DAC In Review

DAC 2025 made one thing clear: AI is no longer a side conversation in semiconductor design, it’s the main event. This year, the shift wasn’t just about speed but reasoning. 

Everywhere we looked, there was evidence that intelligent systems are beginning to do more than assist. They’re starting to participate by asking better questions, making decisions, and even engineering physical outcomes from a string of human-language prompts.  

What stood out most at this year's conference wasn’t just the pace of AI advancement, it was the way it’s beginning to reframe design thinking, enabling new pathways from idea to implementation, and what it means to engineer alongside intelligent systems. 

Engineering Through Language and Logic 

One of the most exciting themes this year was how quickly we’re moving from intent to silicon. There was pertinent focus on the emerging flow: Intent → Spec → RTL → Layout. It’s now possible to input a high-level design intent in natural language and guide that through a series of AI agents, all the way to working RTL and even layout generation. 

At UCSD, the MAGE system showed how powerful this is already—achieving a 95–96% boost in RTL quality compared to generic LLMs. That  means fewer downstream bugs, shorter verification cycles, and faster paths to tape-out. 

And verification itself is evolving, too. We saw explainable AI frameworks that don’t just generate results, they expose why decisions are made. This kind of introspection is essential when engineers are collaborating with autonomous systems and need traceable, human-readable logic behind each outcome. 

Physical Innovation, Powered by Agents 

Another thread that kept resurfacing was that AI is extending beyond simulation and into physical innovation. From battery development to vaccine design, multi-agent systems are forming virtual labs that decompose complex scientific problems, assign specialized tasks, and carry out research loops in parallel. 

In his keynote, William Chappel described how Microsoft’s agents processed 30 million battery chemistries to identify 800 viable materials—with 70% less lithium—before a single sample was made. Another example focused on how a network of AI “scientists” developed eight vaccine candidates in just four hours, with each agent specializing in steps like protein modeling, efficacy analysis, and literature review. 

These were a glimpse of what’s possible when intelligent agents are given real-world tools, context, and autonomy with humans still in the loop to steer and validate. 

What We’re Building Toward 

While most current solutions are still scoped to IP blocks or isolated tools, we’re thinking ahead. Shashank puts it best: We’re entering a world where hybrid teams of AI and humans will not only build chips faster but design systems that weren’t even feasible before. 

For us, that means moving from IP-level verification into full SoC design, integration, and firmware generation. It means adding explainability as a baseline feature. And it means giving engineers tools that learn with them. 

The MooresLabAI Lens 

Everything we heard this week resonated with our journey. We're focused on silicon design, not vaccines or batteries, but we’re solving a similar problem: how do you bring reasoning into engineering workflows without overwhelming the people who use them? 

What we saw at DAC affirmed that our mindset of building intelligent, explainable, and resilient systems is where the entire industry is heading. 

Final Thoughts 

We left DAC energized. The energy on the floor, the curiosity in the sessions, and the conversations we had with so many of you reinforced why we started MooresLabAI. This industry needs tools that make a real difference in how chips get built. 

If you came by our booth or reached out at any point, thank you. We’re grateful for the community that’s forming around this space, and excited for what’s ahead.