BRN Discussion Ongoing

This is a short article which sums up the sort of information being consumed now by the industry every day about Brainchip and its technology. Not sure if it has been posted already as there are so many similar articles across so many publications that come up when I Google search now. I love these days but it was a lot easier when only a few select individuals knew what Brainchip had and what world leading companies were verifying and adopting their AKIDA technology advantages:


My opinion only DYOR
FF

AKIDA BALLISTA
 
  • Like
  • Fire
Reactions: 33 users

Build-it

Regular
The weather has fined up and Brainchip has won the toss and to take advantage of what looks from up here to be a pretty good batting wicket has elected to bat. Opening for Brainchip will be AKD1000 and the IP. We go down to our on field commentator Tata Elxsi for a wicket condition report. Over to you Tata:

Edge AI Solutions

Innovate|Design|Scale
Edge AI Solutions

“Edge AI is here to stay! Artificial intelligence (AI) is powering many real-world applications which we see in our daily lives. AI, once seen as an emerging technology, has now successfully penetrated every industry (B2B & B2C) Banking, logistics, healthcare, defense, manufacturing, retail, automotive, consumer electronics. Smart Speaker like Echo, Google Nest, is one such example of Edge AI solutions in the consumer electronics sector.
AI technology is powerful, and human-kind has set its eye on the path of harnessing its potential to the fullest. Intelligence brought to the device can be very useful and creative.
The key requirements that need to be factored in designing Edge AI architecture are — bandwidth, latency, privacy, security, power consumption. While envisioning an Edge AI solution, these requirements need to be thoroughly weighed in terms of what feature can be traded off and yet be effective.
AI enables the machines to perform cognitive functions such as perceiving, reasoning, and learning similar to humans but much faster and accurately. AI implementation is majorly classified into two phases — Learning and Inference.
Globally, the AI chipset market size is expected to be valued at USD 7.6 billion in 2020 and likely to reach USD 57.8 billion by 2026, at a CAGR of 40.1% during this period. Implementation of AI is the current trend in chip technology, and it’s going to stay that way. Many leading semiconductor companies and venture capitalists see it as the right tech-front for investment.

Opportunities & Challenges

Changing dynamics in terms of hardware consideration for learning and inference has led to the Edge AI hardware market being segmented into CPU, GPU, ASIC, and FPGA. ASICs enable high processing capability with low-power consumption, making them perfectly suited for Edge devices in many applications. It is estimated that an Edge AI architecture inference implemented on ASIC will grow from 30% to 70% and 20% on GPU by 2025. Edge devices are embedded products with resource constraints, and hence, Edge AI implementation needs to be thought of as an application-specific use case. AI-based applications for Edge devices are intelligent robots, autonomous vehicles, smart home appliances, among others. The primary applications that run over Edge AI are related to image/video, sound, and speech, natural language processing, device control system, and high-volume computing.
The global Edge AI software market is estimated to cross $3 Billion by 2027. At this juncture of technology innovations, Semiconductor companies enable AI solutions to realize newer strategies to grow their business and find wider hardware adoption. Many semiconductor companies are no longer seen as just component providers but as complete platform solution providers. Semiconductor companies realize value gained from software and services associated with the chipsets that allow for the rapid adoption of their platforms by the device manufacturer.
To increase their hardware market adoption, semiconductor companies are investing heavily in the software development toolkits integrated with ML/DL frameworks to deliver as a package that allows developers to quickly get started with all components for embedded systems development at ease. This allows the device manufacturer to effectively utilize the silicon resources in a shorter time-span and gain an advantage by being first to market. Edge AI chipset market is witnessing software and associated technology stacks facilitating wider adoption and faster development cycle.
AI processing at the Edge has allowed semiconductor companies and electronic device manufacturers to look beyond the horizon and re-define themselves with innovative solutions. It would be safe to say that Edge AI is driving the digital transformation and guarantees immense potential.”


The interesting aspect of all the assessments of Ai at the Edge is that every potential difficulty that commentators and industry point out Peter van der Made, Anil Mankar and the amazing team at Brainchip past and present have addressed in the AKIDA technology solution.

It is almost as if the Brainchip team had made a decision to identify every possible requirement and difficulty that could ever arise at the edge and then designed on the basis that AKIDA was going to conquer every single one. Ludicrously low power, no need for cooling, ultra high performance, capacity to use all existing networks and convert same to SNN on chip, incremental and on chip learning, unconnected autonomous processing when required, intelligent connection to address bandwidth issues, cyber secure, sensor agnostic, incredible breadth of scalability, digital, chip or IP, training via Tensorflow.

