BRN Discussion Ongoing

[smoothsailing18]

I've been looking a bit further into the Arm Chiplet System Architecture (CSA) to determine how we might tie in to it and this morning I found a very interesting article (published 2 August 2025), which includes an equally interesting video.

At 2.28 mins into the video Eddie Ramirez, Vice President of Marketing at Arm says:

"So what we want to get to is where people talk about having compliant chiplet products to the standard, and the Arm IP will be designed to help enable that compliance as well. And so from our stand-point, we are very actively working on compute sub-systems that can be used to make the compute chiplets. We're also working with spans or IP interconnect, so that not only does it help connect multi-core Arm SOC's, but it also connects to accelerator chiplets as well. ".

Then, in the article there's a diagram which shows an “Acceleration” block, which is where it presumably allows for swap-in accelerators.

It doesn’t prove Akida specifically is in the plan, but if BrainChip delivers Akida 3 as a CSA/UCIe-compliant chiplet, it would fit perfectly into that “Acceleration” block which would be one of the cleanest paths to mass adoption for us IMO.



Set at 2.28 mins


View attachment 90308

Arming a Chiplet System Architecture

Arm’s Chiplet System Architecture, which addresses key chip design challenges, builds on the Neoverse Compute Subsystem and AMBA.

www.electronicdesign.com

Here's a reminder of Arm's chiplet plans as outlined by Renee Haas, CEO of Arm (see full article in the above post).



View attachment 90307

Brainchip needs to be apart of this chiplet revolution and I believe we are

 

[Cirat]

BRN stock is like

“COME ON SHORTER…

Well that was an interesting close ... a single buy of 2,435,328 vol

 

[CHIPS]

It was a sledge aimed specifically at 7, because the last Crypto "helper" was a "German doctor" who he commented on and he's very vocal in here.

I thought the other German's in here would have realised that and not taken personal offence, it's not like I mentioned the War or anything..

My Mother that raised me was German though, so I know boundaries of what can be deemed humor are limited.

Humor often pushes boundaries.

I apologise and am sorry, if anyone took real offence to what I said though.

Accepted . I have 7 on my ignore list; therefore, I cannot see his posts. Without context, your post did not seem nice.
Different nationalities have different kinds of humour.

 

[7für7]

Well that was an interesting close ... a single buy of 2,435,328

it happens several times this week… but don’t worry.. from Monday on, he can start to sell every minute one single share … Noooo he won’t…

Or was it the last punch of a mysterious unknown “captain America”? Who knows!?


GO BRAINCHIP (if you like it or not “potato chips”)

 

[Diogenese]

BluGlass is an Australian company developing high quality lasers for, inter alia, lidar:

https://cdn-api.markitdigital.com/a...38&v=c2533a54e2514fb77a8f93f84db686e1125273e9

...

BLG has a Gallium Nitride (GaN) laser manufacturing technique called RPCVD (Remote Plasma Chemical Vapour Deposition) which operates a few hundred degrees lower than conventional CVD. I'd think that the lower temperature would reduce dopant drift during deposition, producing cleaner edges between P-type and N-type regions, enabling production of narrow spectrum/single frequency lasers.

So the point of this unpaid advertisement is that Akida has a natural affinity for lidar. Imagine that: the world's best digital NN meets the world's best GaN laser in Australia ...

Aurizn has some claims in the AI field:

https://aurizn.co/enterprise/ai-consulting/
...
With our deep expertise in developing AI models, software and outcomes, we provide AI Consulting Services to organisations to help them benefit from their data.

We demystify the process of using your data and applying Machine Learning for improved data-driven decision making.

Machine vision
We unlock transformative potential with AI-enhanced machine vision, enabling workflow automation, intelligent surveillance, smart mining and remote infrastructure monitoring.

Natural Language processing
Employ AI's Natural Language to utilise models like ChatGPT, transforming unstructured data into actionable insights, summarising information, optimising operations and understanding customers.

