{"id":110,"date":"2022-09-05T19:53:01","date_gmt":"2022-09-06T02:53:01","guid":{"rendered":"https:\/\/faculty.engineering.asu.edu\/vidyachhabria\/?page_id=110"},"modified":"2025-01-19T16:25:31","modified_gmt":"2025-01-19T23:25:31","slug":"research","status":"publish","type":"page","link":"https:\/\/faculty.engineering.asu.edu\/vidyachhabria\/research\/","title":{"rendered":"Research"},"content":{"rendered":"\n

Our research focuses on developing novel electronic design automation (EDA) tools and computer-aided design techniques for VLSI systems. Our current interests include the following:<\/strong><\/p>\n\n\n\n

\n

1) Machine learning for EDA and open-source EDA<\/h3>\n<\/div>\n\n\n\n

a) ML EDA Research<\/h4>\n\n\n\n

The semiconductor industry faces several challenges with traditional Electronic Design Automation (EDA) tools, including limited scalability, high complexity, and reliance on heuristic-based approaches that often sacrifice accuracy for speed. At ASU-VDA Lab, we\u2019re addressing these issues by leveraging cutting-edge Artificial Intelligence (AI) and Machine Learning (ML) techniques.<\/p>\n\n\n\n

Challenges with Traditional EDA Tools<\/strong><\/h5>\n\n\n\n
    \n
  1. Heuristic-Based Approaches<\/strong>: These methods tend to prioritize speed over accuracy, leading to suboptimal design solutions.<\/li>\n\n\n\n
  2. High Barrier to Entry<\/strong>: Traditional tools can be difficult for non-experts to use effectively, limiting their accessibility.<\/li>\n\n\n\n
  3. Limited Scalability<\/strong>: With polynomial time complexity, traditional tools struggle to scale with new technologies and large systems, such as those containing trillions of transistors.<\/li>\n<\/ol>\n\n\n\n
    AI-Driven Solutions for Automation and Optimization<\/h5>\n\n\n\n

    To overcome these barriers, ASU-VDA Lab is exploring innovative AI-driven approaches to enhance both automation and optimization in EDA processes.<\/p>\n\n\n\n

    \n

    Leveraging Large Language Models (LLMs) for Automation<\/strong><\/p>\n\n\n\n

    To lower entry barriers and streamline the automation of EDA tasks, we are developing chatbots and specialized datasets for ML-driven EDA script generation. A key example is the OpenROAD Assistant<\/strong>, which fine-tunes the LLaMA2 model to answer questions related to the open-source EDA tool, OpenROAD, and generate scripts to perform physical design tasks, as shown in the figure on the right. To further enhance these models, ASU-VDA is building specialized datasets, including the EDA Corpus<\/strong> for LLMs, BeGAN<\/strong> for layout related data generation and synthetic netlists.<\/p>\n<\/div>

    \"\"<\/figure><\/div>\n\n\n\n
    \n

    AI-Powered Fast Analysis for EDA<\/strong><\/p>\n\n\n\n

    In addition to improving automation, fast and accurate analysis remains a critical challenge in EDA. As part of our research, we\u2019ve developed AI models for inrush current analysis<\/strong>, IR drop analysis<\/strong>, and thermal behavior analysis<\/strong>. These models utilize image-based AI techniques to enable rapid performance insights, significantly improving design efficiency and reducing time-to-market.<\/p>\n<\/div>

    \"\"<\/figure><\/div>\n\n\n\n
    \n

    Graph Neural Networks (GNN) for Netlist Optimization<\/strong><\/p>\n\n\n\n

    Another exciting area of our research focuses on using Graph Neural Networks (GNNs)<\/strong> for optimizing netlists, particularly for tasks such as gate sizing<\/strong> and buffer insertion<\/strong>. GNNs offer the potential to model and optimize complex circuit structures more efficiently, improving performance and reducing the time needed to design optimized systems.<\/p>\n<\/div>

    \"\"<\/figure><\/div>\n\n\n\n

    b) ML EDA Infrastructure<\/h4>\n\n\n\n

    ML for EDA is a growing research area, but it lacks infrastructure. ASU-VDA has developed infrastructure using OpenROAD to provide a playground for EDA researchers. This infrastructure has three key pieces:<\/p>\n\n\n\n

    \n
    <\/div>\n<\/div>\n\n\n\n
    \n