Neuromorphic WG, Mar 5, 2026

Updated Mar 6, 2026 Neuromorphic

Mar 5, 2026

Attended: @petrut-bogdan steve.furber@manchester.ac.uk Pete Bernard EAIF Team Calendar Jonathan Tapson @massimiliano-versace

Attachments EDGE AI Neuromorphic WG EDGE AI Neuromorphic WG kick off meeting – 2025/09/29 08:56 CST – Notes by Gemini 

Summary

Spin Cloud and neuromorphic viability were discussed, with a significant update on the primitive software experience and operational scale of the UT San Antonio system.

Neuromorphic and Spin Cloud Viability
The discussion centered on the viability of Spin Cloud memory technology and the potential for neuromorphic chips to replace Nvidia GPUs for resolving data center power crises. It was noted that Think Cloud is actively manufacturing and selling large machines based on this low-energy AI technology.

UTSA System Software Experience
The UT San Antonio system access was detailed, revealing a system with approximately 365,000 cores and extremely rough, primitive, and unforgiving software. Developing necessary open-service software infrastructure was confirmed to be a massive undertaking, which makes building the hardware system look easy.

New Member and Future Direction
Max Versace, from Analog Devices, joined the group to reconnect with the community and gain access to the latest neuromorphic innovations, viewing the group as a critical antenna. The group decided to organize a proper agenda for future assignments and explore alternative meeting times to accommodate members in Asian time zones.

Details

Decisions

ALIGNED

  • Future Meeting Agenda Planning Process The agenda structure will be organized after Embedded World, communicated in advance, and include input on edge perspectives.
  • Portal Membership List Update Required The group membership list on the portal must be updated to include attending members, pending resolution of user access issues.
  • Alternative Meeting Time Zones Exploration Alternative meeting slots must be identified and communicated via email to accommodate Asian time zones, maintaining flexibility for future scheduling.
  • Next Conference Planning Six-Month Cadence Planning for the next group conference/workshop will target a six-month cadence, including forming a program committee, finding a theme, and opening submissions.
  • Informal San Diego Meeting Planned An informal get-together for attendees at the San Diego meeting in late March is planned.
  • SPAB Meeting One-Slide Summary A one-slide summary of working group activities, upcoming goals, and community asks must be prepared and circulated for input ahead of the SPAB meeting read out.

More details:

  • Discussion on Spin Cloud and Neuromorphic Technology: Steve Furber shared that a lot of the recent focus has been on spin cloud and various forms of memory, acknowledging they have contributed to a paper on spin cloud, a technology they see as entertaining but uncertain regarding its eventual usefulness. They also confirmed that Think Cloud is manufacturing and selling large machines based on this technology, with a focus on low-energy, large AI models. Petrut Bogdan questioned the viability of using neuromorphic chips to replace Nvidia GPUs to solve data center power crises, which Pete Bernard noted as a point of healthy discussion.
  • UTSA Event and Spin Cloud System Access: Pete Bernard provided an overview of the UT San Antonio event, noting it was well-attended and ambitious, focused on providing group access to the neuromorphic system. However, they reported that the system’s software experience was “rough,” “very early and primitive,” and “unforgiving” based on the tutorial for getting workloads onto the system. Steve Furber emphasized that developing the necessary software infrastructure for an open service is a substantial undertaking, noting that their experience suggests it “makes building the hardware system look easy”.
  • Spin Cloud Machine Scale and Workload: When discussing the scale of the Spin Cloud machine at the UT San Antonio event, Pete Bernard described it as having around 48 chips per board and 50 boards, totaling a couple of hundred chips. Steve Furber calculated this setup to have approximately 365,000 cores, which is significantly smaller than the machine at Dresden, which has five and a quarter million cores. The intended application for the UT San Antonio system is primarily targeted toward machine learning (ML).
  • Neurobbench and Future Assignments: Pete Bernard noted that Jason Yik was present at the UTSA event and gave a session on Neurbench, which might help the project move forward. Petrut Bogdan suggested that the ability to run networks on the same hardware could be beneficial for algorithmic benchmarks in Neurobbench, though it is not yet suitable for comparing different systems. Petrut Bogdan committed to organizing a proper agenda after embedded world and communicating it in advance, including getting input from Disha on the edge perspective.
  • Introduction of Leonardo Versace (Max Versace): Leonardo Versace (who clarified their real name is Max) introduced themself as having 30 years of experience in AI, starting in academia and working on AI compute during the DARPA Synapse project. They are currently with Analog Devices (ADI) and will be joining the board of the Age Foundation. They are focused on pushing ADI to adopt ideas in the next wave of innovation, including memory compute and neuromorphic event-based technology.
  • Leonardo Versace’s Goals for the Working Group: Leonardo Versace stated that their primary objective for participating in the working group is to reconnect with the community they had stepped away from while running a software company. A second key goal for them and their team is to gain access to the “latest and greatest innovations” coming out of neuromorphic computing, viewing the group as an antenna to track these developments.
  • Administrative and Membership Updates: Petrut Bogdan reported that they are awaiting a reply from Open Neuromorphic to discuss potential collaboration and integration, noting that this group is looking for direction and funding. Pete Bernard acknowledged a request to add active attendees to the member list on the wiki and confirmed they would fix Petrut Bogdan’s access to allow them to add users. Petrut Bogdan also raised the need to prune the Google calendar invitation list to include only active participants, as the wiki now serves as the primary source for updates for those who only want to be informed.
  • Discussion on Meeting Time Zones: The group discussed the current meeting slot and the need for flexibility to accommodate members in Asian time zones, for whom the current time is very late. Jonathan Tapson confirmed they are willing to have meetings late at night for their time zone to be inclusive of people in Indian time zones. Petrut Bogdan confirmed they would communicate potential alternative slots via email following the meeting.
  • Planning for a Future Conference: The group discussed planning a future conference, potentially with a cadence of roughly six months. Petrut Bogdan suggested planning to form a program committee and inviting submissions for the conference, aiming to find a theme and reviewers.
  • Upcoming Meetings and Events: Jonathan Tapson asked who would be attending the San Diego meeting in the last week of March, which Petrut Bogdan confirmed they would attend. Jonathan Tapson suggested an informal get-together at that event, though they were only 50/50 on attending. Pete Bernard mentioned that a typical working group readout, consisting of one or two slides detailing recent work, goals, and any asks, should be prepared for the SPAB meeting.

Suggested next steps

  • unticked[Petrut Bogdan] Plan Agenda: Organize proper meeting agenda after Embedded World. Communicate new agenda details in advance.
  • unticked[Petrut Bogdan] Invite Speaker: Invite Disha to present expertise on the subject.
  • unticked[Petrut Bogdan] Update Membership: Check and fix user access for adding members to the portal/wiki. Update the membership list with all current attendees.
  • unticked[Petrut Bogdan] Prepare Readout: Create a one-slide working group readout for the SPAB meeting. Circulate slide to the group for input.
  • unticked[Petrut Bogdan] Send Summary: Send follow-up email listing actual action items.