BRN Discussion Ongoing

[Shadow59]

Just been busy in confo meetings about the various colour schemes and numbers of cup holders in our lambo’s and EQ’s on order so we don’t clash at the AGM and the advisability of swimming pools on yachts? The big question of course is, does one stock them with sturgeon or simply have the caviar flown in fresh daily?

Definitely go with the flown in. That's what the helipad is for.

 

[HopalongPetrovski]

Definitely go with the flown in. That's what the helipad is for.

I know dear, but it plays merry havoc with Donald’s hair!

 

[Diogenese]

Thanks Rise for the heads up. Looking forward to this benchmarking! The momentum of the Brainchip low power smarts is trending nicely!

I searched for: low power edge tinyml benchmarking best practices. Lots of info on this, deep stuff.

The Brainchip team will put it in a format that everyone will understand. Should be awesome.

https://www.wevolver.com/article/the-art-and-science-of-benchmarking-neural-network-architectures

The variety of available neural network architectures, configurations, and performance-related parameters makes it challenging to compare alternative ML/DL products and services. To facilitate this, engineers often turn to proven benchmarks for compute intense and data intense systems. For example:

• The Standard Performance Evaluation Corporation (SPEC) family of benchmarks focuses on computational workloads over different computing architectures including cloud-based systems.
• The LINPACK Benchmarks provide measures of a computer’s floating-point rate of execution i.e., they focus on floating-point computing power.
• The TPC consortium includes industry leaders in computing systems and datasets, which have provided a suite of benchmarks for stress testing the transaction processing and data warehousing capabilities of data intensive systems.
• The High Performance Conjugate Gradients (HPCG) benchmark provides novel metrics that enable the evaluation and ranking of High Performance Computing (HPC) systems.

These benchmarks aid in evaluating different components of ML/DL systems, which are typically deployed over computationally intensive platforms. However, they are overly focused on conventional computational workloads and hence not sufficient for training and executing deep learning tasks that are typically more complex.

Hi stuart,

The primary measure of the performance of CPU/GPU is TOPS - tera operations per second - in other words how much power they can burn per second. This does not tell you how long it takes to perform a task or how much power it burns doing it. Such yardsticks are not applicable to Akida which is designed to minimize power consumption and improve speed.

A first meaningful measure of Akida's performance would be inferences per second (IPS) or frames per second (FPS), where the inferences/frames are tested against a standardized set of data. This tells you how fast the chip can perform a task. The second measure would be IPS per Watt or FPS per Watt.

IPS/FPS would clearly demonstrate how much faster Akida is compared to a software CNN run on CPU/GPU, and IPS/FPS per Watt would clearly demonstrate how much power the von Neumann computers use compared to Akida in performing the same tasks. Clearly, the amount of power used is a critical measure for edge devices, and the time is important for applications requiring real time responses.

 

[Murphy]

Gee it’s been quiet in here the last few days , don’t tell me some of you have been made to sign NDA’s

Most probably don't want to get another warning of banning for being off-topic!

 

[Diogenese]

Most probably don't want to get another warning of banning for being off-topic!

Don't worry - @Dreadbot is not an ogre.

 

[Steve10]

Market flat waiting for RBA interest rate decision. If they pause could be start of short squeeze. Toss a coin time.

 

[Dhm]

"The company also announced a new technology called micro tile inferencing, which breaks a neural network into many small pieces so that it can be processed all at once, rather than layer by layer, and very efficiently, saving a great deal of power."


All right - so now you have my attention:

Qualcomm:
WO2023049655A1 TRANSFORMER-BASED ARCHITECTURE FOR TRANSFORM CODING OF MEDIA 2021-09-27

[This is so hot off the press, the ink isn't dry ... 2023-03-30 ... and the specification is not available on line via Espacenet, but you may find it at google Patents]

Systems and techniques are described herein for processing media data using a neural network system. For instance, a process can include obtaining a latent representation of a frame of encoded image data and generating, by a plurality of decoder transformer layers of a decoder sub-network using the latent representation of the frame of encoded image data as input, a frame of decoded image data. At least one decoder transformer layer of the plurality of decoder transformer layers includes: one or more transformer blocks for generating one or more patches of features and determine self-attention locally within one or more window partitions and shifted window partitions applied over the one or more patches; and a patch un-merging engine for decreasing a respective size of each patch of the one or more patches.