My opinion only DYOR
FF

AKIDA BALLISTA
[/QUOTE

Great shot classic cover drive.
Don't change your bat you certainly hit that one out of the middle.

Edge Compute
 
  • Like
  • Haha
Reactions: 4 users

Slymeat

Move on, nothing to see.
Hey Slymeat,

I'm curious, where is the serial number printed? And if you can remember, how soon after these were first advertised on the BRN website did you make the purchase?
If the serial number is an indication that it is really the195th PCIE board packaged and shipped, then there is US$97,500 revenue so far from these cards :)
(I'm hoping you would say that you placed the order within the first few days of the release :) )
Also, I love how there is five zero's at the beginning of the serial number...
@JK200SX The serial number is on the barcode attached to the device and was also on my invoice when I ordered it.

I ordered my board on the very first day they were advertised, probably just a few hours after in fact.
 
  • Like
  • Fire
Reactions: 39 users

JK200SX

Regular
@JK200SX The serial number is on the barcode attached to the device and was also on my invoice when I ordered it.

I ordered my board on the very first day they were advertised, probably just a few hours after in fact.
OK, my bet is they've already sold 2000+ PCIE boards $:)$
 
  • Like
  • Fire
  • Haha
Reactions: 14 users
Hi FF its good to see every one on cricket pitch and commentary is about to start, we all know who are on the other side with big hope of scoring tons but its the players on our side that we dont know much about :p we need 11 players to play the game is about to begin but so far only 6 or 7 players on the field and still few players to come on the field mind you we do need extra player over the stand :) as well . Lets enjoy the game now or shall we wait Bhahahahahah

cheers
🤔
 
  • Like
  • Thinking
Reactions: 2 users

Violin1

Regular
  • Haha
  • Like
Reactions: 10 users
Dell a done deal? Ken Dodds' dad's dog's dead!
I thought it was David Dodd’s dads dogs died . But I could be wrong , unless Ken and David are brothers . Who cares anyway , it was a green day for BRN
 
  • Like
  • Haha
Reactions: 6 users

Foxdog

Regular
The RAAF had a pilotless plane (drone) back in the '60s - Jindivik - used it to tow live fire Ack-Ack targets.

Made by the Government Aircraft factory.

If only we could dig up all those old engineers ...
M8 they were still using them in the 90's. I used to watch them take off.....ah the good old days 🤣
The RAAF had a pilotless plane (drone) back in the '60s - Jindivik - used it to tow live fire Ack-Ack targets.

Made by the Government Aircraft factory.

If only we could dig up all those old engineers ...
 
  • Like
Reactions: 6 users
Thanks for posting that Q.

A technical question. Does Quadric accelerate others AI or do they have their own AI?

I’m questioning if Megachips has bought into Quadric to accelerate Akida. I was of the view, based on an interview with Anil Mankar and Denso staff together that Akida was going to be implemented in Denso technology so querying if that would be why Denso has announced they have signed Quadric.

Are they an add-on to help accelerate the information or decision making after Akida has done it’s work at the sensor?

Denso “Having evaluated Quadric’s q16 processor, its ability to run many types of algorithms efficiently and flexibly allows Quadric’s platform to enable AI in new services and products. We look forward to continuing to work closely with Quadric and plan to integrate their IP into DENSO’s SoCproducts.”

I’d be interested if someone with some technical expertise who could shed light on this please?

Cheers

I’ll start with a caution that I don’t know anything about this technology industry, it’s far from my field of expertise: but I can read and hopefully I have interpreted the words meanings correctly and can answer this one myself.


Introducing the First Generation quadric architecture.​

We built this architecture to meet the demands of various workloads. From Neural Networks to Digital Signal Processing to Computer Vision to the State of the Art that you're developing today we've got you covered. Develop your algorithms once and deploy them on any device that has a quadric architecture instance. When our architecture improves so will your algorithms.

So based on the description above I am surmising the announcement from Denso that they are using Quadric q16 in the development of their vehicles that there is still a chance Akida will be included. That the Quadric chip‘s purpose is to accelerate the algorithms after Akida has done it’s job at the sensor?

The reason I am trying to get to the bottom of this is that Denso is a massive in vehicles, they are part of the Toyota group, and would be great to have onboard.

Another reason I am thinking Denso are using Quadric to work with Akida is that they are both under Megachips banner and Denso is Japanese as well.

