Computer-aided hardware design and test Research goal:
Designing high-performance analog and mixed-signal integrated circuits is a complex and time-consuming process that requires significant expertise. Traditional design methods rely heavily on human intuition and iterative optimization, making them inefficient for modern, increasingly complex circuits. Our research aims to automate the topology synthesis and design optimization of analog and mixed-signal circuits using artificial intelligence, particularly machine learning and reinforcement learning. By developing AI-driven design methodologies, we seek to enhance efficiency, reduce power consumption, and accelerate the innovation of next-generation integrated circuits.
Research approach:
Our approach utilizes a hierarchical AI framework to automate circuit design. A high-level AI agent explores different circuit topologies, determining how key building blocks should be connected to form an optimized architecture. A lower-level AI agent then fine-tunes design parameters to meet performance constraints while minimizing power consumption. To further improve efficiency, we incorporate transfer learning techniques, allowing AI models to adapt knowledge from simpler designs to more complex circuits. Our research demonstrates that machine-learning-based approaches can significantly improve design automation, offering a scalable and efficient alternative to traditional circuit design methodologies.
