An AI system coordinates chip, 3D IC, and PCB design workflows, connecting engineering tools to automate planning, verification, and manufacturing sign-off.
Siemens has introduced the Fuse EDA AI Agent system, an autonomous AI orchestration layer that coordinates semiconductor, 3D IC, and PCB design workflows across design, verification, physical implementation, and manufacturing sign-off. It is designed to automate multi-tool engineering processes in EDA environments while maintaining domain constraints, tool interoperability, and enterprise-grade governance.
The agent operates across Siemens’ EDA portfolio, connecting tools such as Catapult for RTL design, Questa for verification, Aprisa for physical implementation, Solido for custom design, Veloce for hardware-assisted verification, Calibre for sign-off, Innovator3D IC for advanced packaging, Xpedition and HyperLynx for PCB design, and Tessent for test and manufacturability. It executes end-to-end workflows including architecture exploration, testbench generation, place-and-route, timing closure, signal integrity analysis, DFT, and optical proximity correction.
Built on NVIDIA Agent Toolkit and Nemotron models, the system uses a hierarchical agent architecture that breaks engineering tasks into supervised and worker-agent roles. It can dynamically discover and invoke tools through MCP connections, execute multi-step “Agent Skills” workflows, and recover from failures during execution. The goal is to reduce manual coordination across fragmented toolchains and mitigate context limitations in large engineering workflows.
The Fuse EDA AI Agent is layered on Siemens’ existing Fuse EDA AI platform, which provides retrieval-augmented generation pipelines, a multimodal EDA data lake, specialized parsers for design files, configurable access controls, and support for multiple AI models. Its open architecture also allows integration with third-party EDA tools and external systems.
Enterprise deployment is a core constraint of the system. It integrates role-based access control, audit logging, and human-in-the-loop validation steps to ensure traceability and prevent unauthorized data exposure. It is designed to operate within high-performance compute environments, including air-gapped infrastructure, and can integrate with existing job scheduling systems.
Traditional AI systems struggle in EDA workflows due to highly specialized physics-driven data, long and interdependent toolchains, and strict intellectual property requirements. General-purpose models can also suffer from context limits and reliability issues in complex multi-step design tasks. The Fuse EDA AI Agent addresses these constraints by embedding domain-specific knowledge, structured orchestration, and controlled tool execution within a governed enterprise framework.
Click here for the original announcement.