AI Model Development
Using a process of experimentation, we help determine which model and techniques are best suited to solving your business need.



... so three in a bed?

Full disclosure: I do hold BLG shares

 

[7für7]

Already posted? Speculation

If so, sorry..

#mercedesbenz #mbos #glc #electricintelligence | Magnus Östberg

In just ten days, we will open the doors to our very special pre-night celebration at IAA MOBILITY 2025 and reveal our latest new Mercedes-Benz: the all-electric GLC. 🌟 As the brand’s Chief Software Officer, this is an exciting moment for me because it represents another milestone in our...

www.linkedin.com

 

[FJ-215]

Fun Fact , Our Roadmap tells us that in the First Quarter 26 , the AKD 2.0 goes into Production with a custom designed specialised ASIC .



$$$$$$$$$$$$$

View attachment 90302
View attachment 90303

Knowing the above , my guess when Tony Lewis said on LinkedIn that something is coming ,

View attachment 90304


Tony was expecting the Engineering Samples .

View attachment 90305



Is it not time for shareholders to get excited ?


.

Fun Fact , Our Roadmap tells us that in the First Quarter 26 , the AKD 2.0 goes into Production with a custom designed specialised ASIC .



$$$$$$$$$$$$$

View attachment 90302
View attachment 90303

Knowing the above , my guess when Tony Lewis said on LinkedIn that something is coming ,

View attachment 90304


Tony was expecting the Engineering Samples .

View attachment 90305



Is it not time for shareholders to get excited ?


.

Hi Doz,

I see it a little differently. I think his "something coming" comment is related to TENNs computer vision.

Tape out of Akida 2 is long overdue and will most likely be a reference chip rather than production of a full commercially available product. Why tape out a chip when a FPGA is available to test with? I think an actual chip maybe required for the next phase with the AFRL.

All guesses on my part.

 

[IloveLamp]

#spacecomputing | Frontgrade Gaisler

Dávid Juhász attended the opening of ESA Phi-Lab Sweden in Stockholm – the country’s new national laboratory for applied research in AI and edge learning in space systems. ESA Phi-Lab Sweden is part of ESA Phi-Lab NET, the European Space Agency - ESA’s network aimed at accelerating innovation...

www.linkedin.com

 

[manny100]

Latest AI Research Articles & Insights | Neuromorphic Core

Explore cutting-edge AI research, trends, and insights from industry experts. Stay updated with the latest breakthroughs at Neuromorphic Core.

neuromorphiccore.ai

Heaps and heaps of articles and research contained in the link. There is no doubt we are invested in a 'disruptive company' in a new and 'disruptive' industry segment.

 

[smoothsailing18]

From H/C

There’s been a lot of speculation on here, but it’s worth stepping back and looking at what’s actually in play.


1. Intel & the US Govt
The US Government is moving towards a circa 10% equity stake in Intel, converting CHIPS Act grants into ownership. That makes Intel a strategic asset, and it means Washington has influence if the DoD needs particular capabilities delivered quickly (like "get us some neuromorphic chips - NOW!")


2. BrainChip’s position
Akida is a commercially available neuromorphic processor, already proven in real-world use. It offers ultra-low power, edge-AI performance in SWaP-C constrained environments — exactly what defence, aerospace and autonomous systems require. By contrast, Intel’s Loihi 2 remains largely a research platform, not something deployable today.


3. Defence traction
BrainChip already has a footprint with US defence:

• A US$1.8m USAF SBIR contract with Raytheon on radar/micro-Doppler signal processing.
• A DOE SBIR Phase II project with Quantum Ventura (and Lockheed ties) on cybersecurity.
• Evaluation contracts with Bascom Hunter, collaborations with Blue Ridge Envisioneering and ISL, plus work with Airbus/ESA on space-grade deployments.
  
  All of this indicates the technology has cleared early security and vetting hurdles.

