Using Symbolic Regression Metamodels for Integrated Circuits Production
Mónika Grezer, Marina Ţopa, Emilian David, Andi Buzo, Georg Pelz · 2024
The paper introduces a Machine Learning QLattice metamodel that can be used as a more efficient substitute for the circuit simulators in the integrated circuit industry with less compute effort. QLattice represents a Symbolic Regression metamodel that is based on Genetic Programming. The purpose of this metamodel is to find the best mathematical formula that achieves the optimal correlation between the inputs and the output. We made a comparative analysis between two regression metamodels, namely QLattice and Gaussian Process, on synthetic functions that emulate real circuits behaviour, as well as on the datasets used for real-life product development.