Expert-level Leaf Cell Layout Generation via Preference-Optimized LLM

In the field of integrated circuits, leaf cells are the basic units, serving as the fundamental building blocks (e.g., standard cells) that are widely reused in various VLSI designs, forming the basis for more complex circuits. Therefore, the design quality of leaf cell layouts significantly impacts the PPA (Power, Performance, and Area) of the final VLSI designs. To automatically design leaf cell layouts that are close to expert designs, we propose GenLeaf. GenLeaf first utilizes a supervised, performance-aware embedding model to represent layouts and automatically calculate their similarity scores. Since there are expert-designed layouts but no corresponding scripts, we implement Bayesian optimization to generate a layout-script dataset for LLM training. With subsequent supervised fine-tuning and further preference optimization, GenLeaf can generate leaf cell layouts through scripts whose performance closely resembles that designed by human engineers. Experiment results demonstrate that GenLeaf outperforms expert-designed golden layouts across key performance metrics.

(1) Designing the tiny building blocks of chips, called leaf cells, is a challenging task that heavily influences a chip's power use, speed, and size. Even small improvements in these basic components can lead to major gains in the final product, but creating optimal layouts traditionally requires years of human expertise. (2) We developed GenLeaf, an AI system that learns to write scripts to design leaf cells. First, GenLeaf learns to recognize what makes a good layout by studying expert examples. Then, using smart search techniques, it creates its own training data to teach a large language model how to generate high-quality design scripts. Through iterative refinement, the system produces layouts that match, or even exceed, human expert quality. (3) By automating this critical step in chip design, GenLeaf can accelerate the development of more efficient electronics, from smartphones to data centers. It also opens the door for AI to assist in other complex engineering tasks where expert knowledge is scarce but high performance is essential.