A more efficient, cost-effective way to harness renewable energy
A more efficient, cost-effective way to harness renewable energy lead image
Photovoltaic cells harness the renewable energy of sunlight for conversion into electricity. Because the efficiency improvement of standard single-junction solar cells is stagnating, two-junction solar cells are gaining popularity. Dual solar cells can absorb a wider spectrum of sunlight leading to higher efficiency.
Two-junction solar cells can be configured as two-terminal (2T) or four-terminal (4T) tandem devices. While the 2T set-up may be cheaper because of fewer components, the 4T device performs better at varying temperatures and irradiance levels.
Xue et al. devised a novel machine learning-based method, called the neural network, to optimize perovskite-based 4T tandem solar cells utilizing a copper indium selenide bottom cell.
“This approach significantly reduces the computational burden of exploring design possibilities for the perovskite top subcell while showing promising enhancements in predicted device performance,” said Hansong Xue.
The neural network delivered the optimal design space in about 11 hours compared to the standard model simulation, which required six months. The model considered trade-offs between subcell efficiencies and material costs in the 4T configuration while accommodating variability in optical and recombination properties. A potential efficiency improvement of 29.4% to 30.4% and a 50% reduction in fabrication material costs were shown when compared to the best-performing fabricated tandem device.
“The methodology outlined in our work presents an efficient and effective means of optimizing perovskite-based 4T tandem devices, potentially hastening the development of high-performance solar cells,” said Xue.
Source: “Exploring the optimal design space of transparent perovskite solar cells for four-terminal tandem applications through Pareto front optimization,” by Hu Quee Tan, Xinhai Zhao, Akhil Ambardekar, Erik Birgersson, and Hansong Xue, APL Machine Learning (2024). This article can be accessed at https://doi.org/10.1063/5.0187208 .
