At a glance
| Booth | T9404 |
| Country | US |
| Website | www.leafylab.io |
Company profile
Leafy Lab describes itself as a NEXT-GEN DESIGN AGENT and pioneers agentic design optimization for the semiconductor value chain, building tools intended to enable autopilot for chips. The company targets software, verification and IP markets projected to exceed $197B by 2030 and seeks to expand this landscape through proprietary Intelligence Agents. Leafy Lab combines integrated chip-designer intuition with automated simulation pipelines to replace current trial-and-error design approaches with an agentic optimization workflow. Its platform is described as optimizing power, performance and area toward a global optimum, with an example of a 3.6 times improvement in performance. Leafy Lab links this optimization approach to increased first-silicon success and at least 50% faster time-to-market. The company therefore works at the intersection of semiconductor design, verification, layout optimization and predictive device modeling rather than supplying a conventional process tool or material.
Exhibits
Leafy Lab names the offering Agentic Design Platform. It is built as a multi-agent AI framework. The framework pairs designer agents that act as judges with layout agents that act as engineers, assigning complementary roles to agents within the same design-optimization process. Through continuous active-learning loops, the system autonomously navigates complex fabrication rules and trillions of design variables in parallel. Leafy Lab states that the agentic platform can transform design blueprints into optimized, tape-out-ready layouts in hours, compared with a task that previously required months of manual effort from experienced engineers. The platform's objective is not only automation of layout generation but optimization across power, performance and area while incorporating chip-designer intuition and automated simulation pipelines. Leafy Lab also presents Predictive Device AI for semiconductor device modeling. The company notes that standard AI models require millions of data points, while silicon data is notoriously sparse. Its explainable predictive models are described as learning from fewer than 50 measurements and predicting transistor behavior with a 2% testing error. The company contrasts that 2% testing error with traditional models reported at greater than 70% errors, presenting the predictive approach as a way to work with limited silicon data. Together, the Agentic Design Platform and Predictive Device AI address design-layout automation, exploration of very large design-variable spaces, tape-out preparation and transistor-behavior prediction for semiconductor design workflows.
Capabilities and products
- semiconductor agentic design optimization platform
- multi-agent AI chip-layout optimization
- tape-out-ready optimized layouts in hours
- transistor-behavior prediction from fewer than 50 measurements