4. Foundry support
One historical headwind for BrainChip has been access to advanced foundry services. Being part of Intel Foundry’s IP ecosystem helps mitigate that. If the DoD required Akida scaled up rapidly, Intel now has both the manufacturing capacity and government incentive to make that happen.


5. Washington engagement
Recent meetings in Washington (Senate Armed Services, Senate Appropriations, House staffers) show BrainChip is not just engaging with engineers and integrators but also with policymakers. That’s how smaller tech firms build recognition in the federal contracting pipeline.


6. Strategic outlook
While listed on the ASX, BrainChip is fundamentally a US-centric company, with its contracts, partners, and government engagements all pointing that way. Over time, a Nasdaq listing wouldn’t be surprising, though that remains speculative at this stage. But, in case you might think being ASX listed is an impediment - it's not, it could be fixed quickly.


Joining the dots:
The US Government is doubling down on semiconductors, Intel is the anchor, and neuromorphic computing is a recognized strategic need. BrainChip’s Akida is the most deployment-ready neuromorphic IP available. The combination of existing defence contracts, Intel foundry access, and Washington engagement doesn’t guarantee outcomes, but it does place BrainChip in a credible position to benefit if neuromorphic AI becomes a US defence priority.

Some have been critical of Brainchip's board and management lacking commercialization outcomes, but it looks to me like they are making some very astute plays. Looping in home district support via David Min - California's 47th Congressional District in the heart of Orange County (where Brainchip is located) who was present when Dr. Jonathan Tapson BrainChip CTO visited Washington, and no doubt the facilitator - is a smart play. And targeting a customer (DoD) that has both an immediate strategic need and the money to place a big order is text book.

Some serious moving of chess pieces around the board.

 

[7für7]

Already posted? Speculation

If so, sorry..

#mercedesbenz #mbos #glc #electricintelligence | Magnus Östberg

In just ten days, we will open the doors to our very special pre-night celebration at IAA MOBILITY 2025 and reveal our latest new Mercedes-Benz: the all-electric GLC. 🌟 As the brand’s Chief Software Officer, this is an exciting moment for me because it represents another milestone in our...

www.linkedin.com

I asked chatty..


My personal assessment

• >70% probability that Mercedes will integrate some form of neuromorphic technology (e.g. Akida) into MB.OS – maybe not across all vehicles, but as a module for selected functions.
• >50% probability that it will specifically be BrainChip Akida (because of the existing partnership and their prior work together in the EQXX).
• <30% probability that Mercedes will rely solely on conventional SoCs and justify the “Superbrain” hype purely through software. That would be possible, but it would be untypical for a brand that likes to differentiate itself through technology.

In short: I realistically see MB.OS as a hybrid – Nvidia as the main computing platform, plus additional chips like Akida for edge learning. Mercedes won’t market it as “Akida inside” but rather under broader labels like “intelligent personalization” or “hyper-personalized experience”.

 

[Doz]

Dont ask me , Kurt‘s the one banging on about Global Foundaries and their 8nm .

And no , I can’t access this IEEE article , but it sure looks interesting .

Global Foundaries and DoD , like a hand in a glove .

 

[Diogenese]

Hi Doz,

I see it a little differently. I think his "something coming" comment is related to TENNs computer vision.

Tape out of Akida 2 is long overdue and will most likely be a reference chip rather than production of a full commercially available product. Why tape out a chip when a FPGA is available to test with? I think an actual chip maybe required for the next phase with the AFRL.

All guesses on my part.

Hi FJ,

My thoughts are that Akida 2 with TENNs will be workhorse (FJ), while Akida 3 will be the caddy (the car, not golf).

Akida 2 has INT8-bit accuracy, while Akida3 has INT16-bit plus FP32-bit, so Akida 3's silicon footprint will be much larger than Akida 2's. This will mean fewer chips per wafer, hence higher cost per chip to manufacture, maybe 4 times or more. This indicates that Akida 3 is intended for high precision applications where cost is a secondary consideration, like defence/aerospace.

