BRN Discussion Ongoing

[Diogenese]

Is that good or bad

View attachment 95974

It's not Akida.

 

[Fullmoonfever]

It's not Akida.

Thanks for the translation.

 

[HopalongPetrovski]

Very nice


View attachment 95976

Very nice


View attachment 95976

LOOKING GOOD FOR THE AUCTION ATM

 

[buena suerte :-)]

LOOKING GOOD FOR THE AUCTION ATM

Great to see some 'Proper' movement!!

 

[Pom down under]

Great to see some 'Proper' movement!!

But the buy side is looking just as thin as my hair until 0.13

 

[HopalongPetrovski]

Great to see some 'Proper' movement!!

Hopefully dropping out of the ASX300 will mean we eventually fall off the shortee's radar.
I imagine a little more pain to come during witching as the funds make their adjustments and divest, but then, as their interests shift to juicier prey and the company comes out of stealth and starts making announcements, the sp can start to grow organically again.

 

[buena suerte :-)]

Go you good thing

 

[Bravo]

LOOKING GOOD FOR THE AUCTION ATM

Here's hoping we can suspend Isaac Newton’s “what goes up must come down” rule for a while.

Our version of this rule tends to go a bit more like this; "what goes up… slides back down and winds up lower than before."

 

[HopalongPetrovski]

Here's hoping we can suspend Isaac Newton’s “what goes up must come down” rule for a while.

Our version of this rule tends to go a bit more like this; "what goes up… slides back down and winds up lower than before."



View attachment 95981

Might be a good time to break out the trusty prayer helmet Bravo.

 

[smoothsailing18]

Go you good thing

View attachment 95980

Can you tell me please....We're did you find this picture ?

 

[Tony Coles]

In fact, it looks like they use the sequencer patent:

US2025190746A1 SYNCHRONIZED EXECUTION OF NEURAL NETWORK LAYERS IN MULTI-CORE ENVIRONMENTS 20231207

https://github.com/Leonui/tiny-npu?tab=readme-ov-file#npu

Microcode Controller​

The sequencing brain of the NPU. It fetches 128-bit microcode instructions from SRAM, decodes them into engine-specific commands, and dispatches them to the appropriate hardware engine.

A scoreboard tracks which of the 6 engines are currently busy, ensuring instructions only dispatch when their target engine is free. A barrier instruction forces the controller to stall until all engines complete, which is needed between dependent operations (e.g., wait for GEMM before starting Softmax).

GitHub - Leonui/tiny-npu: opensource NPU for LLM inference (this run gpt2) · GitHub

View attachment 95973


For matrices larger than 16x16, the GEMM controller tiles the computation automatically. A [16][64] * [64][256] GEMM is broken into (1)(4)(16) = 64 tiles of 16x16, with partial sums accumulated across the K dimension.

After accumulation, the post-processing unit applies requantization: result_i8 = clamp(round((acc * scale) >> shift)). This keeps all intermediate activations in INT8, minimizing SRAM bandwidth requirements.

In fact, it looks like they use the sequencer patent:

US2025190746A1 SYNCHRONIZED EXECUTION OF NEURAL NETWORK LAYERS IN MULTI-CORE ENVIRONMENTS 20231207

https://github.com/Leonui/tiny-npu?tab=readme-ov-file#npu

Microcode Controller​

The sequencing brain of the NPU. It fetches 128-bit microcode instructions from SRAM, decodes them into engine-specific commands, and dispatches them to the appropriate hardware engine.

A scoreboard tracks which of the 6 engines are currently busy, ensuring instructions only dispatch when their target engine is free. A barrier instruction forces the controller to stall until all engines complete, which is needed between dependent operations (e.g., wait for GEMM before starting Softmax).

GitHub - Leonui/tiny-npu: opensource NPU for LLM inference (this run gpt2) · GitHub

View attachment 95973


For matrices larger than 16x16, the GEMM controller tiles the computation automatically. A [16][64] * [64][256] GEMM is broken into (1)(4)(16) = 64 tiles of 16x16, with partial sums accumulated across the K dimension.

After accumulation, the post-processing unit applies requantization: result_i8 = clamp(round((acc * scale) >> shift)). This keeps all intermediate activations in INT8, minimizing SRAM bandwidth requirements.

