMD system driven by industrial waste heat was conducted using a scaled-up configuration. The scaled-up MAMD system with a size of 23.5 cm by 23.5 cm resembled the smaller setup with the only difference being the heat source at the top. To simulate the utilization of waste heat in an industrial setting, a centrifugal pump was used to circulate hot water from a 78°C water bath. An aluminum heat exchanger was positioned on the top of the scaled-up MAMD system, creating the conditions of simulated waste heat utilization under laboratory conditions and practical waste heat utilization under industrial conditions.

Perovskite minimodule fabrication

The perovskite minimodules used inverted architecture with a layer-by-layer structure of glass/indium tin oxide (ITO)/PTAA {poly[bis(4-phenyl) (2,4,6-trimethylphenyl) amine]}/perovskite (FA_0.97MA_0.03PbI₃)/PCBM {[6,6]-phenyl C61-butyric acid methyl ester}/BCP (bathocuproine)/Ag. Glass substrates (6 cm by 6 cm) were provided with a predeposited ITO layer. The film stack was laser scribed by a 1064-nm nanosecond beam (2 W) to form isolated cell units with a pitch width of 6.3 mm (dead zone width of 147.62 μm) (Fig. 6A). PTAA (Sigma-Aldrich) was blade coated at 15 mm/s atop of the ITO layer by applying a chlorobenzene-based solution (2 mg/ml) onto the substrate in the atmosphere environment. The gap between the blade and the top surface of the glass was set at 80 μm. The environment was kept at 23 ± 1°C and under 50% relative humidity (RH). A short annealing step was then carried out at 60°C for 10 min. The perovskite solution (1.4 M) was prepared by dissolving PbI₂, formamidinium iodide, and methylammonium iodide (1.4 M:1.3 M:0.1 M) into DMF (5.0, 18.3, 46.0, 72.8, 86.3, and 94.2 wt %, recovered from our MAMD system). DMSO was then added to the solution with a mole ratio of DMSO:PbI₂ = 0.6:1. The perovskite film was blade coated in the atmosphere with a coating speed of 20 mm/s and a gap distance of 100 μm. The room environment was kept at 23 ± 1°C and under 30% RH. A dry air knife was installed immediately after the coating head and its movement in synchronization with the coating head so that the wet perovskite film was blown dry shortly after wet cast. The speed of air at the air knife exit was kept at no slower than 1.5 m/s and monitored intermittently with a hot-wire anemometer. The quasidry perovskite film was then transferred to a hot plate to be annealed at 130°C for 5 min. The PCBM film and BCP film were also blade coated sequentially afterward at 15 mm/s and annealed at 70°C for 5 min after the full stack was formed. This was also completed at room temperature of 23 ± 1°C and under 30% RH. A 532-nm nanosecond laser beam (2 W) was used for P2 scribing to expose the top ITO layer for later series connection. A 150-nm Ag electrode was vacuum deposited to finalize the module layer stack. P3 scribing and P4 edge isolation were carried out with the same 532-nm nanosecond beam (2 W).

PSC module characterization

We fabricated a total of 18 perovskite minimodules using two recovered DMF solvents (86.3 and 94.2 wt %) and a commercial DMF solvent (99.5 wt %), with 6 minimodules prepared from each, to enable the statistic evaluation of module performance distribution. The J-V measurements of perovskite minimodules were performed in the glovebox under a N₂ atmosphere. The J-V characteristic curves of minimodules were recorded on the Keithley source unit 2400 under AM 1.5G one-sun intensity illumination (100 mW/cm²) by an AAA solar simulator (SS-100A) from Beijing Sanyou Technology, and the light intensity was calibrated with a standard photovoltaic reference cell. The best perovskite minimodules prepared from the recovered DMF (94.2 wt %) and the commercial DMF (99.5 wt %) were also sent to a third-party test center (National PV Industry Measurement and Testing Center) for efficiency certification and area correction. “Aperture area” was used to determine the PCE of the minimodule, and a metal mask in the shape of an unfilled square was used to define the aperture. The aperture area measured was 19.22 cm², including both the active zone and dead zone of the minimodule but excluding its edge ablation area, which was covered by the mask during J-V measurement. Microscopic images of the laser cut P1-P2-P3 region and active area are presented with the measured P1-P2-P3 width of 147.62 μm and the pitch width of 6.3 mm in Fig. 6A. The ratio of the active area and dead area was then decided to be 41.86, and the calculated geometry filling factor was 97.67%. The optical band gap was determined to be ~1.53 eV on the basis of the Tauc plot analysis and external quantum efficiency curves (fig. S23).

Before the UV aging test, perovskite minimodules were first encapsulated using the polyolefin elastomer encapsulant, polyisobutylene sealant, and tempered glass. Conductive tape was applied on the electrodes of the minimodules to direct the current out from the encapsulated structure. The protected minimodules were subsequently placed within a UV conditioning chamber (Shanghai Riechy GRO-SUV2213T) to undergo continuous illumination at an elevated temperature under open circuit conditions (fig. S24). This sandwich structure of the perovskite minimodule could increase reflections at the air-glass and glass-encapsulant interfaces, leading to a drop in its initial efficiency. The chamber environment was set in accordance with the IEC61215 standard. The UV chamber temperature was set at 60 ± 5°C. The UV irradiation power density on the surface plane of the minimodules was 600 W/m², among which the strength of UVA (320 nm to 395 nm) was 180 W/m², that of UVB (280 nm to 320 nm) was 20 W/m², and the rest (400 W/m²) were in the range of visible light spectrum. The power density was monitored in real time by an irradiation meter kept at the same level of the surface plane, and the total UV irradiation dose was recorded as the integration of the power density against time. An initial value of PCE of the minimodules was tested and recorded using a Keithley 2400 before it was placed inside the UV chamber. Once the UV irradiation dose reached a milestone (e.g., 30 kWh), the minimodule was taken out for the J-V measurement after it naturally cooled down to ambient temperature.

Simulation information

The numerical simulations were conducted by the developed code using Python. The input conditions of the simulations included those of the corresponding experiments. The simulation model and governing equations can be found in fig. S5 and Supplementary Text S1. Our simulation captures heat, vapor, and concentration transport throughout the system. Given the minimal thickness of the hydrophobic porous layer, we have disregarded the impact of its thickness on heat and concentration transport but only considered it in the membrane coefficient. In addition, we assumed a constant temperature within each evaporation layer, with variations occurring across the layers.

Supplementary Materials

This PDF file includes:

Supplementary Text S1 and S2

Figs. S1 to S24

Tables S1 to S9

References