BRN Discussion Ongoing

[Frangipani]

(…) As for neuromorphic research, we know that AFRL has been collaborating with both IBM and Intel for years.

While the AFRL Facilities Book FY25 doesn’t specify whether or not the “3UVPX heterogeneous computing” in their equipment list includes a 3U VPX board with a neuromorphic processor, it could theoretically even refer to IBM’s NorthPole in a 3U VPX form factor (aka NP-VPX):

https://modha.org/wp-content/uploads/2024/09/NorthPole_HPEC_VPX.pdf

View attachment 88776
(…)

View attachment 88777
The following October 2023 exchange on LinkedIn between IBM’s Dharmendra Modha and AFRL Principal Computer Scientist Mark Barnell (who is also named as the Embedded Edge Computing Lab’s Primary POC on page 66 of the AFRL Facilities Book FY25) on the release of NorthPole is evidence that the collaboration between AFRL and IBM did not end with the TrueNorth era.

NorthPole: Neural inference at the frontier of energy, space, and time | Dharmendra Modha | 115 comments

New in Science Magazine, a major new article from IBM Research, IBM introducing a new brain-inspired, silicon-optimized chip architecture suitable for neural inference. The chip, NorthPole, is the result of nearly two decades of work by scientists at IBM Research and has been an outgrowth of a...

www.linkedin.com

View attachment 88778

And yet, less than a week earlier, one of the co-authors, Tarek Taha, will be presenting a differently titled paper by the same group of University of Dayton researchers (plus Simon Khan from AFRL), in which they state that a comparative analysis of their “low-power anomaly detection system for Controller Area Network (CAN)” on Akida and Loihi showed that “Loihi consumes about 91 times less power than Akida”.

https://thestockexchange.com.au/threads/brn-discussion-ongoing.1/post-457988

View attachment 89049


By the way, those University of Dayton researchers are long-time collaborators and co-authors of Steven Harbour from Parallax Advanced Research (headquartered in the same city, namely Dayton, OH), dating back to the time when he was still working at Southwest Research Institute.

Today’s post by Lauren Reinerman-Jones from SwRI (Southwest Research Institute) is further evidence that the US Air Force - while currently evaluating Akida - continues to be interested in Loihi:

It has been great sharing the excitement of our work with Air Force personnel in our #SwRI Dayton office. | Lauren Reinerman-Jones

It has been great sharing the excitement of our work with Air Force personnel in our #SwRI Dayton office. One of our partner’s at #Intel came to support #neuromorphic #AI demos to folks from #AFMC711HPW, #AFRLRYZA, #AFRLRST, and #AFLCMCHBY. Thank you to each of the Air Force personnel that took...

www.linkedin.com

https://www.linkedin.com/in/lauren-reinerman-jones-081496/

 

[Jacob Kiraz]

What ever happened to the Nitendo switch 2?

 

[supersonic001]

Our latest blog explores how Akida’s neuromorphic processor takes advantage of activation sparsity in neural networks to deliver ultra-low-power, high-performance AI for applications where power… | BrainChip

Our latest blog explores how Akida’s neuromorphic processor takes advantage of activation sparsity in neural networks to deliver ultra-low-power, high-performance AI for applications where power, latency, and bandwidth are limited. Unlike conventional accelerators, Akida skips unnecessary...

www.linkedin.com

 

[itsol4605]

Wow! That's interesting!!

China decided yesterday in reconsidering Al investment strategy.

99% Al companies do not make money so far.

99% Al revenues go to NVIDIA itself, across the globe.

The Al risk is escalating tremendously.
— — —
A good time to switch to more modern technology!

https://www.linkedin.com/feed/update/urn:li:groupPost:7036558-7364052669256097792

 

[7für7]

Our latest blog explores how Akida’s neuromorphic processor takes advantage of activation sparsity in neural networks to deliver ultra-low-power, high-performance AI for applications where power… | BrainChip

Our latest blog explores how Akida’s neuromorphic processor takes advantage of activation sparsity in neural networks to deliver ultra-low-power, high-performance AI for applications where power, latency, and bandwidth are limited. Unlike conventional accelerators, Akida skips unnecessary...

www.linkedin.com

View attachment 89925

Can I ask you something?

