Design and Optimization of T-Coil-Enhanced ESD Circuit with Upsampling Convolutional Neural Network

T-coils are widely used in high-speed electrostatic discharge (ESD) circuits to increase bandwidth. Like many other RF/microwave devices, T-coil modeling relies on time-consuming electromagnetic (EM) simulations, which precludes quick design space exploration and fast global optimization. In this paper, a machine learning (ML) model is presented to replace EM T-coil simulations, thereby accelerating T-coil design and optimization. Given the geometry of a T-coil layout, the ML model can infer its S-parameters from 100 MHz to 100 GHz nearly instantly. Finally, this ML model is incorporated into a genetic algorithm (GA), affording a 10× speed improvement in the optimization of a T-coil-enhanced ESD circuit in a 22nm FD-SOI CMOS process.