Agent Factories for High Level Synthesis: How Far Can General-Purpose Coding Agents Go in Hardware Optimization?

1. 1Decompose a design into sub-kernels using an agent factory pipeline
2. 2Independently optimize each sub-kernel using pragma and code-level transformations
3. 3Coordinate multiple autonomous optimization agents to achieve overall design optimization
4. 4Evaluate the optimized design for performance and efficiency gains

This research benefits hardware engineers and AI researchers on a team, as it explores the potential of autonomous agents in optimizing hardware designs, which can lead to improved performance and efficiency

💡 General-purpose coding agents can achieve significant optimization gains in hardware design without requiring specialized training

💡 Coding agents can optimize hardware designs without hardware-specific training!