Neuromorphs!

Some folks are going to work with Rivos to determine if Rivos should invest in neuromorphics or engage with the broader community. Let me (or other folks at Project Phasor) know if you'd like to be part of helping influence that discussion.

I had an offer to join Rivos and the CEO was receptive to a strategy of trying to position Rivos as the best neuromorphic training solution regardless of the user's neuromorphic inference target. However, I am joining NVIDIA and didn't want to let this opportunity go to waste. Especially with Meta's acquisition of Rivos underway and Meta's previous interest in neuromorphics, this is a good time to establish alignment. I think they can fill an important gap in the community.

See email thread below for details and context.

Regards,
Brian

---------- Forwarded message ---------
From: Brian Anderson briandersn@gmail.com
Date: Tue, Oct 21, 2025 at 5:06 PM
Subject: Re: [External] Project Phasor’s future. Also: Rivos, AMD, and Nvidia
To: Blessing Effiong blessing.effiong@slu.edu
Cc: project-phasor-interest@googlegroups.com project-phasor-interest@googlegroups.com

Btw, if anyone would like to take on the responsibility of maintaining the website, LinkedIn page, Youtube channel, email list, and edit permissions for all the associated forms and docs please let me know!

Since there is some interest in continuing with Project Phasor, I encourage folks to self organize and follow up on Blessing's recommendations for next steps. I might be able to serve as an advisor in the future, but will avoid involvement in the short term to give others a chance to step up. In the short term, I will mainly help just by connecting folks.

Especially with Rivos as a potential participant, I think they might appreciate the flexibility to shape things going forward, compared to something that is more established. Especially with something that is more training focused, as Blessing highlighted. At the same time, there are a lot of other efforts like Open Neuromorphic, Edge AI Foundation, and now OCP that will be important to at least collaborate with, if not merge with, farther down the line. OCP in particular has a lot of ties to Meta already, but I haven't kept up with their effort.

Regarding Rivos contacts, these are the folks I would connect people with. I'll introduce relevant folks over email by the end of the week.
Puneet Kumar (CEO) - was supportive of Project Phasor's goals, but is not very familiar with neuromorphics.
David Reveman - A close colleague of mine from our Chrome GPU days. Not very familiar with neuromorphics either, but was on emails where we identified a potential customer of Rivos' that might be interested in event based cameras.
Daniele Castagna - Also a close colleague of mine from Chrome and familiar with Project Phasor.
Peter Tang - I haven't talked to him recently, but he used to work at Intel on neuromorphics and sparse coding and works closely with David and Daniele.
Utku Diril - I haven't talked to him recently, but he is a hardware guy that has shown interest in neuromorphics previously.
I also tried to reach out to folks at Meta, Google, AMD, and NVIDIA and can share contact info with the relevant folks. At Meta, Yann LeCun never responded to my outreach but Garrick Orchard, David Sussillo, Barbara De Salvo, (and Ryad Benosman) might be better contacts. Google has a new Paradigms of Intelligence team that covers bio-inspired AI, but they declined to participate.

Thanks to those who reached out to me through various channels. Everyone's well wishes are very much appreciated!

Best,
Brian

On Fri, Oct 17, 2025 at 10:21 AM Blessing Effiong blessing.effiong@slu.edu wrote:

Brian,

First and foremost, thank you for sharing both the updates and the personal context. And apologies for just responding. I'm deeply sorry to hear about your wife's diagnosis, and I'm glad to hear the doctors caught it early and the initial treatment was successful. Your decision to prioritize stability and family during this time is completely understandable and admirable. Wishing you both all the best as you navigate the road ahead.

Congratulations on your new role at NVIDIA's Advanced AI Systems team. While it's bittersweet to see you step back from Project Phasor, it's exciting that you'll be in a position to influence strategic directions in AI. The neuromorphic community will definitely keep that door open.

On Continuing Project Phasor

I'm committed to helping carry Project Phasor forward. Your work over the past year has built something valuable, and I believe the mission is too important to let it fragment or fade. Here are my thoughts on your discussion points:

Regarding Rivos/Meta: I'm very interested in pursuing the Rivos connection. The timing of their Meta acquisition could actually work in our favor—Meta has resources and a history of engaging with novel compute paradigms. The strategy you outlined (positioning Rivos as a neuromorphic training solution regardless of inference platform) is compelling. I'd be willing to:

Reach out to continue the conversation you started
Leverage the open-source ethos you mentioned to build community engagement
Explore their receptiveness to hardware modeling and architecture search collaboration
Do you have any specific contacts or context you could share that would help warm this introduction?

On Project Phasor's Future: I vote strongly for continuing under the Project Phasor name rather than merging. Here's why:

We've built brand recognition and community identity
The training gap you identified is a distinct focus that doesn't cleanly fit into existing neuromorphic inference organizations
Independence gives us flexibility to work across different hardware platforms
We can still collaborate with existing bodies without losing our identity
Gaps in the Community: Beyond the training gap you've identified, I see opportunities in:

Creating standardized benchmarks that span both training and inference(this is something a group of researchers and I are already working on as a pet project)
Building bridges between neuromorphic research and the broader AI/ML community
Establishing clearer pathways from academic research to commercial deployment(for both hardware and software advancements)
Documentation and education resources for newcomers to neuromorphics
Next Steps: If others are on board, I'd like to propose:

Scheduling a community call in the next 2-3 weeks to align on direction
Setting up a small steering committee to distribute leadership responsibilities
Creating a roadmap for 2025/2026 with concrete milestones
Pursuing the Rivos conversation with your guidance on the best approach
Brian, your leadership and vision have been invaluable. Even as you move into your new role, I hope you'll stay connected as an advisor when bandwidth allows.