There was a false dawn Akida 2 tapeout rumour towards the end of last year. At the time the ViT option was offered only with the larger configurations, so I assume that it required a large number of NPUs. ViT is not mentioned now, as it seems that in playing around with TENNs, they found it could do the job with fewer NPUs.

I think the "something coming" was the Akida 2 FPGA cloud access. This is a great sales tool as it enables potential customers to familiarize themselves with the tech, and try out their models before committing to purchasing the hardware.

So the question is, will we be making Akida 2 chips?

We now have a couple of Edge Boxes, the M2 card, and the PCIe card all with Akida 1. Is there an upgrade in the pipeline?

The early models are provided free, but the advanced TENNs models command a licence fee. The TENNs models can run on an ordinary PC, which is a potential income stream before there are any Akida 2 chips.

The thing is that cost is a major barrier to entry for anyone taking a licence to make Akida. There is a substantial licence fee on top of the tapeout and manufacturing costs. Even if we were to decide to make to order, there is a significant time lag in queueing for time at the foundry on top of the actual manufacturing, testing, pinout and packaging time.

Then there is the question of chip specialization. It's all very well to be able to select the number of nodes (up to 256 nodes - 1024 NPUs), but that involves a separate production run for each variation. So does that lead us back to the likes of ARM chiplets?

As I said, I think anyone who wants/needs Akida 3 would have the readies to pay up front/bankroll production and that's still some way off, but Akida 2? We need a couple of well heeled customers soon ...

 

[manny100]

Just revisiting the 'Jonathan Tapson goes to Washington' linked in post.
" I visited several legislators last week to promote our Federal contracting agenda. The first photo shows me at our discussion with Senator Cindy Hyde-Smith, who is on the Senate Committee on Appropriations; other appointments were with Senator Roger Wicker, Chair of the Senate Armed Services Committee, and Representative David Min, our Laguna Hills CA-47 Rep. I also gave a presentation on BrainChip".
" More seriously - the US AI industry is becoming increasingly integrated with Defense and associated Departments in the US Government, and companies such as Anduril and Palantir are showing the way. BrainChip will be part of this integration - I also visited potential partners in the Beltway on this visit. Looking forward to an exciting future! "
My bold above.
Note he says BrainChip will be part of the US AI integration with defense and associated Departments. He says will be not maybe or might be. Its a Will be. - which is pretty close to a certainty - in his view.

 

[Doz]

Hi Doz,

I see it a little differently. I think his "something coming" comment is related to TENNs computer vision.

Tape out of Akida 2 is long overdue and will most likely be a reference chip rather than production of a full commercially available product. Why tape out a chip when a FPGA is available to test with? I think an actual chip maybe required for the next phase with the AFRL.

All guesses on my part.

Hi FJ , I’m fairly sure that Brainchip has never had any intentions to Tapeout the AKD 2.0 and it looks like someone with considerable financial clout and a need for advanced technology has stepped up to the plate . I imagine that the unknown client will also be aware that ordering in bulk is better value for money .

Now whatever you do don’t ask me who the customer is , or otherwise I might jump to conclusions and say the US Department of Defence .

And my money is on Global Foundaries 22nm FD SOI for the Tapeout of the specific ASIC’s .

All in my opinion ,

 

[Diogenese]

From H/C

There’s been a lot of speculation on here, but it’s worth stepping back and looking at what’s actually in play.


1. Intel & the US Govt
The US Government is moving towards a circa 10% equity stake in Intel, converting CHIPS Act grants into ownership. That makes Intel a strategic asset, and it means Washington has influence if the DoD needs particular capabilities delivered quickly (like "get us some neuromorphic chips - NOW!")


2. BrainChip’s position
Akida is a commercially available neuromorphic processor, already proven in real-world use. It offers ultra-low power, edge-AI performance in SWaP-C constrained environments — exactly what defence, aerospace and autonomous systems require. By contrast, Intel’s Loihi 2 remains largely a research platform, not something deployable today.


