Keynote Speakers

Abstract
Recent years have brought a flood of proposed applications for AI and machine learning in EDA. The potential benefits of AI for EDA, including improved design quality with reduced cost and schedule, have attracted significant efforts and focus across both academia and industry. But we have not seen truly disruptive innovation in this space – yet. This talk will examine what disruptive innovations at the AI-EDA nexus – and for hardware design as well – might look like. First, what are root causes for the absence of visible disruptions from AI-EDA innovations? These include technical and ecosystem challenges, the time and patience needed to reach tipping points, and the presence of active blockers to innovation. Second, where might advances in AI for EDA bring truly transformative, disruptive innovations to benefit the semiconductor and design ecosystem? Here, possibilities span analysis and simulation tooling, the nature of design signoff, the EDA-designer-foundry dynamic, and more. There are also basic categories of disruptive innovations: for the EDA business model; for the worldwide ecosystem of academic researchers, EDA vendors, and designers; and for EDA technology itself. Third, and finally, what will accelerate inevitable disruptive innovations, who might deliver them, and how?

[Related talks and papers can be found at https://vlsicad.ucsd.edu]

Abstract
The era of pervasive intelligence is revolutionizing not only how we innovate, but also the speed at which we innovate. AI is impacting nearly every industry from data centers, mobile, automotive to aerospace and government. The insatiable compute demands coupled with the staggering explosion of intelligent systems is forcing a reinvention of computing and chip design as we know it. This keynote will address how the demand for higher computational power, performance, and efficiency in AI chips is driving the demand for multi-die systems and how this dependency will be a game-changer in pushing the boundaries of innovation.

Abstract
The advancements in AI technology, represented by generative AI, significantly transform our society. On the other hand, the energy consumed by AI technology continues to increase. This increase in society’s overall energy consumption is a critical issue that all researchers involved in information technology must seriously consider. Cyber-physical systems (CPS) provide new value by analyzing information obtained from the physical world in the cyber world and feeding back the results to the physical world. In this talk, CPS for energy management in buildings is introduced as an example of CPS design aimed at solving energy problems. Specifically, our initiatives related to CPS are introduced for managing the energy of the HVAC (Heating, Ventilation, and Air Conditioning), which is said to account for about 40% of the energy consumption in buildings, from design to implementation and on-site experiments.

Abstract
In the AI era, the LLM model size of Generative AI is increasing exponentially, getting closer to human ability, and accordingly, the demand for server’s computing power is also rapidly increasing. However, in preparation for the explosion of demand for AI applications, it is predicted that the server system alone may lack response capabilities, so it is necessary to allocate roles for the AI inference operation of the client system. For this reason, interest in on-device AI is growing, and hardware and software research are actively being conducted to cover “AI inference” by PCs and mobile systems. In this presentation, I would like to introduce the advantages, disadvantages, and development directions of wide IO type high bandwidth memory, processing in memory, and Compute in memory, which are on device-oriented memory solutions.