BRN Discussion Ongoing

[albycam]

@Diogenese - Jem Davis has turned up as lead NED at Literal- Labs AI - https://www.literal-labs.ai they are making some pretty aggressive claims on power saving.. If you have time would you care to comment on this ....

"Literal Labs applies a streamlined and efficient approach to AI that is faster, explainable, and up to 10,000X more energy efficient than today’s neural networks. Similar to NNs in that customers train a dataset, Literal Labs trains Tsetlin machine models specific to customer datasets. Our approach results in an optimised machine model that is then deployed onto the target hardware. The Tsetlin machine model can be deployed as software only or can be accelerated using Literal Labs accelerators. Our benchmarking shows we can achieve 250X faster inferencing than XG Boost using software only, and up to 1,000X faster and up to 10,000X less energy consumption when using hardware acceleration. The company was spun out of Newcastle University by world leaders in Tsetlin machine Dr. Alex Yakovlev and Dr. Rishad Shafik, and led by former Arm CPU division VP and semiconductor startup founder Noel Hurley. "

 

[BrainShit]

This should be announced via the ASX as it’s a commercial product with a $$ amount

They did... in Sep 2023. And the Edge Box is at least a quarter late.

 

[DingoBorat]

@Diogenese - Jem Davis has turned up as lead NED at Literal- Labs AI - https://www.literal-labs.ai they are making some pretty aggressive claims on power saving.. If you have time would you care to comment on this ....

"Literal Labs applies a streamlined and efficient approach to AI that is faster, explainable, and up to 10,000X more energy efficient than today’s neural networks. Similar to NNs in that customers train a dataset, Literal Labs trains Tsetlin machine models specific to customer datasets. Our approach results in an optimised machine model that is then deployed onto the target hardware. The Tsetlin machine model can be deployed as software only or can be accelerated using Literal Labs accelerators. Our benchmarking shows we can achieve 250X faster inferencing than XG Boost using software only, and up to 1,000X faster and up to 10,000X less energy consumption when using hardware acceleration. The company was spun out of Newcastle University by world leaders in Tsetlin machine Dr. Alex Yakovlev and Dr. Rishad Shafik, and led by former Arm CPU division VP and semiconductor startup founder Noel Hurley. "

I can't say anything about their tech, but their aggressive claims mean nothing.

"up to 1,000X faster and up to 10,000X less energy consumption"

The "up to" part is a widely used marketing/promotional gimic, like when you go to a sale, that says "Up to 70% off" and the only thing you can find, marked off that much, is something nobody wants..

10 x better, is still "up to" 10000 times.

Without context, their words mean nothing.

 

[cosors]

Yes, I also like these superlatives when they come from others. For repetition because it's so nice just before the IFS:

ARM:
Arm’s Cortex-M coupled with BrainChip’s Akida delivers unparalleled system performance.
https://armkeil.blob.core.windows.net/developer/Files/pdf/ai-ecosystem-catalogue/BrainChip_Arm-AI-Partner-Solution-Brief-Template-1.pdf

U - nrivaled
U - nparalleled
U - biquitous

Akida is so uuu!
______________________
I would like to add this one
__________________________________________________________________Akida neuron fabric integrates into any Arm

CPU.

Wait a minute @Dio. ...any ARM CPU ... architecture. Am I supposed to take ARM's statement seriously and it doesn't depend on the 22nm (GF) or any nm at all and that's just the qualification process?

IFS

 

[DingoBorat]

Wait a minute @Dio. ...any ARM CPU ... architecture. Am I supposed to take ARM's statement seriously and it doesn't depend on the 22nm (GF) or any nm at all and that's just the qualification process?

IFS

Obviously I'm not Diogenese ..

But my recollection, is that Peter said AKIDA could go down to 7nm process node.

But I'm guessing that's because, that was the smallest viable process node, at that time?

I'm also guessing, that the "neuron fabric" whatever that actually is, would be the limiting factor..

But since it's a "digital" design, there should be no limits? ...

The new smaller process nodes, are completely different "processes" too..

 

[Fact Finder]

Wait a minute @Dio. ...any ARM CPU ... architecture. Am I supposed to take ARM's statement seriously and it doesn't depend on the 22nm (GF) or any nm at all and that's just the qualification process?

IFS

It has been said and written many times since Anil Mankar made this claim after the ARM partnership was announced.

AKD1000 technology IP is process and sensor agnostic and compatible across the major foundries which includes TSMC, Global Foundries, Intel Foundry Services and Renesas Foundries.

It has also been shown to be compatible with the Vorago Hardsil process.

