What do oxide transistors, ferroelectrics, 2-dimensional channel layers, CFETS, Advanced packaging, AI, and tradewinds have in common? They were all topics at the IEEE VLSI Conference in June.
The VLSI conference is held every June alternating between Hawaii, and Kyoto. In 2024 there were three and a half days of workshops, short courses, papers, panel sessions, and networking opportunities. The conference is focused on both technologies – how the chips go together – and circuits, which can range from how things fit together, to a specific aspect of how the circuit operates. At the 2024 conference, 1203 attended, and another 137 will attend online once the proceedings are released. This was the largest VLSI in history to date.
When I think of the VLSI conference and the IEDM conference, I typically think of IEDM as technology that has been released or will be released in the near future to high volume manufacturing. VLSI is a bit more futuristic; however, it seems that the lines are blurring as several papers on chips and technologies have been released for manufacturing.
AI was a key topic at the conference with an evening panel, keynote speakers addressing AI and their outlook on AI, and multiple circuit and technology papers on AI devices. The evening panel session “Will AI Bite the Industry that Feeds it?” featured panelists from academia to implementation, as well as Google’s Gemini.
Moderated by Masato Motomura, Tokyo Institute of Technology; and Chris Mangelsdorf, Analog Devices (retired), panelists included: Serge Biesemans, IMEC; Chidi Chidambaram, Qualcomm; Chet Lenox, KLA; Euicheol Lim, SK Hynix; Azalia Mirhoseini, Stanford University; Rangharajan Venkatesan, Nvidia; Hoi-Jun Yoo, KAIST.). The panelists prepared some remarks on the title question. They also commented on how AI would impact their job field.
The prepared and unprepared responses made for an amusing evening. Soundbites such as, “it is complicated”. “To a memory architect, AI is a gift from God,” “If you don’t use AI, AI will use you.” – most of the panelists commented that AI is a significant turning point in the industry and potentially world history, comparing it to fire and the wheel. Somehow the printing press was left out.
On how AI would impact jobs or how the panelists would work with AI, most thought there would be a job impact, but there was no consensus. Some thought it would help to streamline tasks, and possibly improve robotics so that robots would take over some of the mundane tasks. There was concern over the amount of power needed to train current generation models and hope that AI might help itself to design better more power-efficient chips and models. All agreed that AI should be used to help improve processes in the industry.
The most interesting part of the evening was when Google Gemini was brought out, and the audience had a chance to ask the analysts and Gemini questions. While the panelists sometimes were a bit shy and indirect with their answers, Gemini was clear concise, and to the point.
Two of the keynotes spent considerable time discussing AI. The first was Dr. Ahmad Bahi the CTO of Texas Instruments who spoke on Making Sense at the Edge. Dr. Bahi discussed how a significant amount of AI will be at the edge, and what chips and technology will be needed to support the infrastructure. The key to success will be having the sensing, accelerator, memory, and logic in the same package. The building blocks needed to fit into an advanced package are shown in Figure 2.
The edge designs will be built such that most of the inference can be performed at the edge, and hopefully the software package will be small enough that learning can take place at the edge. Advanced packaging will play a key role in intelligence at the edge due to the components needed for the system. The systems also need to know what data needs to be sent to the cloud and when to send it to the cloud. Most of these ideas have been developing since the IoT was launched, as sensing and processing at the edge is a key component of the IoT.
Kicking off the Wednesday sessions, Maryam Rofougaran CEO and co-founder of Movandi Corporation highlighted the Wednesday Keynote with a discussion on Wireless connectivity in a future hyperconnected world. The audience was led through the history of wireless development leading us to the 5G/6G and the WiFi7 evolution. Ms. Rofougaran commented that the path to future wireless devices will be by disintegrating the SoC and creating multi-chip modules in a 3D package. This will create the speed needed for next-generation wireless, as well as help improve the performance of the radios.
In closing, Ms. Rofougaran left the audience with the thought that data is the oil, 5G is the pipeline, and AI is the refinery. An interesting way to think about the information super highway as the AI journey moves forward.
