Evolution of Optimization Algorithms for Global Placement via Large Language Models

TL;DR

This work tackles the NP-hard global placement problem by introducing an automated framework that uses large language models to evolve optimization algorithms for initialization, preconditioning, and optimization in placement engines, aiming to beat handcrafted heuristics. It combines offline candidate generation with an LLM-driven genetic-evolution loop and an algorithm-level design-space exploration module based on Bayesian optimization, Gaussian process surrogates, and EI acquisition. Across MMS, ISPD2005, and ISPD2019 benchmarks, the evolved algorithms achieve substantial improvements in $HPWL$, with several cases surpassing 10–17% gains and strong generalization to unseen scenarios. The approach is compatible with existing parameter-tuning methods (e.g., AutoDMP) and demonstrates the potential of automatic algorithm design to scale to industry-scale EDA challenges, while offering a practical framework for resource-aware design-space exploration.

Abstract

Optimization algorithms are widely employed to tackle complex problems, but designing them manually is often labor-intensive and requires significant expertise. Global placement is a fundamental step in electronic design automation (EDA). While analytical approaches represent the state-of-the-art (SOTA) in global placement, their core optimization algorithms remain heavily dependent on heuristics and customized components, such as initialization strategies, preconditioning methods, and line search techniques. This paper presents an automated framework that leverages large language models (LLM) to evolve optimization algorithms for global placement. We first generate diverse candidate algorithms using LLM through carefully crafted prompts. Then we introduce an LLM-based genetic flow to evolve selected candidate algorithms. The discovered optimization algorithms exhibit substantial performance improvements across many benchmarks. Specifically, Our design-case-specific discovered algorithms achieve average HPWL improvements of \textbf{5.05\%}, \text{5.29\%} and \textbf{8.30\%} on MMS, ISPD2005 and ISPD2019 benchmarks, and up to \textbf{17\%} improvements on individual cases. Additionally, the discovered algorithms demonstrate good generalization ability and are complementary to existing parameter-tuning methods.

Figures (10)

• Figure 1: Evolution of Optimization Algorithms
• Figure 2: Discovered algorithms example.
• Figure 3: Comparison between DREAMPlace-4.1 and ours.
• Figure 4: LLM-based Algorithm Generation and Evolution Pipeline.
• Figure 5: A Simplified Prompt Example.