What Self-Verifying Means In Agentic EDA Workflows And Why It Matters

Last month, we covered the architectural decisions behind a production-ready EDA AI agent: domain grounding, scalable orchestration across a fragmented tool ecosystem, native interpretation of EDA data formats, and security at the execution layer. These are the foundations upon which successful agentic workflows for EDA are built, but the next layer is something that all engineers will ask about: trust. When an agent is executing a long-running EDA workflow autonomously, making countless interdependent decisions across multiple tools without interruption, how do you know the results can be trusted?

Why deterministic physics-based EDA engines matter

The answer is physics-based EDA engine verification. Physics-based, deterministic EDA engines deliver trusted, signoff-quality results and define the ground truth against which every EDA AI agent decision is validated. Every decision the agent makes is checked against the physics of the design before the next step begins, not at the end of the workflow or after an entire run completes, but continuously, at every step, against the same engines engineers have trusted for decades. Grounding agent decisions in their output is what transforms orchestration into something engineers can build on with confidence.

Why does this matter in EDA? In semiconductor and PCB design, there is no room for error and no tolerance for delay. Every RTL block, every SPICE simulation, every Liberty file matters. Tapeout dates are fixed, respins are expensive, and missed market windows do not come back. For agents operating autonomously across these workflows, there is zero margin for unverified decisions.

How EDA engines and agents work together

The interface between an agent and an EDA engine is fundamentally different from the interface a human designer uses, and building that interface required adding additional capabilities at the orchestration level specifically to interact with agents.

When a human designer is interacting with an EDA tool, they do not need to know all the internals of the engine and what the results are at that level of granularity. An agent, on the other hand, can take advantage of all of that information to reason, act, check with the tool, get feedback, and converge on the correct result. That is what makes continuous physics-based verification possible: the agent is in a continuous cycle with the tools themselves, generating data, verifying it, and if it is going off track, correcting before moving forward.

Agent Skills encode domain knowledge as executable playbooks, giving the agent not just the steps to follow but the context to execute them correctly, with validation built into every step. A scalable MCP architecture enables dynamic tool discovery and orchestration across Siemens’ full EDA portfolio, from high-level synthesis with Catapult, verification with Questa One and Veloce, and custom IC design and verification with Solido, through to physical implementation with Aprisa, signoff verification with Calibre and design-for-test with Tessent, as well as 3D IC integration with Innovator3D IC and PCB design with Xpedition, mitigating context saturation as workflows expand.

The engineer remains in the process throughout. They can see what the agent is doing and the steps it is taking, and where the agent has done something incorrectly, they can augment it, storing that correction as new institutional knowledge in Agent Skills files that then go on to improve every subsequent agentic run. Agent Skills can also be configured to request human-in-the-loop approval before executing consequential steps, for example, checking with the engineer before launching a 5000-core SPICE job or a full DRC run.

What an autonomous, self-verifying workflow looks like in practice

At DAC 2026, Siemens announced self-verifying, long-running EDA AI agents for semiconductor and PCB design, combining Siemens EDA expertise and software with NVIDIA AI infrastructure. Library characterization with Solido Characterization Suite was one of the workflows demonstrated.

Library characterization has historically been one of the most labor-intensive steps in advanced node design and an extremely important one to later stage design phases. Generating and verifying Liberty files across process corners for standard cell and custom IP libraries typically demands multiple full-time engineers, weeks of runtime, and constant manual intervention.

In the demo shown at DAC, the agent receives a single instruction: “Characterize my libraries across all PVTs. Verify, debug, and fix any problems. I am going to bed and will not be available to answer any questions. Resolve any issues on your own.” Then it runs. When it encounters a problem, skill files guide the automated debug sequence. The agent identifies the issue, resolves it independently, and continues without escalation or handoff. Config setup that once consumed hours now takes minutes, pipecleaning that stretched across days wraps up in hours, and the full flow, previously a weeks-long effort, completes in days. Token costs are reduced by approximately 5X to 10X through NeMo Switchyard and Nemotron models, with an additional 25% reduction from NeMo Gym-based optimization of Agent Skills and MCP tools, delivering an overall 10 times faster autonomous characterization workflow, with every decision continuously validated against physics-based EDA engines at every step.

From autonomous tasks to trusted engineering outcomes

Over the past twelve months, each step in Siemens’ agentic EDA journey was a prerequisite for the next: domain grounding, scalable orchestration, and continuous physics-based validation. Physics-based EDA engine verification is the layer that completes the architecture, because it is what allows engineering organizations to move from autonomous task orchestration toward continuously validated engineering outcomes. The result is self-verifying, long-running EDA AI agents that engineers can trust with their most demanding workflows.

To find out more:

• Watch the library characterization demo to see the autonomous EDA workflow in action
• Hear the DAC conversation with Tiff in Tech: how Siemens EDA and NVIDIA are building the AI-native electronic design flow
• Read last month’s article for the full architectural background on what makes an EDA AI agent production-ready

The post What Self-Verifying Means In Agentic EDA Workflows And Why It Matters appeared first on Semiconductor Engineering.