Under the backdrop of the dual carbon strategy, the development of high-efficiency and low-cost photovoltaic technologies has become a core focus for achieving carbon peaking and carbon neutrality goals. Among various photovoltaic technologies, perovskite solar cells (PSCs) have rapidly emerged as a research hotspot in the photovoltaic field owing to their prominent advantages, including facile and low-cost fabrication processes (e.g., solution spin-coating and inkjet printing), high power conversion efficiency (PCE) that has approached or even surpassed that of traditional silicon-based solar cells, excellent stackable integration performance for tandem device fabrication, and favorable compatibility with flexible substrates. These merits render PSCs a promising candidate for next-generation photovoltaic technologies, with substantial potential to promote the diversification and low-carbon transformation of the energy structure. Despite these prominent advantages, the large-scale commercial application of PSCs is still hampered by several key technical bottlenecks. Firstly, the intrinsic material stability of perovskites is inadequate, as they are susceptible to decomposition under environmental stimuli such as moisture, oxygen, illumination, and heat, severely impairing the long-term service lifetime of devices. Secondly, the controllability of the fabrication process remains limited and subtle variations in process parameters (e.g., annealing temperature, solvent ratio, and film-forming atmosphere) can induce significant discrepancies in perovskite film quality and device performance, thereby hindering uniform mass production. Additionally, the lack of reliable packaging technologies further restricts the practical deployment of PSCs in outdoor environments, collectively impeding their further applications. To accelerate the resolution of these aforementioned bottlenecks, in recent years, with the rapid advancement of artificial intelligence (AI), machine learning (ML) technology has been progressively applied to the research of high-efficiency and high-stability PSC devices, benefiting from its robust capabilities in data mining, pattern recognition, and predictive modeling. Specifically, ML can efficiently tackle the complex, high-dimensional data generated in PSC research, thereby overcoming the limitations of the traditional time-consuming and labor-intensive “trial-and-error” research paradigm and providing a new avenue for accelerating PSC technology innovation. Hence, this review systematically surveys machine-learning advances in perovskite material discovery, process optimization and device-performance diagnostics, and outlines future directions in intelligent tandem design and digital encapsulation, offering theoretical and practical guidance for translating laboratory breakthroughs into gigawatt-scale deployment.
Wang et al. (Thu,) studied this question.