AI-Accelerated Atoms-to-Circuits Thermal Simulation Pipeline for Integrated Circuit Design

Transistor scaling and 3D integration have led to high power densities and operating temperatures, which degrade circuit reliability and performance. Layout complexities and nanoscale features have also made temperature difficult to measure, manage, and predict. Thus, new modeling approaches are needed to simulate temperature accurately and efficiently from nanoscale transistors to systems. Here, we present the first AI-accelerated, multiscale, atoms-to-circuits thermal simulation pipeline, illustrating its predictive capabilities with two examples. First, we describe the atoms-to-transistors approach, based purely on ab initio, atomistic materials modeling, and apply it to predict the temperature distribution of an Intel 16 FinFET. Second, we show the transistors-to-circuits approach. This builds from a FinFET thermal model based on the ASAP7 process design kit (PDK) to a full-detail, nanoscale resolution temperature prediction of an active RISC-V core in < 10 minutes, a result that existing tools cannot match. Ultimately, our atoms-to-circuits thermal simulation pipeline enables engineers across the physical design process to accurately and rapidly evaluate the impact of heat on their designs.