My opinion only DYOR
Fact Finder

 

[buena suerte :-)]

BrainChip Begins Accepting Pre-Orders of the Akida Edge AI Box​

Laguna Hills, Calif. – February 20, 2024 – BrainChip Holdings Ltd (ASX: BRN, OTCQX: BRCHF, ADR: BCHPY), the world’s first commercial producer of ultra-low power, fully digital, event-based, neuromorphic AI, today re-opened its product shopping portal and will be accepting pre-orders of its recently announced Akida Edge AI Box. The Akida™ powered Edge AI box is expected to help customers accelerate high-performance AI applications at the Edge in challenging environments where energy-efficiency is essential and cloud connectivity is not guaranteed.

Built in collaboration with VVDN Technologies, the Akida Edge AI Box is designed to meet the demanding needs of retail and security, Smart City, automotive, transportation and industrial applications. The device combines a powerful quad-core CPU platform with Akida AI accelerators to provide a huge boost in AI performance. The compact, light Edge box is cost- effective and versatile with built in ethernet and Wi-Fi connectivity, HD display support, extensible storage, and USB interfaces. BrainChip and VVDN are finalizing the set of AI applications that will run out of the box. With the ability to personalize and learn on device, the box can be customized per application and per user without the need for cloud support, enhancing privacy and security.

“The Akida Edge AI Box is ideally suited to provide the low latency and high throughput processing with ultra-low power consumption – a necessity for the next generation of smart Edge devices,” said Sean Hehir, CEO of BrainChip. “We are excited to officially launch pre- orders of the Akida Edge AI Box and bring this groundbreaking technology to market to empower customers in developing and deploying intelligent, secure and customized devices and services for multi-sensor environments in real time.”

The Akida Edge AI Box starts at $799 and can be pre-ordered by visiting https://shop.brainchipinc.com/.

All looking extremely positive moving forward


“The Akida Edge AI Box is ideally suited to provide the low latency and high throughput processing with ultra-low power consumption – a necessity for the next generation of smart Edge devices,” said Sean Hehir, CEO of BrainChip.

 

[cosors]

That is interesting! I'm off to bed now and get my thoughts straight. See you tomorrow and have a good start to the day! Thanks to you both!

____
...akida at intel's 1.8nm announced next day... snoring

 

[JB49]

All looking extremely positive moving forward


“The Akida Edge AI Box is ideally suited to provide the low latency and high throughput processing with ultra-low power consumption – a necessity for the next generation of smart Edge devices,” said Sean Hehir, CEO of BrainChip.
View attachment 57355

I guess it wasn't sold out as advertised on the website last week. It just wasn't for sale yet.

 

[Food4 thought]

Pre market looking like the sellers are going to try to flood the market again

Need more buyers
Look behind your couch for some change and buy more shares

 

[Esq.111]

Also look under the coffee table ..... I often find a little cash there.

Esq.

 

[buena suerte :-)]

I guess it wasn't sold out as advertised on the website last week. It just wasn't for sale yet.

Looks that way.. I'm thinking that TEAM BRN knew that they were going to be available very Soon and "Sold out" looked much more positive than..
Not available! But now they are most certainly....... FOR SALE!!

Lets go BRN

 

[7für7]

Pre market looking like the sellers are going to try to flood the market again

Need more buyers
Look behind your couch for some change and buy more shares

I said recently……. COCKROACHES!!!!!!

 

[Esq.111]

Good Morning Chippers ,

Now I don't profess to be a charting wizz.....

The three day chart on 1,3,5,15 min & particularly the one hour duration looks Very promising on volume flow for a jolly good surge around midday today.

Not Financial Advice.

Regards,
Esq.

 

[Fact Finder]

That is interesting! I'm off to bed now and get my thoughts straight. See you tomorrow and have a good start to the day! Thanks to you both!

____
...akida at intel's 1.8nm announced next day... snoring

Hi Cosors

The following is a very good guide to takeovers in Australia:

https://www.allens.com.au/globalassets/pdfs/sectors-services/ma/takeovers-in-australia-handbook_feb2023.pdf

My opinion only DYOR
Fact Finder

 

[White Horse]

Obviously I'm not Diogenese ..

But my recollection, is that Peter said AKIDA could go down to 7nm process node.

But I'm guessing that's because, that was the smallest viable process node, at that time?

I'm also guessing, that the "neuron fabric" whatever that actually is, would be the limiting factor..

But since it's a "digital" design, there should be no limits? ...

The new smaller process nodes, are completely different "processes" too..

Hi DB,
Aren't Socionext already using our IP in 7nm.
I think there was a lot of press about it around CES 2023.

