Siemens has rolled out Fuse EDA AI Agent, a domain-specific autonomous agent aimed at automating workflows across semiconductor, 3D integrated circuit (IC), and printed circuit board (PCB) system design, verification, and manufacturing.

The system coordinates multi-tool and multi-agent processes spanning Siemens’ electronic design automation (EDA) portfolio and leverages Nvidia’s Agent Toolkit, advanced Nemotron models, and Nvidia AI infrastructure as part of its core architecture.

Fuse EDA AI Agent expands on the earlier Fuse EDA AI system by integrating a retrieval-augmented generation (RAG) pipeline, multimodal EDA data lake, and specialised file format parsers.

Siemens Digital Industries Software chief AI strategy officer and Siemens EDA senior vice president and general manager Amit Gupta said: “Fuse EDA AI Agent represents the next evolution of our Fuse EDA AI system, moving from in-tool AI capabilities to autonomous, end-to-end workflow orchestration.

“We are delivering intelligent automation across the complete EDA lifecycle, enabling our customers to dramatically reduce design cycles while maintaining the highest quality standards.

“Our open architecture allows customers to integrate their own workflows and models, providing the flexibility required for enterprise-scale AI deployment. This positions the industry to maintain a competitive advantage in an increasingly complex semiconductor and PCB system landscape.”

To support this open approach, Fuse EDA AI Agent offers customisable access controls, supports multiple AI models, and is compatible with third-party integrations, addressing the requirements of complex chip and PCB development workflows.

Across the design lifecycle, from initial planning through verification to manufacturing sign-off, Fuse EDA AI Agent aims to deliver end-to-end automation. In early-stage architectural exploration, design planning, and RTL coding, the agent can automate tasks through Siemens’ Catapult software.

For digital verification, the agent integrates with the Questa One Agentic Toolkit, while for physical implementation it supports place-and-route, timing closure, and power optimisation through Aprisa.

Custom analogue and mixed-signal design is also supported through acceleration features in Solido software while hardware-assisted verification gets a boost from Veloce systems.

For physical verification sign-off, the agent automates DRC violation analysis and resolution using Calibre software. It also manages 3D IC challenges such as power/ground load optimisation and signal path clustering in Innovator3D IC software.

For PCB system design tasks including layout and signal integrity analysis, the tool works with Xpedition and Hyperlynx products. Manufacturing readiness steps are managed via Tessent software for design-for-test workflows as well as Calibre OPC products for optical proximity correction.

Siemens claims that the Fuse EDA AI Agent is a direct response to limitations observed in generic AI solutions when applied to the EDA domain. Standard commercial AI tools lack deep domain expertise needed to interpret dense EDA datasets composed of physics-based information specific to semiconductors and PCBs.

Broader-purpose agentic AI platforms can introduce exposure risks for intellectual property due to weak access controls or inadvertent use of open cloud resources which may leak sensitive data. Furthermore, modern EDA tool chains are highly complex, which is a factor that often overwhelms off-the-shelf models’ context windows and leads to unreliable outputs or model hallucinations.

Fuse EDA AI Agent includes validation logic and task-specific guardrails to prevent workflow errors. Agent Skills enable executable playbooks for complex tasks, allowing users to customise automation sequences for their projects.

The architecture of Fuse EDA AI Agent is built around scalable orchestration via Managed Compute Platform (MCP) connectivity.

A hierarchical structure uses a supervisor agent for planning and worker agents to execute subtasks across Siemens and third-party tools, enabling dynamic tool discovery, error recovery, and efficient user-agent interaction.

Fuse EDA AI Agent is engineered to fit into enterprise IT ecosystems by optimising compute resources across existing scheduling frameworks or high-performance computing (HPC) clusters. Centralised data orchestration allows teams to operate within air-gapped environments while extracting cross-workflow analytics from project datasets.

Security features embedded at the platform level include native support for role-based access control systems, automatically generated audit trails, and human checkpoint capabilities for safeguarding proprietary design data during automated flows. These measures align with requirements for operating in secure enterprise environments where intellectual property protection remains paramount.

The launch coincides with an ongoing partnership between Siemens and Nvidia focused on advancing autonomous workflow agents for semiconductor and PCB system domains.

As part of this relationship, Siemens has optimised Fuse EDA AI Agent to take advantage of Nvidia GPUs as well as Nemotron models tuned for tool-call reliability and reasoning accuracy in multimodal project environments. This joint engineering effort aims at improving RAG accuracy across large-scale datasets while reducing operational costs through more efficient model token usage.

Looking forward, Siemens plans further integration of Fuse Agent with Nvidia’s evolving enterprise-grade agentic stack, including interfaces like NemoClaw that facilitate secure deployment of always-on assistants. OpenShell runtime will enable controlled execution of autonomous agents within industrial environments.

Recently, Siemens introduced the Questa One Agentic Toolkit as part of its Questa One smart verification suite to improve the speed and efficiency of IC design and verification.