I’ve just returned from a very busy week in California visiting Intel and AMD in Folsom before chairing the SpatialML annual meeting in Monterey and then attending the first in-person ACM FPGA conference, also in Monterey. It was a great trip, and I think the theme for me was just how well the tech industry and academia can work together, building on each other’s strengths.
I was delighted to have the first chance to pay an in-person visit to the team built by my former PhD student Theo Drane at Intel in Folsom, where he has been able to gather an exceptionally strong and diverse mix of skills. Intel is sponsoring my PhD student Sam Coward (far left of photo) who also a member of staff in the group. Together we have been working on some really exciting new ways to optimise datapath designs using word-level rewrite rules. While on the trip, we learnt that Sam’s DAC paper has been accepted, so there will be more I will want to say publicly about the development of this work soon!
Also in Folsom, I was able to meet members of the graphics team at AMD, who have recently agreed to support two new PhD students in my lab working on the interaction between microarchitecture and machine learning. I expect them to start in October 2023, with one co-supervised by Deniz Gündüz at Imperial and one co-supervised by Wayne Luk, also at Imperial. We’re planning exciting work, so watch this space.
In 2019 I was awarded a grant by EPSRC to bring together Imperial and Southampton (lead there Geoff Merrett) in the UK with Toronto (Anderson, Betz, Chow) and UCLA (Cong, Ercegovac) and industry (Imagination, Xilinx, Maxeler, NVIDIA, Google, Intel, Arm & Corerain) to reimagine the full stack of machine learning hardware when computing across space, not just time. You can visit the centre website here and see an up-to-date feed of our work on Twitter. Due to COVID, the last time we were able to hold an all-hands meeting was just after kick off, in February 2020, so it was truly a delight to be able to do so again this year. In the meantime, we have been very active and also expanded the centre to include Cornell & Cornell Tech (Abdelfattah, Zhang) as well as Sydney (Boland, Leong) and Samsung AI, also since then Maxeler has been acquired by Groq and Xilinx by AMD. Whereas in 2020, I had primarily focused the workshop on faculty and industry talks, this year I re-focused on hearing the progress that various PhD researchers had made since 2020 as well as very successful 2min lightning talks from all participants to aid networking and build a solid foundation for our researcher exchange programme. In addition, the new participants were given the chance to highlight their work in extended talks.
We had the following speakers:
- Zhiru Zhang, Cornell – new faculty affiliate to the center
- Philip Leong, Sydney – new faculty affiliate to the center
- Sam Bayliss, AMD – industrial partner of the center
- Mohamed Abdelfattah, Cornell Tech – new faculty affiliate to the center
- Aaron Zhao, Imperial – new faculty affiliate to the center
- Jason Anderson, Toronto – affiliated faculty
- Marta Andronic, Imperial – affiliated PhD student
- Vaughn Betz, Toronto – affiliated faculty
- Ben Biggs, Imperial – affiliated PhD student
- Christos Bouganis, Imperial – affiliated faculty
- Yuzong Chen, Cornell Tech – affiliated PhD student
- Jason Cong, UCLA – affiliated faculty
- George Constantinides, Imperial – affiliated faculty
- Sam Coward, Imperial – affiliated PhD student
- Jordan Dotzel, Cornell – affiliated PhD student
- Theo Drane, Intel – industrial partner of the center
- Milos Ercegovac, UCLA – affiliated faculty
- Rajarshi Roy, NVIDIA – industrial partner
- Jon Hare, Southampton – affiliated faculty
- Julie Hsiao, Toronto – affiliated PhD student
- Mike Hutton, Google – industrial partner
- Wayne Luk, Imperial – affiliated faculty
- Geoff Merrett, Southampton – affiliated faculty
- Alex Montgomerie, Imperial – affiliated PhD student
- Zhewen Yu, Imperial – affiliated PhD student
- Paul Chow, Toronto – affiliated faculty
- Stelios Venieris – industrial partner of the center
- Atefeh Sohrabizadeh, UCLA – affiliated PhD student
- Camilo Vega, Toronto – affiliated PhD student
- Anastasios Dimitriou, Southampton – affiliated PhD student
- Andrew Boutros, Toronto – affiliated PhD student
At the meeting dinner, we were also able to celebrate the announcement that our member, Jason Cong, has just won the EDAA Achievement Award, to be presented at DATE 2023. We were able to end the workshop with a productive session, planning collaboration and researcher exchange between our sites over the coming year.
The ACM International Symposium on FPGAs (known as ‘FPGA’) is one of the key dates in my calendar. As a former chair of the conference, I sit on its steering committee, and I have always been a great supporter of the genuine community around this conference, combining a strong academic and industrial presence. It was wonderful to be back in person after two years of virtual FPGAs.
This year two of my collaborators were chairing: Zhiru Zhang (Cornell) was Program Chair and Paolo Ienne (EPFL) was General Chair. They put together a great event – you can read the program here. There were many highlights, but I would particularly mention the paper by my collaborator Lana Josipović and her PhD student Jiahui Xu and collaborators on the use of model checking to remove overhead in dynamically-scheduled HLS, closely related to my student Jianyi’s stream of work which I’ve described in other blog posts. I also had the privilege to serve on the best paper award panel this year, alongside Deming Chen and Viktor Prasanna, an award that went to some work from Jane Li‘s group for their development of a plug-and-play approach to integrating SSD storage into high-level synthesis, presented by the lead author Linus Wong, during the session I chaired on “High-Level Abstraction and Tools”.
It was a personal delight to see so many of my former PhD students, and the definite highlight was getting to meet the first child of one of my former students, born over the period I’ve been out of the loop due to COVID. My former student Sam Bayliss was in attendance and presenting a tutorial on programming the AMD AI Engines developed at using MLIR. My former student Kan Shi (Chinese Academy of Sciences) was also presenting his work together with my other former student David Boland (Sydney) and my current undergrad student Yuhan Diao (Imperial College) on on-chip debugging. Kan is also a part-time YouTuber, and his recent YouTube interview with me was mentioned to me by several attendees – I suspect it has had far wider reach than my publications!
Overall, I have come away from the trip energised from the technical conversations, with far too many ideas for collaboration than I’ll be able to follow up on, and with an even further strengthened belief in the immense promise of industrial / academic collaboration in my field.