Security of Generative AI and Generative AI for Security

Ramesh Karri  - Tandon School of Engineering, New York University 

"JV" Rajendran - Texas A&M University

The rising complexity and tighter production timelines for integrated circuits impose constant challenges. This places immense stress on chip designers and design processes, leading to the emergence of flawed designs with potential vulnerabilities. In a world where computer chips form the backbone of modern life, powering everything from cell phones and cars to traffic lights and pacemakers, even minor errors can have substantial consequences. The unfortunate convergence of high demand and escalating complexity has led to a shortage of qualified engineers. Reports suggest that there are currently 67,000 unfilled jobs in the field, highlighting the pressing need for skilled professionals in this domain.

Over the decades, the field of Electronic Design Automation (EDA) has harnessed the ever-expanding capabilities of machine learning and artificial intelligence across various stages of the chip design process. Tasks ranging from layouts to power and performance analysis, as well as physical design, have all witnessed improvements through programs that are taught rather than explicitly programmed.

This lecture will delve into the upcoming advancements in EDA applications, specifically focusing on the latest form of artificial intelligence—Generative Pre-trained Transformers (GPTs), also known as Large Language Models. We'll illustrate how models like the widely-used ChatGPT can be utilized for tasks such as HDL writing, bug searching and fixing, and even tackling intricate debugging challenges like generating assertions. Instead of confining ourselves to commercial and closed-source tools, we'll demonstrate how you can train your own language models and create designs in a fully open-source manner. We'll also explore the strides made by commercial operators in this domain, such as GitHub Copilot and Cadence JedAI, and reflect on the potential impacts on education and industry—will our designs become more prone to bugs? Will graduating VLSI students be less knowledgeable? All of these aspects will be covered using a diverse set of examples, ranging from simple ones like basic shift registers to complex components such as AXI bus elements and microprocessor designs.