PrefixGPT: Prefix Adder Optimization by a Generative Pre-trained Transformer

2025-11-22

AI Summary

Prefix adder design faces challenges stemming from stringent design rules and an exponentially growing search space. This paper introduces, for the first time, Generative Pre-trained Transformers (GPT) into hardware structure generation, proposing an end-to-end sequential topological modeling framework: circuit topology is encoded via coordinate-based representations, structural validity is enforced through legality masking, and design-rule compliance is autonomously learned via a two-stage pretraining–fine-tuning strategy. Crucially, the method generates high-performance prefix adders directly—without relying on handcrafted features or heuristic rules. Experimental results demonstrate that the best generated designs achieve a 7.7% reduction in area–delay product and an average 79.1% improvement in performance over baselines. These findings validate the feasibility and superiority of large language model–driven automated hardware architecture optimization.

Technology Category

• Machine Learning: Hardware-aware ML
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Search and Optimization: Learning to Search

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Prefix adders are widely used in compute-intensive applications for their high speed. However, designing optimized prefix adders is challenging due to strict design rules and an exponentially large design space. We introduce PrefixGPT, a generative pre-trained Transformer (GPT) that directly generates optimized prefix adders from scratch. Our approach represents an adder's topology as a two-dimensional coordinate sequence and applies a legality mask during generation, ensuring every design is valid by construction. PrefixGPT features a customized decoder-only Transformer architecture. The model is first pre-trained on a corpus of randomly synthesized valid prefix adders to learn design rules and then fine-tuned to navigate the design space for optimized design quality. Compared with existing works, PrefixGPT not only finds a new optimal design with a 7.7% improved area-delay product (ADP) but exhibits superior exploration quality, lowering the average ADP by up to 79.1%. This demonstrates the potential of GPT-style models to first master complex hardware design principles and then apply them for more efficient design optimization.

Problem

• Optimizing prefix adder designs in large search space
• Ensuring valid hardware topology through legality constraints
• Improving area-delay product using generative transformer models

Innovation

• Generative pre-trained Transformer generates optimized prefix adders
• Customized decoder-only architecture with legality mask ensures validity
• Pre-training and fine-tuning enable superior area-delay product optimization