OpenROAD: The Journey So Far and the Roadmap

Prof. Andrew Kahng of UC San Diego delivered a keynote at ORConf on September 15, 2023, presenting a comprehensive overview of OpenROAD — from its origins as a DARPA-funded project to its current status as a widely adopted open-source EDA platform, and the roadmap ahead.

Download the original presentation (PPTX, 39MB)

What is OpenROAD?

OpenROAD was funded by U.S. DARPA as part of the Electronics Resurgence Initiative (ERI), with UC San Diego as the prime contractor. Running from June 2018 through December 2023, the project’s mission was to democratize IC design by directly attacking three barriers:

• Schedule barrier: RTL-to-GDS in 24 hours
• Expertise barrier: No-human-in-loop, tapeout-ready GDS
• Cost barrier: Open-source, eliminating expensive license fees

The motivation was clear: ASIC design in advanced nodes faces enormous barriers of cost, expertise, and risk. Innovators cannot evaluate PPAC (power, performance, area, cost) metrics of their design ideas without access to sophisticated — and expensive — proprietary tools requiring large teams of experts.

The Journey So Far

From Research to Real-World Impact

By 2023, OpenROAD had achieved remarkable traction:

• 600+ tapeouts at process nodes from 180nm down to 12nm
• 18,000+ commits from 84 contributors on the OpenROAD application
• Education and workforce development programs spanning high school to graduate level, including university extension courses

The project gained “virality” across multiple dimensions — from New York design high schools and community colleges, to university programs with Efabless ChipIgnite, VLSI Symposium workshops, integration into multiple open-source EDA flows, and AWS cloud deployment.

Lessons Learned

The keynote shared candid lessons from five years of building open-source EDA:

Building the right team matters. The original plan was to develop OpenROAD entirely with Ph.D. students and post-docs. Within 9 months, the need for experienced EDA architects became clear. EDA veterans were brought in and became essential — coaching students while providing the industrial-strength engineering that production tools demand.

Users are not all the same. The original vision assumed users would actively contribute enhancements and pull requests. In reality, this type of user is rare — vastly outnumbered by those who want a cost-effective alternative to commercial tools without necessarily contributing back to the codebase.

Open-source EDA is still EDA. There has never been a successful EDA tool without amazing application engineers and power users behind it. The strongest contributors combine software skills with the right mindset — often CS-background students from around the world, coached by EDA veterans.

The Roadmap

Cloud and ML Integration

The roadmap envisions a future where tool licenses are unlimited and cloud-native. COPILOT (Cloud Optimized Physical Implementation using OpenROAD Technology) represents the direction of cloud-based chip design. ML challenges include predicting failures and intervening early, alongside practical improvements like AutoTuning.

Early Design Space Exploration

A major focus is enabling architects to explore design spaces earlier in the process:

• Hier-RTLMP (/src/mpl2): RTL and dataflow-driven macro placement that mimics human expert-like decision making
• TritonPart (/src/par): Timing- and constraint-driven partitioning for hierarchical design

Tool Development Priorities

Development priorities are determined by aligning to recurring user issues and requests, striving for the biggest impact to the most users. Key areas highlighted:

• Synthesis improvements: QoR from Yosys + ABC needs improvement — this has been very consistent feedback from users
• CTS usability and QoR: Dedicated developer time is being invested to reduce manual parameter tuning and support hierarchical CTS
• Proxy PDKs: Building out ASAP7/ASAP5 PDK enablements to provide sharable, representative technology benchmarks

Governance and Transition

With the DARPA contract ending December 31, 2023, sustainability was top of mind. Two organizations now anchor the project:

Precision Innovations, Inc. — the principal industrial developer and integrator of OpenROAD. Founded in 2019, staffed by EDA veterans, with key customers including Google and Intel, and partners including Siemens and GlobalFoundries.

The OpenROAD Initiative — a 501(c)(3) nonprofit founded in February 2023 by Tom Spyrou to foster open-source EDA and semiconductor education, ensuring the project’s long-term community governance.

ML Enablement for OpenROAD

A significant portion of the keynote covered OpenROAD’s role as an ML-for-EDA playground:

• CircuitOps: A standardized data format using labeled property graphs (LPG) represented as deep graph library (DGL) objects, enabling ML-friendly data extraction from EDA tools
• ML-centric APIs: Python APIs for extracting timing data, congestion maps, DRC violations, and IR drop as ML-ready datasets
• ChatEDA: An autonomous EDA agent powered by LLMs that provides a conversational interface for human-toolflow interaction — from natural language input through task planning, script generation, and task execution
• RL-based optimization: Training reinforcement learning agents within OpenROAD for automated design optimization

Summary

By its sixth year, OpenROAD had gone “viral” across community engagement, education and workforce development, and commercial use. The formal end of DARPA funding marked not an ending but a transition — with Precision Innovations providing commercial support and the OpenROAD Initiative ensuring community governance and educational mission continuity.

The keynote’s background thoughts offered strategic perspective: infrastructure is becoming commodity (“pick one, move on”), the quality bar matters for adoption, and proxy PDKs are essential for removing barriers to wider use of open-source EDA.