View attachment 33566


Does this not sound a bit like our ViT/TENNs of Gen2?


https://brainchip.com/brainchip-introduces-second-generation-akida-platform/

The second-generation of Akida now includes Temporal Event Based Neural Nets (TENN) spatial-temporal convolutions that supercharge the processing of raw time-continuous streaming data, such as video analytics, target tracking, audio classification, analysis of MRI and CT scans for vital signs prediction, and time series analytics used in forecasting, and predictive maintenance. These capabilities are critically needed in industrial, automotive, digital health, smart home and smart city applications. The TENNs allow for radically simpler implementations by consuming raw data directly from sensors – drastically reduces model size and operations performed, while maintaining very high accuracy. This can shrink design cycles and dramatically lower the cost of development.
Another addition to the second generation of Akida is Vision Transformers (ViT) acceleration, a leading edge neural network that has been shown to perform extremely well on various computer vision tasks, such as image classification, object detection, and semantic segmentation. This powerful acceleration, combined with Akida’s ability to process multiple layers simultaneously and hardware support for skip connections, allows it to self-manage the execution of complex networks like RESNET-50 completely in the neural processor without CPU intervention and minimizes system load.

... but remember what Ella says.

Hi @Diogenese thank you for your optimistic assessment that Qualcomm maybe, just maybe looking at us within their patent lodgement. A fair bit of the jargon that the Qualcomm patent mentions goes both over my head and through my two ears.

I remember when Brainchip recently released Akida 2.0 there was much admiration by some experts about the TENN enhancement and I assumed it was us as first to market . So when Qualcomm's patent got your attention my heart beat a little faster

Are you able to unpack your thoughts about this just a little more? I don't know whether to take my heart pills or open up some champagne!

 

[Steve10]

At its meeting today, the Board decided to leave the cash rate target unchanged at 3.60 per cent and the interest rate on Exchange Settlement balances unchanged at 3.50 per cent.

Let's see what happens now.

 

[Damo4]

At its meeting today, the Board decided to leave the cash rate target unchanged at 3.60 per cent and the interest rate on Exchange Settlement balances unchanged at 3.50 per cent.

Let's see what happens now.

Half expecting my bank to still send me a letter advising my mortgage has increased hahaha

 

[Diogenese]

Hi @Diogenese thank you for your optimistic assessment that Qualcomm maybe, just maybe looking at us within their patent lodgement. A fair bit of the jargon that the Qualcomm patent mentions goes both over my head and through my two ears.

I remember when Brainchip recently released Akida 2.0 there was much admiration by some experts about the TENN enhancement and I assumed it was us as first to market . So when Qualcomm's patent got your attention my heart beat a little faster

Are you able to unpack your thoughts about this just a little more? I don't know whether to take my heart pills or open up some champagne!

Hi Dhm,

My comment about Ella refers to her song " 'tain't what you do, it's the way that you do it" rather than that spurious ChatGPT mash up @Bravo posted.

Until our patent for the TNN/AiT features, which was foreshadowed in a BRN PR, is published, I can't compare the way QC do it with the way Akida Gen 2 does it. I'm sure the BRM Patent Attorneys will be pouring over the QC patent.

That said, Qualcomm do use MACs, so their architecture id different from Akida, so, at the circuit level, their system is different from ours. The question is, what does their patent encompass?

I'd keep the digitalin handy and leave the champagne to chill a little longer.

 

[Dhm]

Hi Dhm,

My comment about Ella refers to her song " 'tain't what you do, it's the way that you do it" rather than that spurious ChatGPT mash up @Bravo posted.

Until our patent for the TNN/AiT features, which was foreshadowed in a BRN PR, is published, I can't compare the way QC do it with the way Akida Gen 2 does it. I'm sure the BRM Patent Attorneys will be pouring over the QC patent.

That said, Qualcomm do use MACs, so their architecture id different from Akida, so, at the circuit level, their system is different from ours. The question is, what does their patent encompass?

I'd keep the digitalin handy and leave the champagne to chill a little longer.

Darn, I was thirsty. Can I use the champagne to wash down the pills?

 

[Diogenese]

Darn, I was thirsty. Can I use the champagne to wash down the pills?

What else would you use?

 

[Fullmoonfever]

Or did you mean tips? Or I'm just a bit slow and that's a play on words?
@Tothemoon24 was about to say the same thing.
The quiet part that is.
Hmm this guy is definitely not a fan
View attachment 33588 View attachment 33589

Was having a look at that ISO to see how hard it is....usually are

This is from the Synopsys website and the ISO is not an actual legal requirement...yet... but a best practice in reality.

Is ISO 26262 Required?
ISO 26262 is not required by law, but many car makers and suppliers follow it to show their commitment to safety and to improve their products safety. Sometimes customers and regulators might require them to prove they follow the standard. But even if it's not required, it's still considered a good practice and following it can improve the safety of car electronic systems and show customers, regulators and end users the company's commitment to safety.