3. Defence traction
BrainChip already has a footprint with US defence:

• A US$1.8m USAF SBIR contract with Raytheon on radar/micro-Doppler signal processing.
• A DOE SBIR Phase II project with Quantum Ventura (and Lockheed ties) on cybersecurity.
• Evaluation contracts with Bascom Hunter, collaborations with Blue Ridge Envisioneering and ISL, plus work with Airbus/ESA on space-grade deployments.
  
  All of this indicates the technology has cleared early security and vetting hurdles.

4. Foundry support
One historical headwind for BrainChip has been access to advanced foundry services. Being part of Intel Foundry’s IP ecosystem helps mitigate that. If the DoD required Akida scaled up rapidly, Intel now has both the manufacturing capacity and government incentive to make that happen.


5. Washington engagement
Recent meetings in Washington (Senate Armed Services, Senate Appropriations, House staffers) show BrainChip is not just engaging with engineers and integrators but also with policymakers. That’s how smaller tech firms build recognition in the federal contracting pipeline.


6. Strategic outlook
While listed on the ASX, BrainChip is fundamentally a US-centric company, with its contracts, partners, and government engagements all pointing that way. Over time, a Nasdaq listing wouldn’t be surprising, though that remains speculative at this stage. But, in case you might think being ASX listed is an impediment - it's not, it could be fixed quickly.


Joining the dots:
The US Government is doubling down on semiconductors, Intel is the anchor, and neuromorphic computing is a recognized strategic need. BrainChip’s Akida is the most deployment-ready neuromorphic IP available. The combination of existing defence contracts, Intel foundry access, and Washington engagement doesn’t guarantee outcomes, but it does place BrainChip in a credible position to benefit if neuromorphic AI becomes a US defence priority.

Some have been critical of Brainchip's board and management lacking commercialization outcomes, but it looks to me like they are making some very astute plays. Looping in home district support via David Min - California's 47th Congressional District in the heart of Orange County (where Brainchip is located) who was present when Dr. Jonathan Tapson BrainChip CTO visited Washington, and no doubt the facilitator - is a smart play. And targeting a customer (DoD) that has both an immediate strategic need and the money to place a big order is text book.

Some serious moving of chess pieces around the board.

Hi FJ,

My thoughts are that Akida 2 with TENNs will be workhorse (FJ), while Akida 3 will be the caddy (the car, not golf).

Akida 2 has INT8-bit accuracy, while Akida3 has INT16-bit plus FP32-bit, so Akida 3's silicon footprint will be much larger than Akida 2's. This will mean fewer chips per wafer, hence higher cost per chip to manufacture, maybe 4 times or more. This indicates that Akida 3 is intended for high precision applications where cost is a secondary consideration, like defence/aerospace.

There was a false dawn Akida 2 tapeout rumour towards the end of last year. At the time the ViT option was offered only with the larger configurations, so I assume that it required a large number of NPUs. ViT is not mentioned now, as it seems that in playing around with TENNs, they found it could do the job with fewer NPUs.

I think the "something coming" was the Akida 2 FPGA cloud access. This is a great sales tool as it enables potential customers to familiarize themselves with the tech, and try out their models before committing to purchasing the hardware.

So the question is, will we be making Akida 2 chips?

We now have a couple of Edge Boxes, the M2 card, and the PCIe card all with Akida 1. Is there an upgrade in the pipeline?

The early models are provided free, but the advanced TENNs models command a licence fee. The TENNs models can run on an ordinary PC, which is a potential income stream before there are any Akida 2 chips.

The thing is that cost is a major barrier to entry for anyone taking a licence to make Akida. There is a substantial licence fee on top of the tapeout and manufacturing costs. Even if we were to decide to make to order, there is a significant time lag in queueing for time at the foundry on top of the actual manufacturing, testing, pinout and packaging time.