 

[Learning]

Thanks for posting this @Learning. Why would BrainChip bother to create a post for another companies PMU on their LinkedIn account? A performance monitoring unit of all things. Is that something that Akida would be used for? Or is there more to it?
Has BrainChip been helping Tachyum with their soon to be released beta version of its prodigy Processor. I haven’t the foggiest but a look at the article below shows that Tachyum have been busy advancing their Prodigee offering.

CEO of Tachyum. “With each new enrichment we are able to incorporate into Prodigy’s software stack, we magnify the ability of a Prodigy platform ready to revolutionize the world.”

Tachyum Upgrades Software Package in Advance of Beta Release of the Prodigy Universal Processor​

LAS VEGAS, February 14, 2024 – Tachyum® today announced that it has upgraded the software stack for the Prodigy® Universal Processor before the anticipated launch of its beta version around the end of quarter. Quality completion of Prodigy’s software stack is a key component as the company continues to advance towards chip production and distribution.
Tachyum software engineers have worked hard to enable the full potential of Prodigy with the development of an ecosystem of applications, system software, frameworks and libraries that are ported to run natively on Prodigy hardware. Once the software package completes its testing and runs cleanly on the FPGA, the company can fully transition to advancing the Universal Processor into production.
The Prodigy software distribution is a completely integrated software stack and package that is ready for deployment “as is.” It is available as a single pre-installed image for Tachyum’s early adopters and customers. Applications have been tested to work out of the box so that customers can immediately start using the reference design. If users encounter any issues during deployment, the software can be quickly and easily restored to its original image.
Included in the software distribution package as part of alpha testing are:

• Latest versions of the QEMU emulator 8.2
• GCC 13.2 (GNU Compiler Collection) and glibc 2.39 (GNU C Library)
• Linux 6.6 LTS (Long Term Support), which contains a large number of changes, updates and improvements

The company also announced plans to switch to the LLVM 18 release once it is available to download. LLVM plays a significant role in every major AI framework. Additionally, it is in the process of adding RAS (Reliability Accessibility Serviceability) in the form of an EDAC (Error Detection and Correction) driver in the next few weeks. Based on customer requests for server applications Tachyum agreed to add bootable SSD RAID next quarter to its UEFI.
As a Universal Processor offering industry-leading performance for all workloads, Prodigy-powered data center servers can seamlessly and dynamically switch between computational domains (such as AI/ML, HPC, and cloud) with a single homogeneous architecture. By eliminating the need for expensive dedicated AI hardware and dramatically increasing server utilization, Prodigy reduces CAPEX and OPEX significantly while delivering unprecedented data center performance, power, and economics. Prodigy integrates 192 high-performance custom-designed 64-bit compute cores, to deliver up to 4.5x the performance of the highest-performing x86 processors for cloud workloads, up to 3x that of the highest performing GPU for HPC, and 6x for AI applications.
“Having a robust software stack tested and ready to go upon the launch of the Prodigy Universal Processor chip is key to rapid adoption by data centers around the world looking to leverage their existing applications while achieving industry-leading performance for hyperscale, high-performance computing and artificial intelligence workloads,” said Dr. Radoslav Danilak, founder and CEO of Tachyum. “With each new enrichment we are able to incorporate into Prodigy’s software stack, we magnify the ability of a Prodigy platform ready to revolutionize the world.”

Hi Slade,

It's interesting that two independent company are promoting each other isn't?

Thanks to @IloveLamp for a reminder.

Let add a bits of speculation to the mix.

This is from there recent white paper. Maybe @Dio can have a look.

"Introduction
Tachyum Prodigy, the world’s first Universal Processor, was designed from the ground up to provide leading-edge AI features to address the emerging demand for AI across a wide range of applications and workloads. Prodigy’s revolutionary new architecture unifies the functionality of CPU, GPGPU, and TPU to address a wide range of workloads, including today’s ever-increasing AI demands without costly and power-hungry accelerators.

In addition to its unified architecture, Prodigy’s AI subsystem incorporates groundbreaking features that deliver high performance and efficiency for AI applications, including the 4-bit TAI exponential data type and multiple levels of sparse matrix processing which enables Prodigy to process large language models (LLMs) with 2-bit effective weights, providing never-before-seen efficiency. In addition, Prodigy integrates up to 16 DDR5 memory controllers to provide unprecedented memory bandwidth and capacity. Prodigy’s powerful AI capabilities enable LLMs to run much easier and cost-effectively than existing CPU + GPGPU based systems. A single 96-core Prodigy with 1 TB of memory can run a ChatGPT4 model with 1.7 trillion parameters, whereas it requires 52 Nvidia H100 GPUs to run the same thing at significantly higher cost and power consumption.

