Read this before investing in BRN, page-55

@pickbrain

What a delicious melange: fact and lie soup with a straw man side.

G'day. This is my first anniversary since I've jumped offthe ‘rocketship’ Brainchip. I'd like to take this moment to update my report. [#### You must have been miffed that you did not sell 3 weeks earlier ####]

Don’t take this as financial advice

[#### The reference to Theranos and Nikola is merely an historical anachronism ####]

as this speech can be as biased as cheerleaders of thisforum. Although my small input won't even budge the scale.

I have raised doubt last year that Brainchip is not whatmany praises it to be.

[#### Did you have any evidence to support those doubts? ####]

The consensus at the time amongst cheerleaders was thatthere will be massive contracts flooding come Christmas 2020. Guess whathappened? Nada. [#### OTC upgrade; Renesas IP contract ####]

Now we’re coming close to the ‘next’ Christmas.

Think of my 5c as Hindenburg Research. In case you don’tknow, it was research done on the company Nikola which exposed fundamentals andrevalued the stock.

Or the WSJ report that exposed Theranos that built a fakeempire using colorful promises of advanced technology to deceive shareholders. [#### clearly not financial advice ####]

Lack of IP SALE

I invested because back then the company targeted thesilicon to be ready in 2018. That was a false promise but anyway, IP was stillavailable. [#### 3^rd Quarter of 2019 to be accurate: https://s3.amazonaws.com/content.stockpr.com/brainchipinc/db/217/1581/pdf/BrainChip+Announces+Akida+Neuromorphic+System-on-Chip.pdf ####]

[#### https://brainchipinc.com/neuromorphic-chip-maker-takes-aim-at-the-edge/ ####]

via The Next Platform January 30th, 2020

When we last spoke with BrainChip in 2018, the company was on the verge of rolling out its FPGA-based spiking neural network (SNN) accelerator, known as Akida (Greek for spike). At that point, the plan wasto get its hardened SoC into the market by 2019. Since then, engineers there have been busy refining the design, and that included a majoraddition. Instead of just supporting a spiking neural network computing model,they’ve integrated the capability to run convolutional neural networks (CNNs)as well.

According to Roger Levinson, BrainChip’s chief operating officer, the CNN support was incorporated to make the solution a better fitfor their initial target market of AI at the edge. Specifically, since convolutional neural networks have proven to be especially adept at picking up features in images, audio, and other types of sensor data using these matrix math correlations, they provide a critical capability for the kinds of AI applications commonly encountered in edge environments. Specific application areas being targeted include embedded vision, embedded audio, automated driving (LiDAR, RADAR), cybersecurity, and industrial IoT. The addition of the CNNcapability did, however, push the company’s chip launch into 2020.

“It delayed getting the Akida SoC out by quite a bit,” Levinson tells The Next Platform, “but was critical to give us what we think is a commercially viable, advanced solution.” ####]

In 2016, BRN acquired a French company called SpikeNet.This was the company that was able to simulate Spiking Neural Network insoftware. Thorpe S J has a lot of published papers on SNN and Spike timingdependent plasticity (STDP) and SpikeNet contributed greatly to the developmentof IP.

Whilst I was invested, I hoped some manufacturer willadopt IP (Brainchip’s SNN patents) and create revenue but that hope wasdiminishing each day.

Long story short, none of the companies took onBrainchip’s IP. [#### Accidental untruth? ####]

Speed the clock to now, Samsung, Apple, Huawei, all thebig tech company utilizes its own Machine Learning Algorithms.

Is there any chance that IP can still be adopted by a bigmobile phone company?

Ask yourself, if it didn’t sell for the last three years,and AI during that time has advanced to the point that all the mobile phonecompanies have their neural fabric in their processor along with algorithm, whydo you think IP will sell now? [#### Ask yourself why Renesas signed an Akida IP licence ####]

There are more reasons than three years ago for IP to notsell.

Is there any chance that AKIDA1000 chip may be adopted bya big mobile phone company?

[#### Straw man 1: Why are we talking about the Akida chip in mobile phones? Anyone who understands the tech would lnow that if Akida were to be incorporated in a mobie phone it would be the Akida IP – in fact, if the order is over 1 million, we recommend the Akida IP instead of the chip. ####]

If you saw PCB board on a mobile phone, you’d know thatit is tiny.AKIDA chip is 15mm by 15mm big.

For the company to allow 15mmx15mm + overhead forAKIDA1000 chip is asking for too much when their smartphone processor alreadycontains neural fabric. [?]

CPU in AKIDA chip is nothing special. It is NOT made byBrainchip. It is Arm-Cortex-M4 and this is a processor built by Armoptimized for discrete processing and microcontrollers. Budget CPU at best.

