Alpha-RTL: Test-Time Training for RTL Hardware Optimization

In three linesAlpha-RTL introduces TTT-RTL, a test-time reinforcement learning framework for LLM-based RTL generation optimization. On RTLLM v2.0 (Nangate 45nm), TTT-RTL reduces PPA product by 65.1% versus reference and outperforms frozen-policy baselines by 26.1%. On XuanTie C910 FPU (Sky130), achieves 59.4% ADP reduction. Adaptive KL-budget controller stabilizes policy updates.

Summary generated by Claude — human-verified