Hi Diogenes, I think you seemed to be confused, does not work that way. Will send you my full analysis of how it works the correct way so you can learn from it. . PS I have no fu@king idea what you wrote nor understand but I love it…

 

[smoothsailing18]

FF

1..AkidaTag© is a complete Reference Platform and development kit launched by BrainChip (on March 10, 2026) to accelerate the design of always-on, battery-powered Edge AI devices. * Hardware Core: It combines the AKD1500 neuromorphic co-processor with Nordic Semiconductor’s nRF5340 wireless SoC. * Form Factor: Often showcased as a small battery-powered "puck" equipped with an accelerometer and microphone. * Capabilities: It enables real-time, on-device processing of motion, vibration, and acoustic data (e.g., keyword spotting, health vitals, anomaly detection) at milliwatt power levels. * Availability: Evaluation units are set for May 2026, with volume production scaling in Q3 2026.

2. Breaking the "Big Tech" Monopoly
You correctly identified that AkidaTag lowers the barrier to entry. This represents a democratization of hardware development: * The "Apple" Barrier: Previously, developing high-end wearables (like the Apple Watch) required massive vertical integration—custom silicon, high-tier algorithm teams, and deep pockets to solve the "always-on" battery drain. * The AkidaTag Shift: It provides a "blueprint" (schematics, firmware, and mobile app). A startup with a 10-person team can now use this platform to build a device that matches—or even exceeds—Apple's battery life for specific health or industrial monitoring tasks. * The Result: We are entering an era of "Specialized AI Wearables." Instead of one generic smartwatch, we will see niche devices: a ring dedicated to sleep apnea, a patch for Parkinson's tremor monitoring, or an industrial tag for motor health.

3. Selling AkidaTag: The StrategyHow does a sales team pitch this to a customer (like a Manager at Huawei or a MedTech startup)?| Strategy | The Sales Pitch ||---|---|| "Land & Expand" | Sell the Reference Kit first (low risk). Once the customer validates their algorithm (e.g., "detecting tank engine sounds"), they sign a high-volume AKD1500 chip order or an IP Licensing deal. || "Solve the Battery Crisis" | "Your current smart sensor dies in 2 days. With AkidaTag, it will last 2 weeks because it only 'wakes up' when a specific event occurs." || "Privacy First" | "No cloud needed. AkidaTag processes sensitive health or military data entirely on-device, making it unhackable via the network and fully private." || "Time-to-Market" | "Don't spend 2 years designing a board. Use our verified blueprint and launch your product in 6 months." |

4. Why Big Companies vs. Small Companies Like It * Small Companies LOVE it: It is their survival ticket. It allows them to compete with tech giants by offering "Single-Task Champions" (devices that do one thing perfectly with 30-day battery life) that big companies find too niche to bother with. * Big Companies WELCOME it: For a giant like Huawei, AkidaTag is a R&D shortcut. It allows their internal teams to rapidly prototype "Moonshot" ideas without redesigning the main SoC. If a prototype works, they integrate the Akida IP into their flagship Kirin chips.

5. Timeline to Watch (2026) * May 2026: First developers get their hands on AkidaTag for evaluation. * Q3 2026: Mass production of the AKD1500 begins. This is when we expect a revenue "ramp-up" for BrainChip (ASX: BRN). * Late 2026: Expect the first consumer products "Powered by Akida" to hit the market.

 

[schuey]

Maybe the Dog house door may be unlatched soon......Im hungry.....

 

[Diogenese]

Hi Diogenes, I think you seemed to be confused, does not work that way. Will send you my full analysis of how it works the correct way so you can learn from it. . PS I have no fu@king idea what you wrote nor understand but I love it…

I am a follower of the Chinese philosopher Confusus.

 

[Tothemoon24]

 

[buena suerte :-)]

Can you tell me please....We're did you find this picture ?

Hi ss18

It was on the Brainchip site!

Now without the 'pre order now' tab!?

 

[Pom down under]

Might be a good time to break out the trusty prayer helmet Bravo.

You don’t need a helmet you look just fine as it is

 

[Pom down under]

Weird as there seems to be the 2 similar sales going through around 4.30 and I’ve not checked everyday. Does anyone have the data to see how long this has been going on for?

 

[Ybbs]

It's working! The lower trend line on the weekly chart has held for now! On the daily chart, it could run to the green target line. Best regards from Germany.

 

[TECH]

As Sean stated "Watch Us Now"........ just 2.5 months into 2026, if you can't sense product development from clients is
slowly accelerating, well, time to crawl out from under your rock!

Nordic Semiconductor, check this company out, operating out of 13 other countries around the globe, just a small fry
who secured $648 million in the last reporting season, why on earth would they engage with us Jacob?

European Space Agency, looking forward to an update from them in the second half of 2026, how many more early
clients will show their wares later this year and into the first half of 2027, all I can say is "Watch Us Now", which can
literally mean, now or over a short period of a month or a quarter or "Watch Us Now" accelerate throughout the
year, nobody actually said the word immediately!!

Our BOD isn't going away anytime soon, so suck it up princess's