POS10-0230
Multi-fidelity Bayesian Optimization for Cost-efficient Simulation-based Design: A Study on Optimization of Block Copolymer Directed Self-Assembly
When and Where
Nov 30, -0001
12:00am - 12:00am
Presenter(s)
Seungha Jeong (Kyung Hee University)
Co-Author(s)
Abstract
Directed self-assembly (DSA) of block copolymers involves a high-dimensional design space with coupled parameters, making it time-consuming and costly to derive optimal conditions through trial-and-error exploration. We present a multi-fidelity Gaussian process active learning framework to balance the cost and accuracy of an experimental design and apply it to dissipative particle dynamics simulations of DSA processes. We dynamically determine the subsequent evaluation point and its corresponding fidelity level by sampling statistically uncertain regions within the design space. Using line edge roughness and pattern placement error as objective functions, we propose designs that minimize the target metrics and demonstrate reduced cumulative computational cost while maintaining predictive reliability. This study offers broad applicability not only to DSA optimization but also to simulation-based design problems across domains where computational resources are limited.
