Multi-Fidelity Surrogate Models for Accelerated Multi-Objective Analog Circuit Design and Optimization
Year: 2025
Type: article
Abstract: Exploring Pareto-optimal trade-offs in analog circuit design is computationally expensive because each candidate evaluation requires time-consuming SPICE simulations. We propose a surrogate-augmented multi-objective optimization framework supporting two operational modes: Surrogate-Guided Optimization (SGO) with periodic truth validation, and Multi-Fidelity Optimization (MFO) with adaptive fidelity promotion. Our pipeline embeds a Graph Neural Network (GNN) surrogate within standard multi-objective evolutionary algorithms (NSGA-II, NSGA-III, SPEA2, MOEA/D) and couples: (i) simu-lator-in-the-loop data generation with caching, deduplication, and selective parallel dispatch of SPICE evaluations via process-based or distributed execution; (ii) uncertain-ty-aware fidelity control with checkpoint/rollback governance to bound predictive drift, tracking Mean Absolute Error (MAE) across validation batches; and (iii) diversity-aware truth selection strategies (greedy farthest-point or crowding distance) that target in-formative regions of the design space. Both operational modes are evaluated across four representative operational amplifier topologies under uniform configurations. The pro-posed approach lowers the number of high fidelity simulations required while main-taining comparable Pareto front quality, indicating a pragmatic path to more sam-ple efficient analog optimization without relying on large scale pretraining or transfer learning.
Source: Electronics
Institutions: Universidad San Pablo CEU, Vrije Universiteit Brussel
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7117312554