If anyone with technical knowledge wants to tell me I’m wrong I’m all ears, as I’m trying to learn.

By the way: great article about Mercedes Benz today FF. Fantastic news!

Cheers,
 
Last edited:
  • Like
  • Fire
Reactions: 22 users

RobjHunt

Regular
Do you really think Dell will be a done deal
Do you? What real response are you after? Some things never change even if the platform does.
 
  • Like
  • Haha
Reactions: 8 users

Guzzi62

Regular
I just been looking in the Brainchip store:

The Development kit- Shuttle PC is still available for just under 10 grand.

Also the PCIe board can be had for 500 bucks.

Lets hope they will get stock soon of the Development kit Raspberry Pi kit as it certainly will be in everyone's best interest to supply eager customers with what they want ASAP.

 
  • Like
Reactions: 8 users

Iseki

Regular
"a soldier firing an NLAW simply points the weapon at a moving vehicle, engages the guidance system and tracks the target for a few seconds before firing. The missile then flies to a point where it predicts the target will be."


is looks to me like Akida. One step learning, two step learning, then pull the trigger.
 
  • Like
  • Wow
Reactions: 4 users

Diogenese

Top 20
I’ll start with a caution that I don’t know anything about this technology industry, it’s far from my field of expertise: but I can read and hopefully I have interpreted the words meanings correctly and can answer this one myself.


Introducing the First Generation quadric architecture.​

We built this architecture to meet the demands of various workloads. From Neural Networks to Digital Signal Processing to Computer Vision to the State of the Art that you're developing today we've got you covered. Develop your algorithms once and deploy them on any device that has a quadric architecture instance. When our architecture improves so will your algorithms.

So based on the description above I am surmising the announcement from Denso that they are using Quadric q16 in the development of their vehicles that there is still a chance Akida will be included. That the Quadric chip‘s purpose is to accelerate the algorithms after Akida has done it’s job at the sensor?

The reason I am trying to get to the bottom of this is that Denso is a massive in vehicles, they are part of the Toyota group, and would be great to have onboard.

Another reason I am thinking Denso are using Quadric to work with Akida is that they are both under Megachips banner and Denso is Japanese as well.

If anyone with technical knowledge wants to tell me I’m wrong I’m all ears, as I’m trying to learn.

By the way: great article about Mercedes Benz today FF. Fantastic news!

Cheers,

Hi SG,

We looked at Quadric a few pages ago: (There are a few more posts following this)

https://thestockexchange.com.au/threads/brn-discussion-2022.1/page-281#post-32194

Hi JB,

Quadric make what they call a "Supercomputer" which integrates the functions of CPU/GPT/AI accelerator using a modified CNN for image classification.

Its basic unit seems to be their Vortex ALU.

It won't be used for doorbells.

The White Paper 2019​

May, 2019

https://www.quadric.io/post/the-white-paper

we founded Quadric to build a product that brings server-class performance to the edge.

1647508064624.png



By 2015, researchers were proving classification results on the Imagenet challenge that exceeded human error rates. One such advanced CNN network architecture, RESNET50 strikes a balance between total computational network complexity and error rate.

1647508699958.png


####################################################################################

https://brainchipinc.com/wp-content...brief_6-How-BrainChip-is-Changing-AI_v1.2.pdf
The Akida Event-Based Neural Processor
The Akida event-based neural processor is a fundamentally different approach that breaks the linear relationship between high power consumption and performance seen in traditional accelerators. The Akida processor is 10x to 30x more energy-efficient than its nearest competitor for inferencing on industry-standard benchmarks such as MobileNet and Google Keyword Spotting DNNs, and is easy to use. Trained on MobileNet’s Imagenet 1000 data set, the Akida neural processor can classify all 1.2 million images, and 1,000 classes, at 30 frames per second within a power budget of just 156 milliwatts in 28nm, compared to several watts for a Google Edge TPU. Audio keyword recognition using the Google keyword database runs at an extremely low power of 150 microwatts
.

30 frames per second in 156 mW = 192 frames per second per Watt. (This is the 2019 Akida, not the improved 2021 commercial version - to be fair, the Quadric figures are from a 2019 white paper). So Quadric is in the same ballpark, but Akida has made home base, while Quadric just making 3rd.

The main difference is that Quadric is a Frankenstein amalgam of CPU, GPU and AI accelerator. Their AI accelerator uses ResNet50, an improved CNN arrangement which enables skip connections when intermediate layers are not required.

