MIT researchers zero in on more durable perovskite solar cells

One of the most likely materials to replace silicon in the construction of solar cells is a material called perovskites. However, one of the biggest issues with perovskites is the tendency for the material to degrade relatively quickly. The usable lifetime of a perovskite-based solar cell has gradually improved from minutes to months, but that's vastly inferior to the decades silicon solar panels can last.

An international team of researchers led by scientists from MIT has developed a new approach to narrowing the search for the best candidates for long-lasting perovskite formulations. This is a challenge due to the vast number of potential combinations. The system has allowed the researchers to zero in on one composition that in the lab has improved on existing versions of perovskite-based solar cells by a factor of 10.

The new formulation was tested under real-world conditions at full solar cell level, not just a lab sample. The perovskite formulation performed three times better than current state-of-the-art formulations. Perovskites are a broad class of materials characterized by the way atoms are arranged in their layered crystal lattice. Layers inside the material are described by convention as A, B, and X, each consisting of various atoms or compounds.

Researchers need to search through a massive number of combinations to find the most likely combination providing longevity, efficiency, manufacturability, and availability for the source materials. One scientist on the project says that you have to consider even just three elements, the most common elements in perovskites that people sub in or out are on the A side of the perovskite crystal structure. The elements can be varied by one percent increments in the relative composition, and the number of steps becomes preposterous.

The team uses a data fusion approach. It's an iterative method using an automated system to guide the production and testing of various formulations. The system uses machine learning to go through the results of those tests combined with first-principles physical modeling to guide the next round of experiments. The system repeats the process and refines the results each time.