Gives a pretty detailed run down on what it is etc.

What is ISO 26262 Functional Safety Standard? | Synopsys Automotive

ISO 26262 mandates a functional safety development process that automotive OEMs and suppliers must follow and document to have their devices qualified to run inside commercial vehicles. It outlines a risk classification system (ASILs) and aims to reduce possible hazards caused by the...

www.synopsys.com

When searching for the ISO I also saw Jan 23 article in Embedded about T2M getting the ISO for certain IP blocks...not us unfortunately haha

Complete ASIL-A,B,C,D and ISO26262 Certified IP Cores from T2M - Embedded Computing Design

T2M released its certified ISO 26262 and AEC-Q100 Automotive Grade Silicon Proven IP Cores as small as 7nm for must have connectivity in automotive applications. T2M supports solutions needing CAN Bus IP Cores and LIN Bus IP Cores certified with ASIL (Automotive Safety Integrity Level) packages...

embeddedcomputing.com

Complete ASIL-A,B,C,D and ISO26262 Certified IP Cores from T2M​

By Chad Cox​

ASSOCIATE EDITOR
EMBEDDED COMPUTING DESIGN
January 24, 2023

But it did make think of T2M as recalled reading something once that BRN engaged them in the early days for Akida.

I presume they still have a relationship / engagement to market and sell Akida as on the Chip Estimate website they had T2M still flagged but maybe outdated?...though is for the IP so would have to be from when we started moving to IP I presume.

Wonder what reach T2M has?

Semiconductor Silicon Proven IP Cores Company - T2M IP

Looking for leadership, state of the art, silicon proven, semiconductor IP & Software for your next SoC

t-2-m.com

T2M Akida-Scalable Neural Network AI Silicon IP | ChipEstimate.com IP Catalog

AKIDA - Neuromorphic Computing

www.chipestimate.com

Log In

Akida-Scalable Neural Network AI Silicon IP

IP Preview

Name:Akida-Scalable Neural Network AI Silicon IP

Provider:T2M

Description:AKIDA - Neuromorphic Computing

Overview:The Akida Neuromorphic IP is the first neuromorphic IP available in the market. Inspired by the biological function of neurons but engineered on a digital logic process, this event-based spiking ...

Category:IP Catalog : Digital Core IP :

Processors : Other

Additional data available! Portability, process node, maturity, features, and more can be viewed by logging in with your ChipEstimate.com account.

 

[Dhm]

What else would you use?

I like your thinking. Hows about we merge Chemist Warehouse and Dan Murphys?

 

[Steve10]

Akida gestation phase at the moment.

Will it be a baby Qualcomm, STM, NXP or TI? How about twins or triplets?

Waiting anxiously pacing up & down.

 

[alwaysgreen]

At its meeting today, the Board decided to leave the cash rate target unchanged at 3.60 per cent and the interest rate on Exchange Settlement balances unchanged at 3.50 per cent.

Let's see what happens now.

Unfortunately that doesn't seem to have changed sentiment.

Going to need more than a rates pause to start a short squeeze.

 

[Steve10]

Interesting chart showing where funds are going to AI start ups. BRN near the bottom.

Sorry this next one is a bit blurry. SiFive on the list with $365.5M funding.

Couldn't find the September 2022 update chart. It's in following video at 12min 25sec.



Intro -
[0:00] The role of AI Hardware -
[0:40] Talk Overview -
[1:50] Intro to me #selfpromotion -
[2:45] AI Hardware: A New Era
Types of AI Hardware -
[4:37] CPUs -
[5:43] GPUs -
[6:56] FPGAs -
[7:37] ASICs
Types of Architectures -
[8:00] Systolic Arrays and Google TPU -
[9:21] CGRAs -
[10:45] Parallel Pipeline / Dataflow
The AI Hardware Startup Market -
[12:04] Current investment levels -
[13:36] @SambaNovaSystems Cardinal SN10/SN30 -
[14:13] @CerebrasSystems Wafer Scale Engine -
[15:36] @Graphcore IPU and Bow -
[17:11] Honorable Mentions: @GroqInc , Lightmatter , Rain, Tachyum
Conclusions -
[18:00] How to Scale -
[19:12] @tenstorrentinc CTO Jim Keller Quote

 

[Damo4]

Interesting chart showing where funds are going to AI start ups. BRN near the bottom.

View attachment 33607

Sorry this next one is a bit blurry. SiFive on the list with $365.5M funding.

