AI Tool MacroAgent Arranges Chip Parts for Faster, Cheaper Chips
Your next smartphone could be faster and last longer on a charge.
MacroAgent is a new framework for macro legalization in chip design, using large language models to create regularity-aware contour algorithms through a four-stage process: clustering, contour generation, template matching, and inter-cluster refinement. On TILOS and Chipyard benchmarks, it delivered 2 to 8 times better layout regularity, 3% to 5% lower routed wirelength with comparable congestion after global routing, and stronger robustness than existing methods. End-to-end tests with Cadence Innovus also showed a 2.9% reduction in routed wirelength and a 68.3% improvement in TNS over the DREAMPlace baseline, plus 1.8% lower routed wirelength when integrated into the Innovus flow.
- MacroAgent uses AI language models to invent new ways of arranging components on computer chips.
- Tests show up to 68% better timing performance and 3-5% shorter wiring compared to leading methods.
- For consumers, this could mean faster devices with longer battery life and lower manufacturing costs.
Why It Matters
Faster, cheaper, more efficient chips mean better phones, laptops, and even cars at lower prices.