Architecture 2.0: Toward Open Source Foundation Models and Datasets for Hardware Design
summary
Argues for “Architecture 2.0”: build open, hardware-specific foundation models trained on open-source datasets (RTL, netlists, layouts, metrics) to cut cost and cycle time; identifies data scarcity, IP limits, and simulator trade-offs (cycle-accurate vs analytical) as the main blockers; delivers a tooling prototype Arch Gym (ISCA ’23) that links ML agents to OpenLane and other simulators and shows many algorithms similarly find optima; prioritizes crowd-sourced data, then fine-tuned models with RTL generation and design-space exploration demos.
AI-written summary · about
chapters (8)
- 0:00Architecture 2.0 intro
- 2:47Defining Architecture 2.0
- 4:37Hardware design datasets
- 6:02ML algorithms for hardware
- 7:25ML-driven design tools
- 9:53Workforce training needs
- 11:01Open-source community
- 12:45Q&A
related talks
26mWhy Has Hardware Infrastructure Diverged From Open Software?
01h 18mReconfigurable Architectures: Generalities,Open-source and emerging-technologies
01h 05mChallenges and opportunities for Open Source Hardware in academia and industry: Roundtable
14mAutomating Design Space Exploration Using Advanced Simulation Technologies
12mOpen source EDA for Machine Learning
10mKeynote: The Open Source Hardware Roadmap
16mOpen-source silicon chips
43mOpen Source Silicon: Where software ends and hardware begins