Generative AI-enabled microstructure design of porous thermal interface materials with desired effective thermal conductivity

The conventional approach to achieve desired effective thermal conductivity (ETC) of porous thermal interface materials (TIM) is processing-microstructure-properties forward analysis, which contains various trial-and-error cycles and is hence inefficient for materials development. Establishing the linkage from ETC to microstructure is essential; however, the recently developed methods including microstructure characterization and reconstruction are suffering from limited accuracy and computational efficiency. To address these problem, in this paper, generative artificial intelligence (AI) model was first implemented to design microstructure of porous TIM with desired ETC. Here, we introduced a representative porous TIM, sintered silver, and a typical kind of generative AI model, conditional generative adversarial network (CGAN), as an example for illustration. The CGAN model can efficiently generate sharp and crisp microstructures of sintered Ag with excellent morphology realism. Besides visual inspection, the ETC values of generated microstructures were evaluated by convolution neural network (CNN) model. It was found that the CGAN model also exhibits satisfactory performance in physical meaning, since the determination coefficient R² between target ETC and CNN predicted ETC values is 0.985. These results confirm the effectiveness of generative AI model capable of synthesizing microstructure of porous TIM with desired ETC, and not limited to porous TIM, the approaches present here can also be generalized and applicable to design microstructure of other porous media and composites.