AI4HW

Machine learning, including large language models, is increasingly being applied across the hardware design flow, from high-level synthesis to physical implementation. Building better models requires diverse training data, reliable feedback from design tools, and the ability to generalize across designs and technologies. This talk argues that open-source EDA provides essential infrastructure for meeting these needs. Open tools, designs, and PDKs can support systematic generation of datasets spanning multiple abstraction levels and enable tool feedback to guide model training and refinement. They also allow researchers to explore how tool versions, configurations, and target technologies affect design outcomes, helping develop models that remain effective beyond a single experimental setup. Reproducible flows provide a common basis for measuring these improvements. The talk will discuss open challenges in data quality, scalability, and transfer to industrial flows, highlighting the role of European open-source EDA initiatives.

Recently, interest in electronic, magnetic and optical materials based on inorganic and organic nano-materials has increased significantly. Such new materials and novel device architectures have the potential to bring the technology beyond the Moore era. The most time-efficient and cost-effective approach to finding the best materials and devices for specific applications is to perform modelling and simulations. In this talk, I will present exciting research carried out by my research group, DeepNano, in the fields of biosensors, molecular electronics, superconducting qubits and the variability of nano-scale transistors to show the importance of material modelling and device simulations not only in academia but also in the industry. Our main aim is to establish a link between the material properties, device architecture and performance using hierarchical, multi-scaled simulation modelling. For example, to evaluate the performance of superconducting qubits, we have developed a simulation flow that links the atomistic simulations with a three-dimensional (3D) numerical device simulator. Also, to simulate the electrical and optical response of biosensors, we have combined analytical and machine learning (ML) methods in a unified simulation framework. In this way, the computation flows and numerical methods developed in my research group can evaluate the material properties and device performance of various heterogeneous devices and systems.

The growing complexity of hardware systems for next-generation wireless communication, sensing, and satellite applications is challenging traditional design methodologies. Artificial intelligence (AI) and machine learning (ML) are emerging as key enablers for accelerating hardware development, reducing design effort, and improving design efficiency. In particular, RF and millimeter-wave circuits pose unique challenges due to their nonlinearities, large design spaces, expensive electromagnetic simulations, and sensitivity to process, voltage, and temperature variations. This talk reviews recent advances in AI/ML-driven design methodologies for RF and mm-Wave circuits and systems, covering device modeling, passive device synthesis, electromagnetic-aware optimization, circuit sizing, layout generation, and system-level co-design. Using RF and mm-Wave design as a representative case study, the talk discusses how AI can be integrated across the hardware design flow and highlights emerging trends toward end-to-end and autonomous design environments, as well as the key challenges and opportunities ahead.