Looking forward to hearing from others on the team about how they'd like to move forward.

Regards,
Blessing.

On Oct 16, 2025, at 1:11 PM, Brian Anderson briandersn@gmail.com wrote:

This Message Is From an External Sender
Do not click links or open attachments unless you recognize the sender and know the content is safe. When in doubt, please report the email to SLUAware by clicking on the Report Suspicious button.
Report Suspicious
Lots of updates! Please reply-all to weigh in.

Our AMD invitation was ultimately unsuccessful, but there is reason to believe Rivos might be a better partner for the community.

Also, I will not be working on Project Phasor after Oct 24 (next week) and am interested in how folks think things should evolve. A lot has changed in a few months.

I am joining NVIDIA
Partially due to changes in my personal life, I am deprioritizing neuromorphics. I will be joining NVIDIA's "Advanced AI Systems" team where neuromorphics are currently out of scope -- HOWEVER, it is the team that helps build internal consensus around upcoming shifts in AI to prioritize. So, if you think there are fresh reasons to prioritize neuromorphics, do not hesitate to reach out!

Points of Discussion
Additional context and updates below, but some questions for folks to think about:
Who would be interested in trying to connect Rivos to the neuromorphic community? How?
Are folks interested in carrying on under the Project Phasor name? Or would it make more sense to combine efforts with an existing org to avoid fragmentation?
What gaps are still left unfilled in the community?
Anything else worth discussing?

RIVOS (/META)
Rivos is a potential ally and complement to the neuromorphic ecosystem. They were just acquired by Meta and are one of the few upstarts targeting AI hardware for training, making them a good surrogate for AMD. They don't accelerate sparsity (yet?), but their beefy RISC-V and GPU cores are tightly integrated and have coherent unified access to lots of cache + HBM + DRAM, which could serve as a friendly starting point from a heterogeneous compute perspective. A strategy might be for Rivos to position themselves as the go-to neuromorphic training solution, regardless of the neuromorphic inference platform one might target. I had an offer to join Rivos and the CEO (Puneet) was receptive to that strategy and the goals of Project Phasor, but I ultimately had to decline given all the uncertainty around the acquisition. Puneet and many folks at Rivos used to work on Chrome and the "os" in Rivos is a nod to open source. Before the acquisition, the CEO indicated they would be receptive to cooperative architecture search and hardware modeling -- they are hungry for feedback about what hardware to build. It's unclear how the acquisition will affect things, but now is proabably a good time to influence how neuromorphics might fit into Rivos+Meta's future.

AMD
I tried hard to pitch Project Phasor to AMD, but never got traction. Michaela Blott was the most responsive internal advocate, having worked on sparse networks for Xilinx FPGAs in the past. She indicated explicitly that AMD would not be able to help with Project Phasor's goals. I highly suspect AMD are investing in neuromorphics internally though for a few reasons: 1) their investment in Liquid AI, 2) their acquisition of Untether AI, and 3) Jens Stapelfeldt said "I know the(sic) some people in the CTO team are working on this" in July before ghosting me - despite 8 (EIGHT!) follow up emails. I was also unable to get a warm intro to Lisa Su, but I did try to email her MIT alum email address twice and never got a response.

Personal Stuff
Sorry I have been MIA at times recently. I tried to update the discord periodically, but those messages are easy to miss. Life got very busy and my wife was diagnosed with breast cancer over the summer. Luckily doctors caught it very early and it seems like they were able to remove it all, but it'll remain a concern into the future. As you can imagine, I am trying to minimize sources of uncertainty which is part of the reason why I didn't roll my dice at Rivos/Meta. But maybe that's an opportunity for someone else here to roll the dice with them. I tried to reach out to a number of folks at Meta, but didn't get much feedback. Garrick Orchard says he isn't aware of any current neuromorphic efforts at Meta and Ryad Bensoman indicated he thinks Meta would still be receptive to a neuromorphic effort even though he left his role as director of neuromorphics at Meta over a year ago.

Looking Forward
I really think a company needs to fill in the neuromorphic training gap to round out and counterbalance any kind of neuromorphic working group. One take away from a Jonathan Ross interview recently was how optimizing for efficient inference has required more expensive training runs. I see neuromorphic computing taking that to the next level where there could be a similar mutually beneficial bifurcation. With AMD and NVIDIA out of the picture, I am hopeful that Rivos might step in - but it'll be up to the community to figure that out as I move on to my new role at NVIDIA.

Wishing everyone here the best. Thanks for resonating with me over the past year and I am looking forward to seeing how things evolve and gaps are filled.

Thanks!
Brian

--
You received this message because you are subscribed to the Google Groups "Project Phasor Interest" group.
To unsubscribe from this group and stop receiving emails from it, send an email to project-phasor-interest+unsubscribe@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/project-phasor-interest/CAKNRqPuZ3YfxjEoQe08CFBDOJ-58-Ts%3DsXFTMx2CPw1vN%3DPr3g%40mail.gmail.com.