Using Artificial Intelligence to Create Better Perovskite Solar Cells

Perovskite solar cells (PSCs) have become an increasingly popular alternative to traditional photovoltaic technologies in the solar industry. These cells, characterized by perovskite-structured materials, offer numerous advantages, including high power conversion efficiencies and cost-effective production. 

 

An example of perovskite thin film. Image used courtesy of Brookhaven National Laboratory

 

However, device stability and scalable manufacturing challenges have hindered their commercialization. To overcome these challenges, researchers have recently turned to machine learning (ML) and explainable artificial intelligence (XAI) to unravel the complexities of perovskite thin-film formation. 

 

Perovskite Manufacturing Challenges

Manufacturing perovskite cells is fraught with challenges and inconsistencies.

One major challenge of producing PSCs involves depositing thin films of perovskite material, which requires precision and uniformity to ensure high performance. However, the current methods for creating these films are unpredictable, often resulting in thickness, composition, and crystalline structure variations across different production batches. These inconsistencies can significantly affect the efficiency and longevity of the solar cells, making it challenging to guarantee a standard level of quality in commercial products.

 

Perovskite cells. Image used courtesy of the Karlsruhe Institute of Technology

 

Moreover, the stability of PSCs remains a concern. While they show high efficiencies in laboratory settings, their performance degrades when exposed to real-world environmental conditions over extended periods. This degradation is often due to factors like moisture, temperature fluctuations, and ultraviolet radiation, which can all change the chemical and physical properties of materials using perovskite.

Ultimately, the perovskite thin-film formation process is poorly understood, complicating efforts to improve it. Traditional approaches to developing these films have relied on trial-and-error methods, leading to gradual and incremental improvements. However, this lack of a systematic understanding impedes the development of standardized and reliable manufacturing processes.
 

Using AI to Reveal Patterns in Perovskite Solar Cells

To tackle these issues, the researchers employed a novel methodology combining deep learning and multiple explainable artificial intelligence methods to analyze high-dimensional data acquired during the PSC manufacturing process.

The research focused on capturing and analyzing in-situ photoluminescence (PL) videos while forming perovskite thin films. These videos revealed complex patterns in the data, which were previously inaccessible to conventional analysis methods, and, using XAI, the team translated these complex data patterns into human-understandable concepts.

 

An overview of the experimental setup. Image used courtesy of Klein et al.

 

The first conclusion from the study was that the researchers established the importance of temporal progression in PL data. They found that this time-based data contained more significant information than spatial data alone, highlighting the necessity of considering dynamic changes during the thin-film formation process.

Another significant conclusion was the correlation between the quality of perovskite films and the PL intensity at the nucleation onset. The study demonstrated that a higher PL peak intensity at this stage led to superior-quality films. This insight is pivotal in optimizing the perovskite film formation process for better overall solar cell performance. Additionally, the research revealed that the timing and intensity of PL peaks during the manufacturing process were crucial in determining the final quality of the PSCs. Specifically, high PL peak intensity at the start of chamber venting induced thicker and rougher films.

Finally, the study concluded that superior crystal growth, characterized by a steeper PL intensity decay during the crystallization phase, correlated with higher-performing PSCs. This finding suggests that how crystallites grow and merge into larger crystals, thus reducing grain boundaries, is key to enhancing charge carrier extraction and, consequently, the efficiency of PSCs.

 

Insights Into Perovskite Films

Combining ML with XAI marks a significant departure from traditional trial-and-error techniques, instead providing a data-driven understanding of the thin-film formation process. This approach proved pivotal in discovering intricate patterns in complex data sets beyond the capability of traditional analyses. XAI helped make these patterns understandable, translating them into actionable conclusions for material scientists.