Siemens Adds Self-Verification for Long-Running EDA Agents
Siemens recently expanded its strategic partnership with Nvidia to add new AI capabilities to the Fuse EDA AI Agent, enabling self-verifying agentic AI workflows for semiconductor and printed-circuit board (PCB) design. The goal is to move toward agents that can remain active throughout long-running semiconductor and PCB design workflows.
The new capability combines Siemens' electronic design automation (EDA) software with Nvidia's AI open models, agent software, secure runtime, and accelerated computing stack.
It also spans Siemens' EDA technologies, including Catapult high-level synthesis, custom IC design and verification with Solido, Aprisa physical implementation, Calibre signoff verification, and Tessen design-for-test.
The Fuse EDA AI Agent system, now integrated into Siemens' recently launched Intelligence Center X, supports agent creation and orchestration in its enterprise industrial AI environment. The self-verifying AI workflows enable autonomous and long-running EDA AI agents to continuously validate their decisions against proven engineering tools. It allows the agent to monitor a workflow, inspect the results, change its configuration, and run other operations while checking its progress against established EDA engines.
This is what Siemens calls self-verification. The language model does not, by itself, determine whether a design is correct. Instead, the agent uses simulation, verification, and physical design tools to produce the engineering evidence for the next decision.
For example, library characterization using Siemens' Solido Characterization Suite with agentic AI can reduce characterization turnaround by more than 10× and token cost by 5× to 10×, according to the company. These agentic workflows automatically generate and verify Liberty files via Solido Characterizer with Solido LibSPICE, Solido Generator, and Solido Analytics.
The journey toward self-verifying agents
At DAC 2025, Siemens introduced the Fuse EDA AI System, which provided the data and model infrastructure to use generative AI with EDA data.
The Fuse EDA AI Agent, announced at Nvidia GTC in March 2026, added an execution layer. The supervisory agent can break an engineering objective into tasks, delegate work to specialized agents, and connect to EDA applications through Model Context Protocol (MCP) interfaces.
This new expansion, released at DAC 2026, adds long-running agents that can keep working after the first tool call. For example, an agent can launch a simulation, inspect the returned logs, identify a failure, modify the setup, and try again.
The Fuse EDA AI Agent translates an engineer's objective into a plan and the specialized agents select tools and execute the steps. Extending the agentic AI workflows with Nvidia's Nemotron 3 Ultra model provides reasoning for planning and multi-step execution, while the Nvidia NeMo Gym open library is used to evaluate agent behavior to improve results and optimize workflows. It also leverages Nvidia OpenShell secure runtime to provide a secure environment for running the agents.
But the engineering result comes from Siemens' EDA tools. For instance, a model may decide which SPICE analysis to run or if another process, voltage, and temperature (PVT) corner is needed, but the simulator produces the electrical result.
"The goal is not to replace physics with AI," said Timothy Costa, general manager for industrial and computational engineering at Nvidia. The objective is to connect AI reasoning with accelerated simulation and physics-aware models and keep trusted engineering tools as the source of validation, he added.
How a long-running agent handles library characterization
Library characterization shows why this type of agent can be useful even when the underlying simulations are already automated.
In the case of a standard-cell, memory, or custom-IP library, it has to be characterized across different operating conditions before its timing and power models can be used. Engineers first configure the library and run preliminary simulations to confirm that the environment is working correctly.
Amit Gupta, senior vice president and general manager of Solido Custom IC and chief AI strategy officer, at Siemens EDA, Siemens Digital Industries Software, described this process as "pipe cleaning."
If simulations fail because of an incorrect setup or convergence issue, an engineer has to inspect the result, change the configuration, and rerun the task. Once the setup is stable, the full characterization run begins.
Engineers continue to monitor failures, review outliers, and decide if additional PVT corners need to be simulated before the Liberty models can pass quality checks. "It comes out to about a four-to-six-week process for a library," Gupta said.
The Solido Characterization agent automates much of this supervisory loop. Gupta said an engineer can initiate the process with a single prompt. The agent can launch hours-long runs, monitor them, react to failures, change configurations, rerun jobs, and perform final quality assurance steps.
The workflow combines Solido Characterizer, Solido LibSPICE, Solido Generator, and Solido Analytics. Characterizer manages the characterization tasks, LibSPICE runs the circuit simulations, Generator produces the Liberty models, and Analytics examines the results and supports outlier analysis.
The productivity gain comes from reducing the manual work between simulations. Nvidia CUDA-X libraries can also accelerate numerical operations used by the simulation and physical verification tools to reduce the time between an agent's action and the feedback it needs for the next decision.
Nvidia's PhysicsNeMo open-source PyTorch framework for physics AI models provides another layer where physics-aware models can give the agent fast estimates to explore a design and decide where a higher-fidelity simulation is required. But Costa said those models are not a substitute for engineering validation. According to him, trusted solvers, test data, and domain-specific criteria are the final references.
Cross-domain autonomy remains the larger engineering challenge
The Solido characterization example is a workflow with well-defined inputs, tools, and acceptance checks. Extending the same autonomy across several EDA domains is more difficult.
Gupta said Siemens is currently focused on long-running agents within individual areas such as custom IC design, digital verification, design-for-test, and Calibre physical verification.
"What we see as the next step is being able to stitch those product families together," he added.
Siemens has demonstrated an early example, where Calibre detects a design rule violation, and the agent invokes Aprisa semiconductor software for a digital physical design repair or Solido for an analog repair.
This moves the agent from identifying a problem to changing the design, and it also creates new verification requirements. Fixing a local design rule check violation can create a timing, congestion, or signal integrity problem somewhere else in the design.
The challenge is more than exposing additional applications through MCP. Even a cross-domain agent has to understand how different representations of the same design correlate to one another.
Gupta explained that Siemens is developing EDA ontologies that map relationships between different domains and improve the models' ability to understand the different data types. This context is necessary because a change in one stage may invalidate results elsewhere.
Human judgment remains part of the design and verification workflow. When an agent repairs a design rule violation, it can create another physical problem, he added. Engineers therefore need to observe the agent's actions, pause or stop execution, and redirect the workflow.
Near-term deployment of this capability will remain domain-scoped and supervised. But more full-scale autonomy comes when EDA companies can preserve engineering intent, while agents can move across multiple tools, data models, and verification steps.
This shows that agentic EDA is not replacing the physics-based tools used to design and verify chips. The agent can keep the workflow moving and decide what should happen next, while the deterministic EDA engines and human engineers determine if the results can be trusted.
The expanded EDA AI capabilities will be available in upcoming releases of Siemens' AI-native EDA portfolio.
Related articles
Siemens Agentic Toolkit Automates Chip Verification Workflows
A Look at Agentic AI in the EDA Engineering Workflow
The post Siemens Adds Self-Verification for Long-Running EDA Agents appeared first on Embedded.