Master's thesis

Verilog RTL Generation with AI Agents and Formal Verification

Hochschule München University of Applied SciencesAEMY

Verilog RTL Generation with AI Agents and Formal Verification

Hochschule München is among the largest German universities of applied sciences, located at the heart of Munich, Germany. The working group "AEMY" focuses on safe, secure and smart systems. Our main working areas include RISC-V processor design, WebAssembly runtime development and open source chip design tools.

Large Language Models can generate hardware descriptions in Verilog from natural-language specifications, but the generated code often contains subtle functional errors. Most existing approaches check this code only through simulation, which can miss corner cases.

In this thesis, you will build an agent pipeline in which AI agents generate Verilog designs that must satisfy given SystemVerilog Assertions (SVA). Open-source formal tools such as Yosys and SymbiYosys either prove these properties or return counterexamples, which the agents then use to repair the design. The approach applies the generate–verify–repair loop of our tool ACSLGen, which uses AI agents for C++ specifications, to hardware design. The goal is Verilog that is formally verified rather than only tested. Potential areas of focus include:

  • Designing agents for RTL generation, debugging and repair
  • Comparing feedback types: compiler errors, simulation results, bounded model checking and k-induction
  • Condensing counterexample traces into feedback the agents can use
  • Evaluating on existing benchmarks (e.g., VerilogEval, RTLLM) extended with formal properties
  • Comparing local open-weight LLMs with respect to success rate, iterations and runtime

This project offers hands-on experience in agentic AI, hardware formal verification and open-source EDA tools. For further information or to express your interest, feel free to reach out.