Then there is the question of chip specialization. It's all very well to be able to select the number of nodes (up to 256 nodes - 1024 NPUs), but that involves a separate production run for each variation. So does that lead us back to the likes of ARM chiplets?

As I said, I think anyone who wants/needs Akida 3 would have the readies to pay up front/bankroll production and that's still some way off, but Akida 2? We need a couple of well heeled customers soon ...

Hi Smoothsailing,

Thanks for this.

I had subconsciously written off Intel, but Intel/DoD does tie in with my rich customer hypothesis, even if Intel does not have the readies itself.

 

[Diogenese]

#spacecomputing | Frontgrade Gaisler

Dávid Juhász attended the opening of ESA Phi-Lab Sweden in Stockholm – the country’s new national laboratory for applied research in AI and edge learning in space systems. ESA Phi-Lab Sweden is part of ESA Phi-Lab NET, the European Space Agency - ESA’s network aimed at accelerating innovation...

www.linkedin.com

View attachment 90334

All these LEOs are going to stuff up terrestrial astronomy, and put a few astrologer's noses out of joint too.

Should be able to pick up a cheap 2nd hand observatory soon.

 

[Deadpool]

It was a sledge aimed specifically at 7, because the last Crypto "helper" was a "German doctor" who he commented on and he's very vocal in here.

I thought the other German's in here would have realised that and not taken personal offence, it's not like I mentioned the War or anything..

My Mother that raised me was German though, so I know boundaries of what can be deemed humor are limited.

Humor often pushes boundaries.

I apologise and am sorry, if anyone took real offence to what I said though.

LOL

 

[Deadpool]

The above post, all in good humor, please don’t take offense, our dear Deutsch friends

 

[FJ-215]

Hi FJ,

My thoughts are that Akida 2 with TENNs will be workhorse (FJ), while Akida 3 will be the caddy (the car, not golf).

Akida 2 has INT8-bit accuracy, while Akida3 has INT16-bit plus FP32-bit, so Akida 3's silicon footprint will be much larger than Akida 2's. This will mean fewer chips per wafer, hence higher cost per chip to manufacture, maybe 4 times or more. This indicates that Akida 3 is intended for high precision applications where cost is a secondary consideration, like defence/aerospace.

There was a false dawn Akida 2 tapeout rumour towards the end of last year. At the time the ViT option was offered only with the larger configurations, so I assume that it required a large number of NPUs. ViT is not mentioned now, as it seems that in playing around with TENNs, they found it could do the job with fewer NPUs.

I think the "something coming" was the Akida 2 FPGA cloud access. This is a great sales tool as it enables potential customers to familiarize themselves with the tech, and try out their models before committing to purchasing the hardware.

So the question is, will we be making Akida 2 chips?

We now have a couple of Edge Boxes, the M2 card, and the PCIe card all with Akida 1. Is there an upgrade in the pipeline?

The early models are provided free, but the advanced TENNs models command a licence fee. The TENNs models can run on an ordinary PC, which is a potential income stream before there are any Akida 2 chips.

The thing is that cost is a major barrier to entry for anyone taking a licence to make Akida. There is a substantial licence fee on top of the tapeout and manufacturing costs. Even if we were to decide to make to order, there is a significant time lag in queueing for time at the foundry on top of the actual manufacturing, testing, pinout and packaging time.

Then there is the question of chip specialization. It's all very well to be able to select the number of nodes (up to 256 nodes - 1024 NPUs), but that involves a separate production run for each variation. So does that lead us back to the likes of ARM chiplets?

As I said, I think anyone who wants/needs Akida 3 would have the readies to pay up front/bankroll production and that's still some way off, but Akida 2? We need a couple of well heeled customers soon ...

Hi Dio,

Had always wondered about the cost of tape out on Akida 2. Assumed they would do the 3 versions on a multi project wafer if it was possible.

Trying to remember where Anthony Lewis said they didn't want to produce a physical chip as it might scare off IP sales?