Uncategorized

The Verification Evolution: How AI Is Raising the Stakes for Chip Design

verification

 

Every improvement in semiconductor design productivity enables faster iteration, and AI has transformed design like nothing else before it. Ironically, the speed of AI-enabled design is making engineering easier and harder—with verification sitting squarely in that gap.

Having more freedom for architectural exploration and iterating faster inherently demands tighter interaction between design and verification. Verification has always been a foundational discipline of semiconductor engineering, ensuring products behave according to specification before tapeout and sparing teams from mistakes that cost precious time and millions in rework.

But AI is changing the nature of the role because instead of simply increasing the volume of work, it is increasing the complexity of the work itself. 

As AI accelerates development, verification increasingly determines whether all that speed translates into trustworthy products or a rapidly growing tail of flawed designs. That’s why verification is becoming one of the semiconductor industry’s most strategic positions in the engineering value chain. 

 

With Systems so Complex, No Part of Design Can Succeed in Isolation 

 

Modern semiconductors have evolved beyond monolithic blocks integrated post-design. Architects are increasingly conceiving complex, heterogeneous designs where dozens of functional blocks operate in parallel under strict power, performance, and thermal constraints. This makes it more critical than ever for every layer to work in sync from day one. 

A functional block can perform flawlessly in isolation while the overall system behaves unpredictably once every layer begins interacting. Features that appear correct in unit testing may produce unexpected corner-case failures once deployed under real customer workloads. 

Verification teams now spend increasing amounts of time understanding how designs behave under realistic operating conditions.  

Emulation and prototyping allow software development long before production of silicon exists, exposing system-level interactions that synthetic testing cannot reveal. Success is measured less by whether an individual block passes its testbench and more by whether the entire platform performs predictably under load. 

That broader system view is also changing how industry leaders think about EDA tools. As Srikanth Jagabathula, CEO of Xcelsa Labs, notes: 

“Let’s not be constrained by how the systems and tools exist today because they were actually designed with a different set of assumptions forty years ago. Rather, let’s come with like a completely new perspective, an AI first perspective… we are taking an end to end approach wherein instead of focusing on single steps, we are trying to move the whole outcomes.”

 

 

AI is Solving Major Problems—And Creating New Ones 

 

Today, AI-powered tools generate assertions, optimize test coverage, accelerate regression analysis, and identify coverage gaps with a speed that human engineers might take weeks to uncover. As this work becomes increasingly automated, engineers gain time back to focus on other high-value activities. 

That acceleration is tangible. Jagabathula describes the traditional design loop as inherently iterative:

“This iterative process is very, very slow, time consuming. Engineers try something. They run the tool, which takes a long time. They look at the output, try to mentally figure out using their expertise, how to what to do next. They try other things and finally it converges… What [our platform] does is it brings AI to significantly accelerate this iteration process.” 

But here’s the paradox: The same capabilities that enable faster iteration and allow engineers to evaluate far more architectural possibilities creates additional design behavior that has be verified, tested for assertions, and brought back to beginning to evaluate through an architectural lens. 

It’s the classic “garbage in, garbage out” conundrum. 

The answer isn’t simply running more regressions. Verification has always been about establishing confidence, which ultimately depends on judgment. AI can suggest what to build, generate test cases, and speed up debug cycles. But it can’t determine whether a team is asking the right questions in the first place. 

That’s why verification engineers have to guard against teams creating mountains of unusable code. And why the best verification engineers aren’t distinguished by the sheer volume of regressions they execute.  

They best ones are characterized by the questions they ask before anyone else thinks to ask them. That judgment remains fundamentally human, and AI is also helping. 

 

Verification Is Moving Upstream 

 

verification

 

Increasingly supported by AI-enabled verification tooling, verification engineers are becoming more involved in the decisions that shape successful design long before RTL is complete.

Architectural partitioning, interface definitions, hardware/software boundaries, and overall system planning all influence the final objective of a design, and forward-thinking companies are recognizing that the verification strategy itself has become a more important part of the upstream considerations in designs. 

Rather than scaling verification teams late in the design cycle, forward-thinking organizations are integrating verification expertise early, hiring these key team members early.

This is also changing how engineering teams collaborate.  Verification engineers are increasingly collaborating with architects, designers and firmware teams further upstream, where their experience and judgment can help identify system risks while the architecture is still fluid.   

As hardware and software co-design becomes mandatory, success depends less on optimizing siloed functions and more on cross-disciplinary collaboration throughout development. 

 

What Companies Look For in the Modern Verification Engineer 

 

verification skillset

 

As chip complexity escalates, engineering profiles evolve. Technical depth in System Verilog, UVM, or formal verification will remain essential for a long time, and yet companies are prioritizing engineers with broader system-level awareness than they have in the past.

Forward-looking Engineering leaders are evaluating and hiring for: 

  • Hardware/Software Co-Design Fluency: Understanding how hardware choices impact software performance—and vice versa. 
  • AI Tool Integration: Knowing where automation drives speed versus where human inspection is mandatory. 
  • Cross-Domain Collaboration: Moving comfortably between architecture, RTL, firmware, validation, and customer applications. 

In a nutshell, they’re looking for verification engineers who understand how hardware decisions influence software behavior, and how software influences hardware performance, who are comfortable collaborating across a wider scope of functions, and who ultimately design for end applications rather than simply for functional correctness.

Jagabathula points to one underlying trait that separates top performers:

“Engineers need the curiosity to look around, learn from adjacent teams, and have the humility to realize how much they don’t know. A narrow focus—’I’m really good at this one task and don’t want to think beyond it’—becomes a bottleneck in modern engineering.“

 

The Path Forward 

 

It’s tempting to view verification as a discipline that simply becomes more efficient as AI improves, but the reality is more complex. 

Jagabathula summarized the larger ambition of the industry with this powerful statement: “The big vision is ‘hardware at the speed of software’.”

Staying competitive in the AI era won’t simply require using faster or better design tools. It will demand that engineering organizations build the right teams as chip development accelerates.  

At SBT, we’re seeing that the organizations building the most capable and productive design teams are integrating verification into their talent strategies earlier. We’re also seeing that the most desirable verification roles are going to engineers who have accumulated deep experience, are actively integrating AI into their flows, and are equipping themselves with systems-level exposure.  

 

 

If you’re building this type of team, we’d love to hear from you and learn what’s working. If you’re going to be building this type of team soon, we’d love to talk and share what we’re seeing across the semiconductor industry. 

Contact

"*" indicates required fields

This field is for validation purposes and should be left unchanged.