Analog Design Automation: The Final Frontier for Chips
Analog Design Automation: The Final Frontier for Chips

Semiconductor innovation historically followed two parallel tracks: digital design became increasingly automated, while analog design resisted this trend.
Now, modern digital chip development relies heavily on sophisticated EDA tools that automate synthesis, place-and-route, timing closure, verification, and increasingly even floorplanning. But even though the analog/mixed-signal layer is responsible for translating physical signals into usable computation, its design has remained stubbornly manual by comparison.
That’s finally beginning to change.
A new generation of companies is attempting to automate this difficult, intuition-driven aspect of semiconductor engineering using AI-assisted design workflows, reinforcement learning, and software-defined analog development platforms. If they succeed, the implications could extend far beyond engineering productivity.
The shift could alter who gets to build custom silicon in the first place, because once analog design becomes more programmable, the barriers that historically kept custom chip development inside elite semiconductor organizations might collapse.
Why Analog Chip Design is Hard to Automate

The semiconductor industry automated digital design years ago because digital systems are inherently abstractable.
In contrast, analog circuits operate within the messy realities of physics. Since voltage variations, thermal drift, parasitics, noise, signal integrity, electromagnetic coupling, and manufacturing variability all interact, even tiny layout changes can materially impact circuit behavior.
That complexity is why analog engineering has remained one of the most specialized disciplines in semiconductors—and why analog design has resisted “push-button” automation. Elite analog engineers are often valued not simply for technical knowledge, but for their intuition developed over years of iteration-heavy design work.
Researchers have been attempting to solve this problem for years. In a 2020 paper introducing AutoCkt, researchers described analog design as suffering from a “long human-in-the-middle iteration loop that requires expert intuition.” The paper demonstrated that reinforcement learning could optimize analog circuits much more rapidly than traditional approaches.
Google’s DeepMind soon accelerated industry interest in AI-driven chip design. In a 2021 Nature paper, researchers wrote: “In under six hours, our method automatically generates chip floorplans that are superior or comparable to those produced by humans in all key metrics.”
While that work focused primarily on floorplanning rather than analog design specifically, it helped validate an idea now reshaping semiconductor engineering: AI systems can increasingly optimize problems that historically depended on elite human intuition.
But can AI really automate technical expertise out of semiconductor design?
Leaving the Lab: How AI is Changing Semiconductor Design
The focus of market conversation is quickly shifting from whether AI can enhance analog design to how much of analog design can be driven by software and models. The real value of the technology isn’t merely that it speeds analog layout generation, but that it compresses the entire custom silicon development cycle.
Research momentum is accelerating quickly. Multiple recent papers have demonstrated reinforcement-learning approaches capable of achieving near-human or human-level analog optimization performance across various design tasks (see, for example, Domain Knowledge Based Automated Analog Circuit Design with Deep Reinforcement Learning).
But this isn’t just an academic thing. AI-assisted workflows are also being applied to analog layout generation, floorplanning, simulation optimization, yield-aware design, verification acceleration, mixed-signal integration, and iterative tuning reduction.
This is exactly where we see the next wave of Electronic Design Automation (EDA) startups stepping in.
Astrus is one example. They’re applying AlphaGo-inspired reinforcement learning directly to the physics-heavy world of analog layout. By utilizing physics-aware AI models that learn through first-principles simulations rather than copying human examples, Astrus can generate thousands of production-ready analog layouts in a fraction of the time it takes a manual engineering loop.
That’s a profound shift for the semiconductor industry.
Unlike prior automation attempts, these modern AI systems improve as they absorb more design data, making iterations faster and opening the opportunity for architects and engineers to explore designs that previously seemed unrealistic or inaccessible.
That radically changes the economics of semiconductor development.
OEMs Want More Custom Silicon…And Why That Matters to Analog Design

The timing of this shift coincides with exploding demand for custom silicon.
Hyperscalers have already proven the strategic value of vertical silicon integration through products like Google TPU, AWS Trainium and Inferentia, Microsoft Maia, and Meta MTIA. Increasingly, however, the custom silicon movement is spreading well beyond the tech giants.
Across robotics, automotive systems, industrial automation, AI inference, networking, photonics, edge computing, and power management, companies increasingly want hardware optimized for their specific workloads.
Custom silicon is becoming strategically important because general-purpose compute is often too power-hungry or inefficient for emerging AI workloads.
McKinsey recently noted in its AI Infrastructure Analysis that hyperscaler AI infrastructure growth is “accelerating the adoption of more power-efficient hardware,” and breaking this bottleneck requires moving away from legacy, manual development flows.
Celera Semiconductor is a prime example of a company redefining this space with an AI-accelerated silicon development platform.
Their unique approach creates digital twins of analog functions, delivering custom analog ICs and complex power management integrated circuits (PMICs) up to 10x faster than traditional vendors. This automated, AI-searchable design intelligence transforms custom analog from a high-risk gamble into a scalable, accessible competitive advantage.
The main bottleneck that AI automation seems ready to break is this: If analog development cycles shrink dramatically, smaller engineering organizations may suddenly gain access to capabilities that previously required massive dedicated design teams.
Traditional Design Firms May Be Facing a Structural Shift
This transition creates uncomfortable but important questions for fabless semiconductor companies, foundries, and design service companies alike.
Scarce expertise, long iteration cycles, highly specialized workflows, complex tooling, and severe talent bottlenecks have benefited firms specialized in analog designs. But if AI-assisted analog workflows reduce development friction, customers may increasingly choose to build silicon capabilities in-house.
Creative Strategies recently described the current AI-driven semiconductor expansion as a “giga cycle,” noting that custom ASIC development is surging. Broadcom has projected its custom silicon business could exceed $100 billion by the end of the decade.
Obviously, semiconductor design organizations won’t disappear, but where value emerges in the system will shift. That mirrors broader trends already visible across AI infrastructure.
As we discussed in our post on photonic fabrics, the semiconductor industry increasingly rewards engineers who can bridge multiple disciplines rather than operate within isolated silos.
Analog Design Automation Changes the Talent Equation (Nothing Eliminates It)

Importantly, none of this means that the role of analog engineers will be eliminated. In fact, it may open the door to more newcomers—who aren’t just skilled in analog or mixed-signal or digital design, but whose skills span the entire system.
The modern semiconductor engineer needs to understand the interactions between design, packaging, manufacturability, software tooling, AI models, and system-level architecture.
In our practical experience, automation often shifts engineering talent upward into higher-level architectural problems rather than eliminating it. The issue isn’t AI vs humans; it’s how to make it all happen with a lean team of experts who collaborate across disciplines.
The companies that dominate this transition won’t just have the best AI tools. They’ll be positioned to use AI as a wedge to create organizational agility and gain serious competitive leverage.
SBT’s Take: The Next Semiconductor Bottleneck Could Be the Org Chart
If AI-driven analog workflows continue to scale—and they will—then the divide between engineers with skills honed by years of handcrafted analog engineering and those who have quickly mastered the fundamentals and leverage the tools most effectively will all but vanish.
Analog design is one of the last artisanal sectors of hardware engineering. But the future demands lean, agile, multidisciplinary engineering teams that bridge the gap between architectural intent and physical silicon using AI-native platforms.
The only question now is: how is your organization preparing?
If you have questions about what this talent shift looks like and how you can build a team that doesn’t just compete, but leads… let’s talk.
Contact
"*" indicates required fields