1647513839516.png



Also remember Akida's performance at key word spotting:
1647510358747.png


IPS = inferences (identified words) per second

Opinions and research expressed are my own as an unlicensed individual. External links are not endorsed. Do your own research or consult a licensed financial advisor.

Stock Disclosure:
Held. Edit stock disclosure
Reply
Report Edit Delete
 
  • Like
  • Fire
Reactions: 16 users
Hi SG,

We looked at Quadric a few pages ago: (There are a few more posts following this)

https://thestockexchange.com.au/threads/brn-discussion-2022.1/page-281#post-32194

Hi JB,

Quadric make what they call a "Supercomputer" which integrates the functions of CPU/GPT/AI accelerator using a modified CNN for image classification.

Its basic unit seems to be their Vortex ALU.

It won't be used for doorbells.

The White Paper 2019​

May, 2019

https://www.quadric.io/post/the-white-paper

we founded Quadric to build a product that brings server-class performance to the edge.

1647508064624.png



By 2015, researchers were proving classification results on the Imagenet challenge that exceeded human error rates. One such advanced CNN network architecture, RESNET50 strikes a balance between total computational network complexity and error rate.

1647508699958.png


####################################################################################

https://brainchipinc.com/wp-content...brief_6-How-BrainChip-is-Changing-AI_v1.2.pdf
The Akida Event-Based Neural Processor
The Akida event-based neural processor is a fundamentally different approach that breaks the linear relationship between high power consumption and performance seen in traditional accelerators. The Akida processor is 10x to 30x more energy-efficient than its nearest competitor for inferencing on industry-standard benchmarks such as MobileNet and Google Keyword Spotting DNNs, and is easy to use. Trained on MobileNet’s Imagenet 1000 data set, the Akida neural processor can classify all 1.2 million images, and 1,000 classes, at 30 frames per second within a power budget of just 156 milliwatts in 28nm, compared to several watts for a Google Edge TPU. Audio keyword recognition using the Google keyword database runs at an extremely low power of 150 microwatts
.

30 frames per second in 156 mW = 192 frames per second per Watt. (This is the 2019 Akida, not the improved 2021 commercial version - to be fair, the Quadric figures are from a 2019 white paper). So Quadric is in the same ballpark, but Akida has made home base, while Quadric just making 3rd.

The main difference is that Quadric is a Frankenstein amalgam of CPU, GPU and AI accelerator. Their AI accelerator uses ResNet50, an improved CNN arrangement which enables skip connections when intermediate layers are not required.

1647513839516.png



Also remember Akida's performance at key word spotting:
1647510358747.png


IPS = inferences (identified words) per second

Opinions and research expressed are my own as an unlicensed individual. External links are not endorsed. Do your own research or consult a licensed financial advisor.

Stock Disclosure:
Held. Edit stock disclosure
Reply
Report Edit Delete
Thanks for the reply Dio.

I don’t understand a lot of the terminology but I guess you‘re saying they are heavy on power use. If that’s the case why are Denso using them as the are car parts manufacturers and I thought we were trying to reduce power usage to improve efficiency?

How would they be used with Akida in a vehicle? Or because Denso have chosen Quadric that rules Akida out?

Sorry for asking silly questions but I can’t get my head around some of these concepts.

If the answers too convoluted that’s fine. It’ll all come out in the wash one day regardless and I’m confident of Brainchips profitability. I‘m just curious and hate not understanding things.

Cheers
 
  • Like
Reactions: 4 users

Diogenese

Top 20
Thanks for the reply Dio.

I don’t understand a lot of the terminology but I guess you‘re saying they are heavy on power use. If that’s the case why are Denso using them as the are car parts manufacturers and I thought we were trying to reduce power usage to improve efficiency?

How would they be used with Akida in a vehicle? Or because Denso have chosen Quadric that rules Akida out?

Sorry for asking silly questions but I can’t get my head around some of these concepts.

If the answers too convoluted that’s fine. It’ll all come out in the wash one day regardless and I’m confident of Brainchips profitability. I‘m just curious and hate not understanding things.

Cheers
On the figures they quote, they are quite good on power usage compared to the also-rans. They quote ~ 130 images/second/watt.

Akida (engineering sample) was 192 images/second/watt. Akida 1000 (commercial) has been said to have better performance than expected, but I haven't seen the figures. So the old Akida seems to be about 50% better than Quadric on this measure, assuming they were both tested on the same dataset of images.

