Evaluation of the Capabilities of Generative AIs in Converting HLS Code to RTL

• Grant number: 24/21130-4
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: March 01, 2025
• End date: June 30, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Vanderlei Bonato
• Grantee: Giordano Santorum Lorenzetto
• Supervisor: Oliver Sinnen
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: University of Auckland, New Zealand
• Associated to the scholarship: 24/07574-7 - Evaluation of the Ability of Generative and Public-Use Artificial Intelligence Models for RTL Code Generation, BP.IC

Abstract

This project investigates the use of generative artificial intelligence models, such as ChatGPT (OpenAI) and Gemini (Google), for converting High Level Synthesis (HLS) code to Register Transfer Level (RTL). The objective is to compare the quality and functional correctness of RTL code generated by AI from HLS code, such as C++ used by Vitis HLS, with those produced from natural language prompts. For this, the open-source RTLLM repository, which provides circuits and testbenches for verifying the functionality of the generated code, will be used. Additionally, a comparison will be made between the performance of AIs and traditional HLS tools, such as AMD Vitis HLS. The evaluated metrics include resource usage, throughput, latency, and the pass@k metric to measure the functionality of the generated code. This study aims to contribute to the understanding of the potential of generative AIs in hardware development, offering a detailed analysis of their applicability and efficiency compared to traditional methods.