Artificial Intelligence Based Bandgap Voltage Reference Design
Savvas Karipidis, Andi Buzo, Georg Pelz, T. Noulis · 2025
In this work artificial intelligence algorithms are selected and applied to the problem of minimizing the temperature coefficient of a bandgap reference voltage. The optimization problem is being split into two parts to achieve higher efficiency. In the first part, the operating point of the transistors is being sought with the dual annealing algorithm, whose integrity on working on ill-defined integer surface vindicates its selection. This algorithm's output is next algorithm's starting point, from which a better-conditioned surface is encountered. Nelder-Mead method is used to descent to the minimum. Entire process takes a few hundreds function calls, making it fast and efficient, achieving a temperature coefficient of 30 ppm.