Efficient yield optimization with limited gradient information

TL;DR

This paper introduces a modified Newton-Monte Carlo method for yield optimization that effectively handles both uncertain variables with gradients and deterministic variables without, improving efficiency over derivative-free methods.

Contribution

It presents a novel mixed strategy combining gradient-based and derivative-free approaches for yield optimization with mixed variable types.

Findings

The modified method outperforms derivative-free approaches in efficiency.

Numerical comparisons demonstrate the effectiveness of the mixed strategy.

The approach handles uncertain and deterministic variables simultaneously.

Abstract

In this work an efficient strategy for yield optimization with uncertain and deterministic optimization variables is presented. The gradient based adaptive Newton-Monte Carlo method is modified, such that it can handle variables with (uncertain parameters) and without (deterministic parameters) analytical gradient information. This mixed strategy is numerically compared to derivative free approaches.