They are using ALUs (arithmetic logic units), distributed memory, and ResNet50, which is a modified CNN arrangement with some form of data compression, if I recall correctly.

It seems to be directed to fairly heavy lifting, but I suppose they envisage it as a one-stop shop for automotive.
"we founded Quadric to build a product that brings server-class performance to the edge."

I don't see them as a danger re Mercedes, who are fairly committed to their do-it-yourself (with a couple of mates) system.

Where we may cross swords would be in the 64-chip Akida array type applications.
 
Last edited:
  • Like
  • Fire
Reactions: 16 users
On the figures they quote, they are quite good on power usage compared to the also-rans. They quote ~ 130 images/second/watt.

Akida (engineering sample) was 192 images/second/watt. Akida 1000 (commercial) has been said to have better performance than expected, but I haven't seen the figures. So the old Akida seems to be about 50% better than Quadric on this measure, assuming they were both tested on the same dataset of images.

They are using ALUs (arithmetic logic units), distributed memory, and ResNet50, which is a modified CNN arrangement with some form of data compression, if I recall correctly.
Thanks Dio.

I am laughing because you are speaking on so much a higher level than I can grasp.

My knowledge of computers how to switch it off for a minute and then turn it back on to reboot it!

It’s not your fault I can’t grasp the concept. Thanks for humouring me.

For example I don’t know what an ALU is, I’m guessing it uses maths to compute its decisions.

But it’s like the scene in the Simpson’s when Marge is talking and Homers thought bubble has a chimpanzee clanging symbols!

It’ll come out one day what Denso are doing when they release their product and I’ll go: “Ohhh, now I get it.”

Cheers!
 
  • Like
  • Haha
Reactions: 12 users
Thanks Dio.

I am laughing because you are speaking on so much a higher level than I can grasp.

My knowledge of computers how to switch it off for a minute and then turn it back on to reboot it!

It’s not your fault I can’t grasp the concept. Thanks for humouring me.

For example I don’t know what an ALU is, I’m guessing it uses maths to compute its decisions.

But it’s like the scene in the Simpson’s when Marge is talking and Homers thought bubble has a chimpanzee clanging symbols!

It’ll come out one day what Denso are doing when they release their product and I’ll go: “Ohhh, now I get it.”

Cheers!
1647946399202.gif
 
  • Haha
  • Like
Reactions: 19 users

Diogenese

Top 20
Thanks Dio.

I am laughing because you are speaking on so much a higher level than I can grasp.

My knowledge of computers how to switch it off for a minute and then turn it back on to reboot it!

It’s not your fault I can’t grasp the concept. Thanks for humouring me.

For example I don’t know what an ALU is, I’m guessing it uses maths to compute its decisions.

But it’s like the scene in the Simpson’s when Marge is talking and Homers thought bubble has a chimpanzee clanging symbols!

It’ll come out one day what Denso are doing when they release their product and I’ll go: “Ohhh, now I get it.”

Cheers!
The ALU is the bit where the number-crunching happens in the old fashioned von Neumann (vN) CPU. It adds two multi-bit binary numbers - usually between 8-bits and up to 128 bits or more, and can implement multiplication.

As far as I understand it, Quadric has distributed ALUs among numerous cells each with their own memory to avoid the (peak hour) vN bottleneck which occurs when all the data is stored in a large memory and must be transferred over a common 32 line (for example) cable (bus) to the single processor.

The operative interconnection of the processors can be reconfigured electronically.

########################################################### (SG-proof fence)

This is the patent for Quadric processor architecture:

US10474398B2 Machine perception and dense algorithm integrated circuit

1647949425187.png




A circuit that includes a plurality of array cores (110),
each array core of the plurality of array cores comprising:
a plurality of distinct data processing circuits; and

a data queue register file (130);

a plurality of border cores (120), each border core of the plurality of border cores comprising: at least a register file, wherein:
at least a subset of the plurality of border cores encompasses a periphery of a first subset of the plurality of array cores; and
[ii] a combination of the plurality of array cores and the plurality of border cores define an integrated circuit array
.

A good proportion of the array is used by the "border" processors (cores) 120 in organizing the data inputs/outputs of the working processors 110.
 
  • Like
Reactions: 12 users

Foxdog

Regular
Ah yeah, thanks Dio - that clears it up 🤔
 
  • Haha
  • Like
  • Fire
Reactions: 12 users

Dallas

Regular
Vielleicht interessant. Hello ID Spracherkennung 95 Prozent genau 🦧🤖😁
 
Top Bottom