Reproduce “ARCS: Autoregressive Circuit Synthesis with Topology-Aware Graph Attention and Spec Conditioning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tushar Dhananjay Pathak
ARCS is an autoregressive system that rapidly synthesizes complete, SPICE-simulatable analog circuit designs, including both topology and component values, conditioned on target specifications. It generates structurally valid designs in milliseconds, leveraging topology-aware graph attention and a novel Group Relative Policy Optimization method.
Key insights
- Incorporating circuit-graph inductive biases and decoupling structural and numerical predictions in transformer architectures improves simulation validity for generated designs.
- Normalizing reinforcement learning rewards on a per-topology basis is crucial for effective multi-topology learning, preventing the model from favoring 'easier' circuits.
- Grammar-constrained decoding not only guarantees structural validity but can also accelerate inference by preventing the generation of lengthy, invalid token sequences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.