David vs. Goliath: Can Small Models Win Big with Agentic AI in Hardware Design?

TL;DR

This paper demonstrates that small language models, when combined with an agentic AI framework involving task decomposition and iterative feedback, can achieve near-LLM performance in hardware design tasks at a fraction of the cost.

Contribution

It introduces a novel approach combining small models with agentic workflows to efficiently tackle complex hardware design problems, challenging the notion that bigger models are always better.

Findings

Small models with agentic AI match large model performance on hardware design tasks.

Agentic workflows reduce computational costs significantly.

The approach enables adaptive learning in complex design environments.

Abstract

Large Language Model(LLM) inference demands massive compute and energy, making domain-specific tasks expensive and unsustainable. As foundation models keep scaling, we ask: Is bigger always better for hardware design? Our work tests this by evaluating Small Language Models coupled with a curated agentic AI framework on NVIDIA's Comprehensive Verilog Design Problems(CVDP) benchmark. Results show that agentic workflows: through task decomposition, iterative feedback, and correction - not only unlock near-LLM performance at a fraction of the cost but also create learning opportunities for agents, paving the way for efficient, adaptive solutions in complex design tasks.