Agentic AI Meets Formal Verification in the Era of Agentic Hardware Development
Organizations: Hong Kong University of Science and Technology
Abstract
As AI agents become increasingly capable of generating and optimizing hardware circuits, ensuring the correctness of AI-generated designs becomes more important. Meanwhile, advances in AI lower the barriers to developing and applying formal methods, creating new opportunities to automate and improve hardware verification. This invited paper presents an overview of the methodologies and applications of hardware formal methods in the AI era. First, we examine how agentic AI advances hardware formal methods across three levels: specification, proof construction, and reasoning engines. Second, we discuss how formal methods support reliable AI-driven hardware development through RTL generation, debugging, and correctness-preserving optimization. We review representative research alongside our recent contributions, highlighting key technical advances and insights from experimental results. Finally, we discuss future research opportunities toward proof-native hardware development, where formal reasoning could become an integral part of autonomous hardware design.
Figures & tables
| Statistic | Value | Eval. Dimension | Metric |
|---|---|---|---|
| Design Scale | 83 | Syntax Check | Compile Pass |
| Func. Categories | 13 | Formal Provability | FPV Proven Rate |
| Avg. RTL Scale | 2.9k LoC | Coverage | COI/Formal/Proof |
| Avg. Spec Scale | 10k tokens | Mutation Testing | Mutant Kill Rate |
| Category | Direction | Representative Work | Key Contribution |
| Formal Specification | Property- based | AssertLLM ( Yan et al., 2025 ) | Automates SVA generation from specification documents and waveform. |
| AssertLLM2 ( Wu et al., 2026 ) | Evaluates assertion quality via formal verification, coverage, and mutation testing. | ||
| AssertionForge ( Bai et al., 2025b ) | Combines specification and RTL context for assertion generation. | ||
| LASSO ( Ankireddy et al., 2025 ) | Uses formal feedback to improve assertion coverage. | ||
| FVEval ( Kang et al., 2025 ) | Benchmarks assertion generation using syntax and formal verification. | ||
| Reference model-based | FormalRTL ( Li et al., 2026 ) | Uses a supplied C reference model to guide RTL generation and equivalence-based correction. |
| Category | Direction | Representative Work | Key Contribution |
| Generation | Property-guided generation | OpenLLM-RTL ( Liu et al., 2024 ) | Uses formal verification to select training examples for RTL generation. |
| CorrectRTL ( Wang et al., 2026a ) | Uses generated assertions and formal feedback for iterative RTL correction. | ||
| Formal-model- guided generation | FormalRTL ( Li et al., 2026 ) | Uses C reference models for equivalence-guided RTL generation and correction. | |
| CktFormalizer ( Xiong et al., 2026 ) | Uses typed circuit representations and theorem proving for hardware generation. | ||
| HINT ( Cheng et al., 2026 ) | Introduces an executable hardware-intent representation with pre-RTL checking and RTL-lowering obligations. | ||
| Proof2Silicon ( Jha et al., 2025 ) | Uses verifier-guided learning and formally verified intermediate programs for hardware generation. |
Formal Hardware Verification
Formal hardware verification uses mathematical methods to prove that RTL designs satisfy safety, temporal, security, and equivalence properties.
See all 33 papers