, , , ,
ChipMATE is a groundbreaking multi-agent framework for RTL code generation that addresses the limitations of existing API-based systems. These systems often assume the availability of a golden testbench at generation time, rely on closed-source APIs that do not align with chip vendors' security requirements, and cannot be trained on vendors' proprietary RTL codebases. To bridge these gaps, ChipMATE introduces a novel approach inspired by industrial practices where correctness emerges from cross-comparison between independently written RTL modules and reference models. The framework pairs a Verilog agent with a Python reference-model agent, enabling mutual verification of outputs without relying on a golden oracle. ChipMATE incorporates a backtrack-based inference workflow to prevent error propagation across turns and implements a two-stage training pipeline. In the first stage, each agent is individually trained to maximize its code-generation capability, followed by joint training to enhance collaboration between the agents effectively. To support the training process, ChipMATE utilizes a hybrid data-generation framework that produces 64.4K high-quality reference model training samples. The framework achieves impressive results, with pass rates of 75.0% and 80.1% on VerilogEval V2 using 4B and 9B base models respectively. Notably, ChipMATE outperforms all existing self-trained models and even surpasses DeepSeek V4 with 1600B parameters. The authors have made both the code and model weights publicly available on GitHub (https://github.com/zhongkaiyu/ChipMATE), showcasing their commitment to transparency and reproducibility in research. With its innovative approach to multi-agent training via reinforcement learning, ChipMATE represents a significant advancement in RTL code generation that holds promise for enhancing industrial practices in this domain.
Summary- ChipMATE is a special tool that helps create computer code in a better way than before.
- It uses a new idea based on how things are done in real companies, where being correct comes from checking different parts of the code against each other.
- ChipMATE has two agents, one using Verilog and the other using Python, to check if the code works right without needing a perfect example to compare to.
- It also has ways to fix mistakes and trains its agents in two steps to make them better at their job.
- ChipMATE makes its own data for training and does really well compared to other similar tools.
Definitions
- Framework: A structure or system that helps organize and solve problems.
- Verification: Checking if something is correct or works as expected.
- Inference: Figuring out something based on what you already know.
- Pipeline: A series of steps or stages in a process.
- Transparency: Being open and clear about what you are doing.
Introduction
ChipMATE is a research paper that introduces a novel multi-agent framework for RTL code generation. This framework addresses the limitations of existing API-based systems and aims to improve industrial practices in this domain. The authors have made their code and model weights publicly available on GitHub, showcasing their commitment to transparency and reproducibility in research.
The Need for ChipMATE
Existing API-based systems for RTL code generation have several limitations that hinder their effectiveness. These systems often assume the availability of a golden testbench at generation time, rely on closed-source APIs that do not align with chip vendors' security requirements, and cannot be trained on vendors' proprietary RTL codebases. As a result, there is a need for an innovative approach to overcome these challenges and improve the efficiency of RTL code generation.
Inspiration from Industrial Practices
The authors of ChipMATE drew inspiration from industrial practices where correctness emerges from cross-comparison between independently written RTL modules and reference models. This approach allows for mutual verification of outputs without relying on a golden oracle, making it more robust and adaptable to different scenarios.
The Framework
ChipMATE pairs a Verilog agent with a Python reference-model agent, enabling effective collaboration between the two agents. The framework incorporates a backtrack-based inference workflow to prevent error propagation across turns, ensuring high-quality results. Additionally, ChipMATE utilizes a hybrid data-generation framework that produces 64.4K high-quality reference model training samples to support its training process.
Training Pipeline
The training pipeline in ChipMATE consists of two stages - individual training and joint training. In the first stage, each agent is trained individually to maximize its code-generation capability. Then in the second stage, joint training is performed to enhance collaboration between the agents effectively.
Impressive Results
The results achieved by ChipMATE are impressive, with pass rates of 75.0% and 80.1% on VerilogEval V2 using 4B and 9B base models respectively. This outperforms all existing self-trained models and even surpasses DeepSeek V4 with 1600B parameters.
Availability
One of the key strengths of this research is its commitment to transparency and reproducibility. The authors have made both the code and model weights publicly available on GitHub (https://github.com/zhongkaiyu/ChipMATE), making it accessible for other researchers to replicate their results.
Conclusion
In conclusion, ChipMATE is a groundbreaking multi-agent framework for RTL code generation that addresses the limitations of existing API-based systems. Its innovative approach inspired by industrial practices, efficient training pipeline, and impressive results make it a significant advancement in this field. With its availability on GitHub, ChipMATE holds promise for enhancing industrial practices in RTL code generation and paving the way for future research in this domain.
Existing API-based agentic systems for RTL code generation are fundamentally misaligned with industrial practice: they assume a golden testbench is available at generation time, rely on closed-source APIs incompatible with chip vendors' air-gapped security requirements, and cannot be trained on vendors' proprietary RTL codebases, leaving valuable internal data unused. Recent self-trained models address the deployment constraint but remain single-turn generators that overlook the critical role of verification in real industrial flows. To bridge these gaps, we present ChipMATE, the first self-trained multi-agent framework for RTL generation. Inspired by industrial practice where correctness emerges from cross-comparison between independently written RTL modules and reference models, ChipMATE pairs a Verilog agent with a Python reference-model agent that mutually verify each other's outputs without any golden oracle. We design a backtrack-based inference workflow to prevent error propagation across turns, and a two-stage training pipeline that first trains each agent individually to saturate its code-generation capability, then trains the team jointly to collaborate effectively. To support the training, we further build a hybrid data-generation framework that produces 64.4K high-quality reference model training samples. ChipMATE achieves 75.0\% and 80.1\% pass@1 on VerilogEval V2 with 4B and 9B base models, outperforming all existing self-trained models and even DeepSeek V4 with 1600B parameters. Our code and model weights are publicly available in https://github.com/zhongkaiyu/ChipMATE.
