Data-driven optimization of 3D battery design

To power microelectronics for the internet-of-things applications, high-performance miniature batteries, called microbatteries, are critically important. Given their limited size, the three-dimensional design of microbatteries is key to maximizing their performance. Therefore, a computational strategy to identify the target battery architecture has major implications for performance improvement. In this paper, we propose a data-driven 3D battery optimization system at the full cell level that combines an automatic geometry generator based on Monte Carlo Tree Search and highly accurate machine-learning-based performance simulators. The performance of the proposed method is demonstrated by designing high-performance 3D batteries with more than 5.5 times efficiency compared with the approach based on a randomized algorithm. One of the designed geometries displayed greater power and energy densities due to more than 10% reduced internal resistance than the reported state-of-the-art geometry at the current density of higher than 15.8 mA/cm². The results demonstrate the effectiveness of the method.