Applying machine learning to chip design and manufacturing: interview with Lorenzo Servadei

Lorenzo Servadei and his team at Sony AI are focused on researching and developing machine learning models to aid chip design and manufacturing. In this interview, Lorenzo tells us more about Electronic Design Automation, and how machine learning has been added into the mix to further advance the field of semiconductor chip design.

What was your inspiration for pursuing a career in AI and semiconductors?

When I was pursuing my Master’s degree, I studied subjects related to traditional computer science and algorithmics – before AI was seen as a specific area of study – which led me into the field of software development. While working in software development, I had the opportunity to join a semiconductor company that was seeking AI experts, which allowed me to explore the algorithmic aspects of AI. I’ve always been fascinated by the idea of cross-level optimization, where improvements at one layer of the tech stack can drive progress at another. What drew me further into this field was the opportunity to connect knowledge from different communities – AI researchers, semiconductor engineers, and Electronic Design Automation (EDA) specialists – and seeing how much value can be created when these different disciplines interact with one another.

Applying AI in engineering is particularly challenging because it requires raw machine learning capability, deep domain expertise, trustworthiness, and the ability to handle highly constrained environments. This makes the intersection between AI and semiconductor design both demanding and rewarding.

Today, with designs and technologies becoming increasingly complex, design automation is taking on an even more central role. The ability to leverage AI for design automation is opening up new possibilities, helping engineers manage complexity, accelerate development, and push the boundaries of what’s possible in semiconductor innovation.

I believe that both AI and semiconductors have been, and will continue to be, key drivers of the global economy, so working at their intersection feels like a meaningful way to contribute to the future.

How is Electronic Design Automation pushing semiconductors to new levels of efficiency and innovation?

Electronic Design Automation (EDA) has been a critical enabler of semiconductor progress for decades, constantly pushing the boundaries of what’s possible. Today, we’re seeing another transformation, fueled by advances in computational power and the integration of AI.

The first wave of AI in EDA, what you could call EDA 1.0, focused on using machine learning for fast estimations and surrogate models. These methods provided valuable accelerations in areas like performance prediction, design space exploration, and yield estimation. They didn’t fundamentally change the way design was done, but they offered powerful shortcuts that saved time and resources.

In 2025, we’re entering EDA 2.0. Here, AI isn’t just a tool for estimation; it’s beginning to shape scripting, design functionalities, and even the logic that drives automation flows. Instead of simply assisting engineers with predictions, AI is starting to participate more actively in the design process itself, enabling new levels of adaptability, creativity, and efficiency in handling even more complex technologies.

To realize this potential, we still need careful research that allows us to understand when and how these new methods genuinely improve upon established approaches, while not losing sight of the fundamentals that have made EDA such a strong discipline already. It’s the balance between leveraging breakthrough technologies like AI and staying grounded in proven engineering principles that will continue to push semiconductors to new levels of efficiency and innovation.

How does your research aim to advance machine learning and expand what’s possible for chip designers?

I believe there are three interesting directions in AI for chip design. First, using neural networks to accelerate multi-physics models can help designers explore more complex systems with faster turnaround. Second, developing and selecting the best optimization algorithms allows designers to tackle design trade-offs more efficiently and reliably. And third, generative AI for physical implementation opens up entirely new ways of approaching chip layout and design space exploration. Together, these directions could significantly expand what designers can achieve in terms of speed, quality, and creativity. Our research, at a high level, explores large language models (LLMs) and physics-informed neural networks to generate, optimize, and improve state-of-the-art image sensor electronics.

How do you think AI will contribute to chip design in the future?

If researchers can demonstrate clear benefits, AI has the potential to fundamentally reshape chip design processes. It could accelerate tasks that are traditionally very time-consuming, making them much faster, while also unlocking new solutions that weren’t previously feasible. Over time, this could not only improve efficiency, but also change the way designers think about solving problems.

One of the most exciting aspects about AI in chip design is that AI opens the possibility of addressing multiple objectives and multiple design stages within a single system. Instead of design operations following a strict waterfall pattern, moving step by step from one stage to the next, they can evolve into a more holistic, interconnected process. AI can help capture interactions across abstraction levels, enabling optimization that simultaneously considers performance, power, area, manufacturability, and reliability.

This shift from sequential workflows to integrated, holistic design operations could be transformative, redefining how future chips are conceived and built.

How do you see AI helping semiconductor companies maintain a competitive advantage in such a rapidly evolving industry?

There is still much to explore regarding how AI can impact the semiconductor industry. It is an incredibly competitive landscape, so speed and efficiency are crucial. AI can help organizations accelerate internal processes and reduce execution times, which translates directly into faster time-to-market. At the same time, it can improve product quality by enabling more accurate modeling and better optimization. And perhaps most importantly, AI has the potential to fuel innovation by generating creative ideas that keep companies ahead of the curve.

About Lorenzo

Lorenzo Servadei is the Head of AI for Chip Design at Sony AI, leading research on applying Machine Learning to chip design and manufacturing. Previously, he was Head of Machine Learning for Sensors at Infineon Technologies AG, where he earned his Ph.D. in Computer Science with Johannes Kepler University Linz, focusing on AI for hardware–software co-design. His work has appeared in top venues such as DAC, DATE, TCAD, IROS, and ICASSP. He also lectures and leads a research group in Machine Learning for Design Automation at the Technical University of Munich, where he obtained his Habilitation.

Find out more

These papers were recently presented at MLCAD, the 7th ACM/IEEE International Symposium on Machine Learning for CAD:

• GENIE-ASI: Generative Instruction and Executable Code for Analog Subcircuit Identification, Phuoc Pham, Arun Venkitaraman, Chia-Yu Hsieh, Andrea Bonetti, Stefan Uhlich, Markus Leibl, Simon Hofmann, Eisaku Ohbuchi, Lorenzo Servadei, Ulf Schlichtmann, Robert Wille
• Schemato – An LLM for Netlist-to-Schematic Conversion, Ryoga Matsuo, Stefan Uhlich, Arun Venkitaraman, Andrea Bonetti, Chia-Yu Hsieh, Ali Momeni, Lukas Mauch, Augusto Capone, Eisaku Ohbuchi, Lorenzo Servadei

AIhub is dedicated to free high-quality information about AI.

AIhub is dedicated to free high-quality information about AI.