View attachment 33608


Couldn't find the September 2022 update chart. It's in following video at 12min 25sec.



Intro -
[0:00] The role of AI Hardware -
[0:40] Talk Overview -
[1:50] Intro to me #selfpromotion -
[2:45] AI Hardware: A New Era Types of AI Hardware -
[4:37] CPUs -
[5:43] GPUs -
[6:56] FPGAs -
[7:37] ASICs Types of Architectures -
[8:00] Systolic Arrays and Google TPU -
[9:21] CGRAs -
[10:45] Parallel Pipeline / Dataflow The AI Hardware Startup Market -
[12:04] Current investment levels -
[13:36] @SambaNovaSystems Cardinal SN10/SN30 -
[14:13] @CerebrasSystems Wafer Scale Engine -
[15:36] @Graphcore IPU and Bow -
[17:11] Honorable Mentions: @GroqInc , Lightmatter , Rain, Tachyum Conclusions -
[18:00] How to Scale -
[19:12] @tenstorrentinc CTO Jim Keller Quote

Where does the $27.8M figure come from? Is it based on the obligation to LDA?
I'm guessing most of the others are funded in such a way due to not being a publicly listed company?
Sorry for all the questions, just don't understand the chart.

Edit: or have I completely missed the point and this is just cash reserves?

 

[Steve10]

Where does the $27.8M figure come from? Is it based on the obligation to LDA?
I'm guessing most of the others are funded in such a way due to not being a publicly listed company?
Sorry for all the questions, just don't understand the chart.

Not sure. Might be better to send him an email. Will be good to have this guy review Akida.

 

[Space Cadet]

Was having a look at that ISO to see how hard it is....usually are

This is from the Synopsys website and the ISO is not an actual legal requirement...yet... but a best practice in reality.

Is ISO 26262 Required?
ISO 26262 is not required by law, but many car makers and suppliers follow it to show their commitment to safety and to improve their products safety. Sometimes customers and regulators might require them to prove they follow the standard. But even if it's not required, it's still considered a good practice and following it can improve the safety of car electronic systems and show customers, regulators and end users the company's commitment to safety.

Gives a pretty detailed run down on what it is etc.

What is ISO 26262 Functional Safety Standard? | Synopsys Automotive

ISO 26262 mandates a functional safety development process that automotive OEMs and suppliers must follow and document to have their devices qualified to run inside commercial vehicles. It outlines a risk classification system (ASILs) and aims to reduce possible hazards caused by the...

www.synopsys.com

When searching for the ISO I also saw Jan 23 article in Embedded about T2M getting the ISO for certain IP blocks...not us unfortunately haha

Complete ASIL-A,B,C,D and ISO26262 Certified IP Cores from T2M - Embedded Computing Design

T2M released its certified ISO 26262 and AEC-Q100 Automotive Grade Silicon Proven IP Cores as small as 7nm for must have connectivity in automotive applications. T2M supports solutions needing CAN Bus IP Cores and LIN Bus IP Cores certified with ASIL (Automotive Safety Integrity Level) packages...

embeddedcomputing.com

Complete ASIL-A,B,C,D and ISO26262 Certified IP Cores from T2M​

By Chad Cox​

ASSOCIATE EDITOR
EMBEDDED COMPUTING DESIGN
January 24, 2023

But it did make think of T2M as recalled reading something once that BRN engaged them in the early days for Akida.

I presume they still have a relationship / engagement to market and sell Akida as on the Chip Estimate website they had T2M still flagged but maybe outdated?...though is for the IP so would have to be from when we started moving to IP I presume.

Wonder what reach T2M has?

Semiconductor Silicon Proven IP Cores Company - T2M IP

Looking for leadership, state of the art, silicon proven, semiconductor IP & Software for your next SoC

t-2-m.com

T2M Akida-Scalable Neural Network AI Silicon IP | ChipEstimate.com IP Catalog

AKIDA - Neuromorphic Computing

www.chipestimate.com

Log In

Akida-Scalable Neural Network AI Silicon IP

IP Preview

Name:Akida-Scalable Neural Network AI Silicon IP

Provider:T2M

Description:AKIDA - Neuromorphic Computing

Overview:The Akida Neuromorphic IP is the first neuromorphic IP available in the market. Inspired by the biological function of neurons but engineered on a digital logic process, this event-based spiking ...

Category:IP Catalog : Digital Core IP :

Processors : Other

Additional data available! Portability, process node, maturity, features, and more can be viewed by logging in with your ChipEstimate.com account.

The good thing is we supply IP. It's up to the others to meet ISO. Saves us a shit load of money.

SC