 

[Frangipani]

Yes, that statement actually comes from a paper presented at the IEEE ISVLSI conference in July 2023. The paper is titled "Comparative Analysis of Low-Power Anomaly Detection on Neuromorphic Chips for Controller Area Network (CAN) Data", and was co-authored by Tarek Taha (University of Dayton) and Simon Khan (AFRL), among others.
In the study, they implemented an anomaly detection model on both Loihi and Akida chips, using the same CAN dataset. The results showed that Loihi consumed around 91 times less power than Akida for this specific task.
However, it's important to note that this is not a general benchmark. The test was very specific to a certain use-case in vehicular networks, and the result depends heavily on the software setup, optimization level, and even how the power consumption was meassured. Also, Loihi is more of a research platform, while Akida targets edge applications with different design priorities.
So the result is interesting, but should'nt be overgeneralized.

Hi @Baneino,

could you please share with us where you found evidence of that July 2023 IEEE ISVLSI conference paper titled “Comparative Analysis of Low-Power Anomaly Detection on Neuromorphic Chips for Controller Area Network (CAN) Data”, which you referred to in your above post?

Neither is it included in Tarek Taha’s extensive list of publications (which, while not quite up to date, should be exhaustive with respect to those published in 2023) nor could I find any trace of it when googling the paper’s title as such (except for your 28 July TSE post, that is).

The only conference paper listed on Tarek Taha’s website, which appears to be relevant to the topic, happens to be the 2024 paper “Unsupervised Anomaly Detection for Automotive CAN Bus on the Intel Loihi”:

Publications

This is also the only paper on this topic I was able to find under Simon Khan’s Google Scholar profile, who - according to you - is one of the other co-authors.

Said 2024 conference paper was not a comparative analysis, though, plus the authors stated in their abstract that “To the best of our knowledge, this is the first low-power, unsupervised anomaly detection system using the Loihi or any other neuromorphic processor”, which obviously wouldn’t make sense if they had already published an earlier paper comparing Loihi and Akida.
(https://ieeexplore.ieee.org/document/10651224/)

Then there is also the 6 August 2025 conference paper titled “Anomaly Detection of CAN bus messages on neuromorphic hardware” that I referred to in April, which equally appears to contradict the existence of a prior comparative paper by the same authors (cf. section underlined in red):

It appears not everyone at AFRL resp. collaborating with AFRL would agree that Akida is always the superior choice:

Anomaly detection of CAN bus messages on neuromorphic hardware | SPIE Optics + Photonics

View presentations details for Anomaly detection of CAN bus messages on neuromorphic hardware at SPIE Optics + Photonics

spie.org

View attachment 82236


View attachment 82235
View attachment 82234

Tarek Taha

View attachment 82233

Is that mysteriously elusive 2023 conference paper you claimed existed by any chance a GenAI hallucination?

 

[Bravo]

An XPU, or "X Processing Unit," is a broad term encompassing various specialized processors, including CPUs, GPUs, DPUs, and NPUs, designed for high-performance computing and AI/ML applications

What SoftBank, Government Investments Could Mean for Intel's Business—and Stock

Intel shares jumped on Tuesday as investors responded to signals the Trump administration may throw the company’s troubled foundry business a lifeline.

www.investopedia.com

EXTRACT ONLY

 

[FJ-215]

The share price is currently at 19.25c, but there are 5x 1 volume buy-orders sitting at 19c.

Why on Earth would you be willing to wait for 19c for 1 unit, doesn't make sense, haha!

My guess..

They be shorters..

 

[Diogenese]

An XPU, or "X Processing Unit," is a broad term encompassing various specialized processors, including CPUs, GPUs, DPUs, and NPUs, designed for high-performance computing and AI/ML applications

What SoftBank, Government Investments Could Mean for Intel's Business—and Stock

Intel shares jumped on Tuesday as investors responded to signals the Trump administration may throw the company’s troubled foundry business a lifeline.

www.investopedia.com

EXTRACT ONLY


View attachment 89928

White knights or circling vultures?

https://www.globaltimes.cn/page/202508/1341346.shtml

US govt reportedly to take 10% stake in Intel
Move could erode trust in chip giant, change relationship between govt and businesses: experts
By GT staff reportersPublished: Aug 20, 2025 10:27 PM

• 
  
• 
  
• 
  
• 
  
• 

 

[7für7]

Is it today the lowest volume I ever saw on brainchip ? Or is it the lowest volume I ever saw on Brainchip today?

 

[smoothsailing18]

Prophesee

 

[Bravo]

White knights or pack of vultures?

https://www.globaltimes.cn/page/202508/1341346.shtml

US govt reportedly to take 10% stake in Intel
Move could erode trust in chip giant, change relationship between govt and businesses: experts
By GT staff reportersPublished: Aug 20, 2025 10:27 PM

• 
  
• 
  
• 
  
• 
  
• 

HI @Diogenese,

Do you think BrainChip, Arm and IFS could be collaborating to create an Akida 3 chiplet on Intel’s 18A?

