Coco: An Agentic Copilot for the Hardware--Software Co-Design Lifecycle

2026-10-01

AI Summary

This study addresses the issues of large language model hallucination and analysis overload caused by missing data in hardware-software co-design. To this end, it proposes an intelligent agent platform tailored for TPU architects. Methodologically, the work establishes a “retrieve rather than memorize” data paradigm, formulating co-design as an independent agent domain. By integrating relational storage, SQL grounding, a typed API tool library, and multi-agent workflows, the platform enables orchestration-free tool composition and experiment automation. Experimental results demonstrate that the proposed platform significantly shortens simulation and insight cycles while improving expert adoption rates, thereby providing effective support for high-confidence system design decisions.

Abstract

Co-designing ML models and the accelerators that run them is an unusual reasoning task: architects must draw confident, high-stakes conclusions about systems that do not yet exist, and the pace of both model evolution and hardware cadence means the analysis burden grows every quarter. The evidence behind each decision--hundreds of gigabytes of fresh simulation sweeps over novel design points--is by construction absent from any LLM's pretraining corpus, and there is no external literature to retrieve; naive "chat-with-your-data" approaches hallucinate exactly where correctness matters most. We present Coco (Copilot for Codesign), an agentic platform deployed with TPU architects that accelerates the co-design lifecycle of setting up experiments, sweeping simulators, and deriving insights. Coco is built as four layers: (i) a datastore that automatically registers every simulation sweep into a normalized relational schema, so agents ground every number in a SQL query rather than scraping heterogeneous files; (ii) a library of tools with typed APIs that agents compose without human orchestration; (iii) agents that encode recurring analysis workflows--most notably iso-execution analysis, which compares systems at matched execution configurations, including swept-but-dominated points off the Pareto frontier; and (iv) a platform UX whose navigation state doubles as agent context. We report early deployment experience toward a reduction in time-to-simulation and time-to-insight, and argue that co-design is a distinct agentic domain: its data must be retrieved rather than memorized, its workflows are recurring but context-dependent, and expert adoption hinges on UX that balances IDE-style control with interactive exploration.

Problem

• Hardware-Software Co-Design
• ML Accelerators
• Agentic Copilot
• Simulation Data Analysis
• LLM Hallucination

Innovation

• Hardware-Software Co-Design
• Agentic Platform
• LLM Hallucination Mitigation
• Iso-Execution Analysis
• Relational Datastore