ChipMATE: Self-Trained Multi-Agent Framework for RTL Generation
ChipMATE introduces an innovative multi-agent framework for generating RTL code, aiming to resolve discrepancies between current API-based systems and real-world industrial practices. In contrast to earlier methods that depend on the availability of a golden testbench during the generation process and utilize closed-source APIs, ChipMATE functions without a golden oracle and accommodates air-gapped security needs. It integrates a Verilog agent alongside a Python reference-model agent that validate each other's outputs, drawing inspiration from industry-standard cross-comparison techniques. This self-training framework allows for the use of proprietary RTL codebases and enhances single-turn generators by adding verification. The research paper can be found on arXiv with the ID 2605.12857.
Key facts
- ChipMATE is the first self-trained multi-agent framework for RTL generation.
- It pairs a Verilog agent with a Python reference-model agent for mutual verification.
- No golden testbench is required at generation time.
- Supports air-gapped security requirements for chip vendors.
- Can be trained on proprietary RTL codebases.
- Addresses misalignment with industrial practice.
- arXiv paper ID: 2605.12857.
- Overcomes limitations of single-turn generators.
Entities
Institutions
- arXiv