If Akida 3 is delivered as a standards-compliant chiplet, you would think it would dramatically broaden adoption, making it much easier for any multi-die SoC/SiP program - not just Arm’s - to plug Akida in.

This could also open the door to other players such as NVIDIA, AMD, Qualcomm, MediaTek, SoftBank (e.g. Stargate), Tier-1 automotive suppliers, and defense primes.

From a strategic standpoint, it seems both technically feasible and mutually beneficial:

• BrainChip would gain wider adoption by positioning Akida 3 as a “drop-in” neuromorphic accelerator across multiple ecosystems.
• Intel would expand its foundry business by offering differentiated AI chiplets on 18A, attracting new customers (including the US govt, defense primes, etc.) and reinforcing its chiplet marketplace ambitions.
• Arm would strengthen its SoC value proposition by pairing Arm CPUs with complementary AI chiplets (like Akida), giving customers more modular design options.

Do you think Akida 3 could support UCIe or Arm’s Chiplet System Architecture (CSA)?

 

[Baneino]

Hi @Baneino,

could you please share with us where you found evidence of that July 2023 IEEE ISVLSI conference paper titled “Comparative Analysis of Low-Power Anomaly Detection on Neuromorphic Chips for Controller Area Network (CAN) Data”, which you referred to in your above post?

Neither is it included in Tarek Taha’s extensive list of publications (which, while not quite up to date, should be exhaustive with respect to those published in 2023) nor could I find any trace of it when googling the paper’s title as such (except for your 28 July TSE post, that is).

The only conference paper listed on Tarek Taha’s website, which appears to be relevant to the topic, happens to be the 2024 paper “Unsupervised Anomaly Detection for Automotive CAN Bus on the Intel Loihi”:

Publications

View attachment 89924

This is also the only paper on this topic I was able to find under Simon Khan’s Google Scholar profile, who - according to you - is one of the other co-authors.

Said 2024 conference paper was not a comparative analysis, though, plus the authors stated in their abstract that “To the best of our knowledge, this is the first low-power, unsupervised anomaly detection system using the Loihi or any other neuromorphic processor”, which obviously wouldn’t make sense if they had already published an earlier paper comparing Loihi and Akida.
(https://ieeexplore.ieee.org/document/10651224/)

Then there is also the 6 August 2025 conference paper titled “Anomaly Detection of CAN bus messages on neuromorphic hardware” that I referred to in April, which equally appears to contradict the existence of a prior comparative paper by the same authors (cf. section underlined in red):




View attachment 89927

Is that mysteriously elusive 2023 conference paper you claimed existed by any chance a GenAI hallucination?

Thank you very much for pointing this out and for your thorough check.
I went back to verify the source again, and I have to admit that the IEEE ISVLSI 2023 paper I mentioned (“Comparative Analysis of Low-Power Anomaly Detection on Neuromorphic Chips for Controller Area Network (CAN) Data”) cannot be found in any official records (IEEE Xplore, Tarek Taha’s publication list, Google Scholar). The only trace I could find was in a discussion forum, and I probably relied on that too quickly, which may have given the impression that this was a properly published paper.

If I made an error here or expressed it in a misleading way, I would like to aknowledge and correct that: as far as I can see now, the only verifieable publication in this domain is the 2024 paper (“Unsupervised Anomaly Detection for Automotive CAN Bus on the Intel Loihi”).

I really appreciate you raising this, since it helps keeping the discussion factually correct.

 

[Diogenese]

HI @Diogenese,

Do you think BrainChip, Arm and IFS could be collaborating to create an Akida 3 chiplet on Intel’s 18A?

If Akida 3 is delivered as a standards-compliant chiplet, you would think it would dramatically broaden adoption, making it much easier for any multi-die SoC/SiP program - not just Arm’s - to plug Akida in.

This could also open the door to other players such as NVIDIA, AMD, Qualcomm, MediaTek, SoftBank (e.g. Stargate), Tier-1 automotive suppliers, and defense primes.

From a strategic standpoint, it seems both technically feasible and mutually beneficial:

• BrainChip would gain wider adoption by positioning Akida 3 as a “drop-in” neuromorphic accelerator across multiple ecosystems.
• Intel would expand its foundry business by offering differentiated AI chiplets on 18A, attracting new customers (including the US govt, defense primes, etc.) and reinforcing its chiplet marketplace ambitions.
• Arm would strengthen its SoC value proposition by pairing Arm CPUs with complementary AI chiplets (like Akida), giving customers more modular design options.

