ChipCraftBrain: Validation-First RTL Generation via Multi-Agent Orchestration

TL;DR

ChipCraftBrain is a multi-agent framework that combines symbolic-neural reasoning and adaptive orchestration to improve RTL code generation from natural language, achieving high correctness on benchmarks and FPGA validation.

Contribution

It introduces a novel adaptive multi-agent orchestration with hybrid reasoning, knowledge augmentation, and hierarchical decomposition for more accurate RTL generation.

Findings

Achieves 97.2% pass@1 on VerilogEval-Human, matching state-of-the-art.

Reaches 94.7% pass@1 on CVDP, outperforming single-shot baselines.

Successfully generates FPGA-validated RISC-V SoC modules with full correctness.

Abstract

Large Language Models (LLMs) show promise for generating Register-Transfer Level (RTL) code from natural language specifications, but single-shot generation achieves only 60-65% functional correctness on standard benchmarks. Multi-agent approaches such as MAGE reach 95.9% on VerilogEval yet remain untested on harder industrial benchmarks such as NVIDIA's CVDP, lack synthesis awareness, and incur high API costs. We present ChipCraftBrain, a framework combining symbolic-neural reasoning with adaptive multi-agent orchestration for automated RTL generation. Four innovations drive the system: (1) adaptive orchestration over six specialized agents via a PPO policy over a 168-dim state (an alternative world-model MPC planner is also evaluated); (2) a hybrid symbolic-neural architecture that solves K-map and truth-table problems algorithmically while specialized agents handle waveform timing and general RTL; (3) knowledge-augmented generation from a 321-pattern base plus 971 open-source reference implementations with focus-aware retrieval; and (4) hierarchical specification decomposition into dependency-ordered sub-modules with interface synchronization. On VerilogEval-Human, ChipCraftBrain achieves 97.2% mean pass@1 (range 96.15-98.72% across 7 runs, best 154/156), on par with ChipAgents (97.4%, self-reported) and ahead of MAGE (95.9%). On a 302-problem non-agentic subset of CVDP spanning five task categories, we reach 94.7% mean pass@1 (286/302, averaged over 3 runs), a 36-60 percentage-point lift per category over the published single-shot baseline; we additionally lead three of four categories shared with NVIDIA's ACE-RTL despite using roughly 30x fewer per-problem attempts. A RISC-V SoC case study demonstrates hierarchical decomposition generating 8/8 lint-passing modules (689 LOC) validated on FPGA, where monolithic generation fails entirely.