Oracle's ORACLE framework slashes analog chip design time 20x
ORACLE cuts analog circuit design time from weeks to hours using RL + LLMs...
Deep Dive
ORACLE is an open-source reinforcement learning framework for multi-objective analog circuit design optimization, using vector-valued rewards and preference-aware conditioning instead of scalar rewards. It also employs large language model-guided action selection to filter suboptimal actions. Across multiple circuit topologies with 2,000 test cases, ORACLE reduces runtime by 20.4x–104.4x compared to state-of-the-art approaches, meets 99.9% of target specifications, and achieves 5.1x–318.6x better figure of merit in the resulting output specs.
Key Points
- ORACLE is an open-source RL framework from University of Michigan and Oracle for analog circuit design optimization
- Achieves 20.4x-104.4x faster runtime on 2,000 circuit topologies while meeting 99.9% of specifications
- Uses vector-based rewards, preference conditioning, and LLM-guided action filtering for superior performance
Why It Matters
Could cut analog chip development cycles from months to days, accelerating AI hardware innovation