Do you think Akida 3 could support UCIe or Arm’s Chiplet System Architecture (CSA)?

Hi Bravo,

Last time I looked, Intel were having problems getting 18A to work properly - very low yield per wafer.

In olden times, Intel would have been able to throw fistfulls (fistsfull) of cash at the problem but now they are strapped for cash, and the sharks are circling. Clearly there has been a backdoor deal with US gvt - Intel is too big to fail. Under the Chips Act, the gov would want IFS to succeed so maybe that's where their investment has gone. Of course, Intel also needs help with AI.

Nvidia was blocked from buying ARM, so this may prevent any merger, but collaboration is possible, it's just that there could be lots of fingers in the pie. Maybe IFS will be hived off - it's already notionally customer-agnostic. This would leave Intel pretty much as it was before IFS in 2024.

As for Akida 3, I see this as being the Rolls-Royce option (aerospace, defence, ...) because of its INT16/FP32 capability. This will take up a lot more wafer real estate per chip, making it much more expensive to make, say >3X. I think Akida 3 will be a specialist high precision option, while Akida 2 will be the "standard" workhorse.

I don't know if 18A is suitable for RadHard.

The advantage of chiplets is that it would make Akida available for a much larger range of applications without needing specialized IP SoC designs/wafer tapeout/production for each customer.

 

[TECH]

Hi @TECH,

sorry to disillusion you, but this is not a new Accenture patent!

@Fullmoonfever had already posted about it more than a year ago.
This patent titled “Self-learning Neuromorphic Gesture Recognition Models” was filed on 22 November 2022 and first published on 23 May 2024:

https://thestockexchange.com.au/threads/brn-discussion-ongoing.1/post-432356

View attachment 89907
View attachment 89904


I also recalled referring to it in a November 2024 post:

https://thestockexchange.com.au/threads/brn-discussion-ongoing.1/post-442579

View attachment 89905




The only novel thing about the patent is that it has now finally been granted, after initially getting rejected back in February (cf. the meaning of B2 after the patent number as well as below patent history):

View attachment 89906

https://register.epo.org/help?lng=en&topic=kindcodes

View attachment 89909

https://register.epo.org/ipfwretrieve?apn=US.202217992205.A&lng=de

View attachment 89908




As far as I am aware, Accenture has referred to Intel’s Loihi as a “neuromorphic research chip” in all their relevant patents at some point. Please provide proof of the contrary, thank’s!

Thanks, Frangi for your detailed post, yes, I did note the earlier challenges made against the claims laid out in the earlier patent, but having this patent "officially" granted 3 weeks ago is important to acknowledge, whether our company will be the beneficiary of anything remains to be seen, and my speed-reading has always got me in trouble, I'll go digging because I'm sure Intel isn't always referenced as an research chip, my question would be, why not compare apples with apples, i.e. commercially available technology with commercially available technology, Intel has more than enough issues going on internally, who really knows if all their research chips will ever see the light of day.

I reserve my opinion, and thanks once again for posting...do you personally feel from all your research that BRAINCHIP will surprise many by actually succeeding in this space, on a commercial footing that is ?

Best regards.......Tech (Perth)

 

[Diogenese]

Hi Bravo,

Last time I looked, Intel were having problems getting 18A to work properly - very low yield per wafer.

In olden times, Intel would have been able to throw fistfulls (fistsfull) of cash at the problem but now they are strapped for cash, and the sharks are circling. Clearly there has been a backdoor deal with US gvt - Intel is too big to fail. Under the Chips Act, the gov would want IFS to succeed so maybe that's where their investment has gone. Of course, Intel also needs help with AI.

Nvidia was blocked from buying ARM, so this may prevent any merger, but collaboration is possible, it's just that there could be lots of fingers in the pie. Maybe IFS will be hived off - it's already notionally customer-agnostic. This would leave Intel pretty much as it was before IFS in 2024.

As for Akida 3, I see this as being the Rolls-Royce option (aerospace, defence, ...) because of its INT16/FP32 capability. This will take up a lot more wafer real estate per chip, making it much more expensive to make, say >3X. I think Akida 3 will be a specialist high precision option, while Akida 2 will be the "standard" workhorse.

I don't know if 18A is suitable for RadHard.

The advantage of chiplets is that it would make Akida available for a much larger range of applications without needing specialized IP SoC designs/wafer tapeout/production for each customer.

One thing which adds urgency to the US/Intel deal is the threat to TSMC posed by the potential takeover of Taiwan.