[#### Straw man 2: The ARM Cortex is not built by ARM – ARM licence their IP. Akida is processor agnostic so it can be matched with any processor the customer wants. In fact, the ARM processor is not part of the Akida IP licence. The processor used with Akida is only used for setup/configuration – it does not get involved in the SNN data processing. ####]

To put things into perspective, Samsung’s silicon on themobile Exynos 1080 has 8 high performing processors. 4 x Arm-Cortex-A78 and 4xCortex-A55 on the single chip. 8 Processors on the single-chip and A78 is aflagship processor from Arm that is designed to handle complex compute tasks.

[#### … wouldn’t want to go to a wine and chalk night at your place … How long would that run on a battery powered edge device? ####]

Do you think they will fancy the idea of wasting15mmx15mm footprint to take on an inferior CPU?

SNN accelerator vs SNN processor If you read the forum,whenever someone asks what people think about AI chip from another company, youoften hear that they are inferior because they are just accelerators. So whatexactly does it mean?

It means the chip uses SNN algorithm

[#### “algorithm”? - Akida 1000 has a silicon spiking neural network using standard CMOS hardware circuits ####]

but it does not have a processor built-in to the chip soit requires a separate processor to run the device. But as I said above, theprocessor used in AKIDA is nothing fancy, it is just an off-the-shelfArm-Cortex-M4 processor.

An advantage of having a processor is that if thedesigner is building a device from the ground up and it demands a simpleprocessing task by CPU, such as face detection doorbell, it can utilize justAKIDA chip without needing a separate processor.

[#### At last a morsel of truth ####]

But if the designer wants to adopt AI chip to speed up MLalgorithms whilst retaining their existing architecture (CPU) to control allthe rest of the accessories, then the accelerator suffices. Having an inferiorM4 processor taking precious footprint is added burden.

[#### Straw man 3: Akida does better ML on its own. If we’re still talking mobile phones, (precious footprint), or any other bulk product the Akida IP does not come with the ARM cortex. ####]

Thus if the customer already has working silicon in theirdevice and just wants something to speed up its AI application, they can chooseeither accelerator or processor.

[#### and you know that the accelerator is as good as Akida? Or that the processor will handle ML more efficiently than Akida? ####]

So there are advantages and also disadvantages of beingan SNN processor. It’s not superior in any means.

[#### if you were aware of all these deficiencies, why did you buy BRN in the first place ####]

Is SNN a prerequisite in Edge devices?

Well, the general theory behind SNN is that it will consume less power than its counterpart.

But if the power consumption of a few milliwatts is negligible (in the case of electric car, CCTV) then any edge AI chip can be utilized.

[#### Do you know how much power is consumed by ADAS computers? Image processing using CNN is a major load of ADAS. ####]

For example, NDP10x chip can compete with AKIDA in speechrecognition with similar power consumption and accuracy. If you’re developingsmart speaker then it is easy to buy NDP10x chip with a footprint of1.4mmx1.8mm than it is to buy AKIDA with 15mmx15mm as you won’t use otherfunctions such as visual/olfactory/tactic that is offered by AKIDA as yourdevice won’t simply use those features. BTW, this is the chip that Amazon usesfor voice assistant “Alexa”.

So if you want to simply build an AI speaker, hearingaid, etc when space is always a constraint, would you choose AKIDA or somethingthat is 1% the size of the AKIDA chip?

Other company that is already utilizing the smart doorlock that BRN wants to tap into is Kneron. Kneron is Chinese AI edge companyalso targeting Smart City, Smart Vehicle, Financial, Security as their customerbase. When CEO of BRN said they will not compete in Chinese market, it couldrelate to protecting patents but also interpreted as the market is alreadycompetitive as Chinese government grants subsidies to their own AI companies tofuel growth. KL720 is able to compute in 8bit, and 16bit and can beconfigured to CNN, RNN, LSTM. This means KL720 can process 4K videos and drivenatural language processing (NLP) for translators and AI assistants which AKIDAcan’t do.

The biggest difference between Loihi and AKIDA1000 Thebiggest difference between Loihi and AKIDA1000 lies in Loihi’s ability to doRecurrent Neural Network (RNN) and having multi-compartment dendrite function.This means Loihi is technically more advanced.

[#### The biggest difference between Liohi and Akida is that Loihi is still in the lab ####]

Intel opened up a cloud system where potential customerscan try different models. Intel research has signed up over 100 membersincluding Accenture, Airbus, GE, Berkeley, and Washington State University,Lenovo, Logitech, Mercedes-Benz and Prophesee.

[#### The best thing about Loihi is that it is a red herring for potential Akida competitors ####]