IIT Guwahati's LEAP Framework Revolutionizes IC Design with 50% Faster Runtime

Researchers at the Indian Institute of Technology Guwahati have unveiled a groundbreaking machine learning framework named 'LEAP,' marking a significant advancement in Electronic Design Automation (EDA).

This innovative tool is set to transform the Integrated Circuit (IC) design process, a crucial aspect of the $600 billion semiconductor industry that underpins modern electronic devices.

IC design relies on EDA software to convert high-level designs into a manufacturing format called Graphic Design System (GDS). Traditionally, this process is fraught with complexity and relies on heuristic methods that, while quick, often result in less-than-optimal solutions.

The team behind LEAP, including Prof. Chandan Karfa and Dr. Sukanta Bhattacharjee from IIT Guwahati, along with their BTech students Chandrabhushan Reddy Chigarapally and Harshwardhan Nitin Bhakkad, and collaborator Dr. Animesh Basak Chowdhury from New York University, has developed a solution that significantly enhances design efficiency.

LEAP revolutionizes the technology mapping process by intelligently prioritizing promising configurations, reducing the number of options considered by over 50 percent. This targeted approach speeds up the design process and improves circuit performance.

Prof. Karfa said, "Our framework not only accelerates the mapping process but also boosts circuit performance. We've cut the EDA tool's runtime by 50 percent and achieved a 2 percent reduction in clock period without increasing circuit area."

The LEAP framework narrows down configurations by estimating delays and selecting the top ten options for each node, compared to traditional methods that evaluate around 250 configurations. This improvement results in a 50 percent reduction in runtime and a 51 percent decrease in the number of configurations checked. LEAP achieves comparable performance to exhaustive methods while using 63 percent fewer configurations, substantially enhancing the efficiency of the open-source ABC EDA tool.

Published at the ACM/IEEE International Conference on Computer-Aided Design (ICCAD 2024), this research promises to impact the semiconductor industry by reducing design times and enhancing performance. The LEAP framework paves the way for faster, more efficient electronic devices with lower energy consumption, driving innovation across technology sectors.