TSMC is also setting up a US foundry under the Chips Act.

https://pr.tsmc.com/english/news/3210

HSINCHU, Taiwan, R.O.C. Mar. 4, 2025 – TSMC (TWSE: 2330, NYSE: TSM) today announced its intention to expand its investment in advanced semiconductor manufacturing in the United States by an additional $100 billion. Building on the company’s ongoing $65 billion investment in its advanced semiconductor manufacturing operations in Phoenix, Arizona, TSMC’s total investment in the U.S. is expected to reach US$165 billion. The expansion includes plans for three new fabrication plants, two advanced packaging facilities and a major R&D team center, solidifying this project as the largest single foreign direct investment in U.S. history.

BRN already has a connexion with TSMC, and with Global Foundries, so there will be a few options.

I see some of the financial soothsayers are drawing comparisons between the AI boom and the Dot.COM boom of the early 2000s. Now there will be casualties from the AI boom, but the AI boom is founded on bedrock, unlike the hot air of a lot of the Dot.COM boom pretenders, so it's a matter of choosing the right horse.

 

[Pom down under]

Probably the worse looking buy and sell side I’ve seen for a long time, just hope whoever is orchestrating this f.ks off sooner than later as it’s bloody depressing

 

[CHIPS]

Okay, it's odd that that is the only article "I" could find on the Net in relation to Loihi 3..
But I did find this, from April last year and actually from Intel..

Intel Builds World’s Largest Neuromorphic System to Enable More Sustainable AI

www.intc.com

"Hala Point packages 1,152 Loihi 2 processors produced on Intel 4 process node in a six-rack-unit data center chassis the size of a microwave oven. The system supports up to 1.15 billion neurons and 128 billion synapses distributed over 140,544 neuromorphic processing cores, consuming a maximum of 2,600 watts of power. It also includes over 2,300 embedded x86 processors for ancillary computations"

The numbers seem familiar ..
Somebody tell me what the frack is going on....
The previous article looks like it's..

View attachment 89894

So this "Hala Point" is the size of a microwave oven and is already in 4nm?!
It ain't getting smaller any time soon, is my guess...

GROK says this:

There is currently no publicly available information about a Loihi 3 chip from Intel. The latest data refers to the Loihi 2, introduced in September 2021 as Intel's second-generation neuromorphic research chip. Built on the Intel 4 process (previously known as 7 nm), it supports up to one million neurons and offers significant improvements over the original Loihi, including faster processing, better energy efficiency, and enhanced connectivity options.

Intel is currently focused on advancing and applying Loihi 2, along with its open-source Lava software framework for neuromorphic applications. There are no indications in available sources that a Loihi 3 is in development or has been announced. If you have specific rumors or sources about a potential Loihi 3, please share them for a more targeted response.

Intel's most recent major advancement in this area is the Hala Point system, announced in April 2024, which deploys 1,152 Loihi 2 processors to create the world's largest neuromorphic system with 1.15 billion neurons.

Projections for 2025 suggest Intel plans to integrate Loihi 2 into consumer devices for edge AI applications, but no announcements or details about a Loihi 3 have surfaced in official sources or recent reports.

The company continues to focus on advancing Loihi 2 through research collaborations, the open-source Lava software framework, and systems like Hala Point for sustainable AI.

 

[CHIPS]

Today’s post by Lauren Reinerman-Jones from SwRI (Southwest Research Institute) is further evidence that the US Air Force - while currently evaluating Akida - continues to be interested in Loihi:

It has been great sharing the excitement of our work with Air Force personnel in our #SwRI Dayton office. | Lauren Reinerman-Jones

It has been great sharing the excitement of our work with Air Force personnel in our #SwRI Dayton office. One of our partner’s at #Intel came to support #neuromorphic #AI demos to folks from #AFMC711HPW, #AFRLRYZA, #AFRLRST, and #AFLCMCHBY. Thank you to each of the Air Force personnel that took...

www.linkedin.com

View attachment 89919

https://www.linkedin.com/in/lauren-reinerman-jones-081496/

View attachment 89920

View attachment 89921

If the US Government is really going to buy 10% of Intel, this makes sense. The US Air Force must then take a closer look at Loihi 2, but if Akida is much better, they will not be forced to use Loihi 2, I suppose.

 

[CHIPS]

Why spend all that $$$$ when your not going to use what you buy

Buying Intel means securing the semiconductor industry in the US, jobs, reputation, and it is also part of MAGA. The Government always makes such a decision without consulting the parties needing the material. Even Trump wants the best drones, Marine, Air Force etc. He would not settle for less if somebody told him that Akida is better.

And they also cannot wait for years until Loihi 2 is ready for use and tested thoroughly. Akida is already there and tested!