Author: Harry Foster, Siemens EDA
For decades, advances in electronic design automation (EDA) went hand in hand with faster engines. Verification teams relied on more powerful simulators, scalable formal tools, and higher-capacity solver engines to keep pace as designs grew.
This approach is now reaching its limit.
In modern systems-on-a-chip development, the real bottleneck is no longer the raw throughput of the tools. It's coordination. Verification has become a continuous, adaptive process. Engineers must interpret results, refine intent, and adjust strategies across multiple tools and iterations. Increasingly, productivity is limited by workflow complexity, not computing power.
From automation to workflow intelligence
Traditional automation enables stable inputs and predictable flows. RTL verification doesn't work that way. Designs evolve, specifications change, and intermediate results often reveal new risks.
This is where agentic AI is gaining ground.
Instead of optimizing individual steps in silos, agentic systems operate through workflows. Agents monitor the verification status, plan specific actions, execute tasks, and synthesize the results.
Integration matters. AI that sits outside the toolchain, analyzing logs or generating scripts, can increase review effort and decrease confidence in the results. In contrast, integrated approaches allow AI-generated actions to be evaluated using the same coverage models, semantic criteria, and checks required for final validation.
Keeping humans in the decision-making process
Despite rapid advances in AI, full autonomy remains unrealistic for RTL verification. And it's not even desirable.
The goal is not to replace engineers, but to reduce manual coordination so they can focus on decision-making and risk mitigation.
Verification decisions depend on incomplete specifications, implicit assumptions, and technical compromises that require human expertise. Determining "good enough" results is not an absolute.
Therefore, effective agentic systems are designed to be human-centered. AI agents assist by proposing actions, executing tasks, and providing information. Engineers retain control over all decisions affecting intent, scope, and final approval.
In practice, this means establishing explicit approval points in the workflow. AI accelerates execution and analysis, but the final validation authority remains with the engineer.
(Source: Siemens Digital Industries Software)
A solid foundation for reliable workflows
Implementing this model requires more than simply adding AI features to existing tools. It depends on an architecture that exposes verification engines in a structured and semantically meaningful way.
The engine's native interfaces allow agents to run tools, obtain results, and directly observe the system's state. Instead of relying on unstructured input, these interfaces provide controlled entry points, such as running simulations, querying coverage, and performing failure analysis.
Context matters. By maintaining relationships between designs, tests, assertions, and historical results, agentic systems can reason through iterations. This continuity is critical in complex verification environments where each step builds upon previous results.
Where does agentic AI add value?
The first applications of agentic AI are already improving productivity in various RTL workflows:
- RTL Development: AI-assisted code generation allows you to align design goals with verification requirements and detect issues early.
- Lint and static analysis: Smarter configuration and contextual filtering help reduce noise and focus attention on real problems.
- Cross-clock domains (CDC): Iterative analysis and refinement accelerate convergence towards robust asynchronous designs.
- Verification planning: AI agents can translate changing specifications into structured plans that adapt as designs change.
- Depuration: By correlating the waveforms, assertions, and records of all performances, the possible underlying causes are discovered more quickly.
In all these cases, a consistent pattern emerges: limited automation, a robust context, and human oversight. The result is a quantifiable increase in productivity without compromising rigor.
Risk management and maintaining trust
Like any automation, agentic AI introduces new risks. Less constrained systems could propagate errors or conceal intentions.
The solution is not to limit capacity, but to strengthen the structure.
Bounded actions restrict what AI agents can do. Engine-qualified validation ensures that results meet production standards. Structured interfaces reduce ambiguity. Mandatory human review maintains accountability at key decision points.
This aligns with the reality of verification, the responsibility for which ultimately lies with engineering teams. In this model, AI augments expertise rather than replacing it.
Looking to the future
Agentic AI in electronic design automation is constantly evolving, but the direction is clear. The current goal is to reduce friction in planning, execution, and analysis, while maintaining engineer control.
In the long term, it could evolve into a broader orchestration, but only if supported by transparency and robust validation.
As designs become more complex, the greatest benefits will come from improving how work is done, not just the speed of execution of the tools. Agentic AI represents a shift toward workflow intelligence, helping teams move faster while preserving the rigor needed for final validation.
For verification teams facing ever-increasing iteration overload, this change comes at a perfect time.
