Coco boosts hardware and software design with an intelligent assistant
Coco: An Agentic Copilot for the Hardware--Software Co-Design Lifecycle
Programming LanguagesArtificial Intelligence
Summary
Designing both machine learning hardware and the models they run is very complex and requires analyzing huge amounts of new simulation data every few months. The authors present Coco, a smart assistant that helps engineers organize data, run experiments, and analyze results efficiently without making mistakes. Coco understands recurring tasks and provides an interactive interface that fits the experts' workflows. Early tests with TPU designers show it speeds up decision-making during co-design.
What this means in practice
- •For hardware architects: Shorten the time to set up and analyze hardware simulations for designing ML accelerators by automating data organization and workflows with smart agents.
- •For software performance engineers: Improve insights into joint software-hardware trade-offs during design by using a tool that compares system configurations systematically and interactively.
Authors
Samuel Kushnir, Kavya Sreedhar, Yeshwanth Reddy Pogula, Amir Yazdanbakhsh, Narges Shahidi, Ming Liu, Varun Gohil, Ravi Iyer, Parthasarathy Ranganathan, Christina Delimitrou, Suvinay Subramanian
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.