Saturday 01 February 2025
A team of researchers has made significant strides in developing an artificial intelligence system that can optimize high-level synthesis (HLS) designs for chip production. The system, called Agentic-HLS, uses large language models to predict the performance and resource utilization of HLS designs.
HLS is a crucial step in the development of integrated circuits, as it transforms high-level programming languages into low-level hardware descriptions. However, this process can be time-consuming and labor-intensive, requiring significant expertise and computational resources.
The Agentic-HLS system tackles these challenges by integrating machine learning models with specialized tools for reasoning and optimization. The system uses a combination of graph neural networks (GNNs) and large language models to analyze and optimize HLS designs.
One key innovation of the Agentic-HLS system is its ability to use agentic reasoning, which allows it to iteratively review and refine its predictions based on available training data. This process enables the system to learn from its mistakes and improve its performance over time.
The researchers tested their system on a range of HLS designs and achieved impressive results, outperforming previous approaches by a significant margin. The system’s accuracy in predicting design performance was particularly noteworthy, with an average root mean square error (RMSE) score of just 4.21.
The Agentic-HLS system has the potential to revolutionize the field of chip design, enabling designers to create more efficient and effective integrated circuits. By automating the HLS process, the system could also reduce the time and cost associated with chip development, making it easier for companies to bring new products to market.
The researchers are now planning to extend their work by incorporating synthesizer version considerations into the system, which they believe will enable even more accurate predictions. They also hope to explore the use of larger language models, such as GPT-4o1, to further improve the system’s performance.
Overall, the Agentic-HLS system represents a significant step forward in the development of artificial intelligence for chip design, and its potential applications are vast and exciting.
Cite this article: “Artificial Intelligence System Optimizes Chip Design Process”, The Science Archive, 2025.
Artificial Intelligence, High-Level Synthesis, Chip Design, Language Models, Graph Neural Networks, Agentic Reasoning, Integrated Circuits, Machine Learning, Predictive Modeling, Optimisation