This paper presents the Prodigy ATX Platform, focusing on the hardware architecture, target applications, and how it will democratize AI for those who wouldn’t normally have access to sophisticated AI models. The Prodigy ATX Platform allows everyone to develop and run cutting edge AI models for as low as $5,000 in an entry-level platform SKU configuration featuring a 48-core Prodigy and 256 GB of DDR5 memory."

White Papers | Tachyum

Technical deep dives into Tachyum Prodigy and its features, abilities, use-cases, development, and architecture.

www.tachyum.com

Learning

 

[Fact Finder]

Hi DB,
Aren't Socionext already using our IP in 7nm.
I think there was a lot of press about it around CES 2023.

Hi WH
I missed that but entirely possible because the former CEO Mr. Dinardo actually said AKD1000 could scale to 14nm and 7nm when 7nm was the smallest at a major foundry. Since then Peter van der Made has mentioned 5nm and Anil Mankar has spoken about 4nm. Given that Brainchip sold only two nodes to Renesas and AKD1000 is 80 nodes I as a technophobe would think that as Anil Mankar has stated you can run keyword spotting on 1 node it just increases latency you could easily put 1 node on 2nm at TSMC but more likely much more. Where is @Diogenese?

My opinion only DYOR
Fact Finder

 

[Diogenese]

@Diogenese - Jem Davis has turned up as lead NED at Literal- Labs AI - https://www.literal-labs.ai they are making some pretty aggressive claims on power saving.. If you have time would you care to comment on this ....

"Literal Labs applies a streamlined and efficient approach to AI that is faster, explainable, and up to 10,000X more energy efficient than today’s neural networks. Similar to NNs in that customers train a dataset, Literal Labs trains Tsetlin machine models specific to customer datasets. Our approach results in an optimised machine model that is then deployed onto the target hardware. The Tsetlin machine model can be deployed as software only or can be accelerated using Literal Labs accelerators. Our benchmarking shows we can achieve 250X faster inferencing than XG Boost using software only, and up to 1,000X faster and up to 10,000X less energy consumption when using hardware acceleration. The company was spun out of Newcastle University by world leaders in Tsetlin machine Dr. Alex Yakovlev and Dr. Rishad Shafik, and led by former Arm CPU division VP and semiconductor startup founder Noel Hurley. "

Hi alby,

There isn't much technical detail available other than Wiki's discussion on Tsetlin machines.

https://en.wikipedia.org/wiki/Tsetlin_machine

A Tsetlin machine is a form of learning automaton collective for learning patterns using propositional logic. Ole-Christoffer Granmo created[1] and gave the method its name after Michael Lvovitch Tsetlin, who invented the Tsetlin automaton[2] and worked on Tsetlin automata collectives and games.[3] Collectives of Tsetlin automata were originally constructed, implemented, and studied theoretically by Vadim Stefanuk in 1962.

The Tsetlin machine uses computationally simpler and more efficient primitives compared to more ordinary artificial neural networks.[4]

As of April 2018 it has shown promising results on a number of test sets.[5][6]
...
The Tsetlin automaton is the fundamental learning unit of the Tsetlin machine. It tackles the multi-armed bandit problem, learning the optimal action in an environment from penalties and rewards. Computationally, it can be seen as a finite-state machine (FSM) that changes its states based on the inputs. The FSM will generate its outputs based on the current states.

... which is above my pay grade.


There are no published patents for Literal Lbs.

The LL people did publish a paper in Norway in 2022. The abstract is public:

Resultat #2068852 - A Convolutional Tsetlin Machine-based Field Programmable Gate Array Accelerator for Image Classification - Cristin

Resultat #2068852 - A Convolutional Tsetlin Machine-based Field Programmable Gate Array Accelerator for Image Classification - Cristin

app.cristin.no

Vitenskapelig Kapittel/Artikkel/Konferanseartikkel

2022

A Convolutional Tsetlin Machine-based Field Programmable Gate Array Accelerator for Image Classification​

• Svein Anders Tunheim
• Jiao Lei
• Rishad Ahmed Shafik
• Alexandre Yakovlev og
• Ole-Christoffer Granmo

TITTEL

A Convolutional Tsetlin Machine-based Field Programmable Gate Array Accelerator for Image Classification

SAMMENDRAG

This paper presents a Field Programmable Gate Array (FPGA) implementation of an image classification accelerator based on the Convolutional Tsetlin Machine (CTM). The work is a concept design, and the solution demonstrates recognition of two classes in 4 × 4 images with a 2 × 2 convolution window. More specifically, there are two sub-Tsetlin Machines (TMs), one per class. A single sub-TM employs 40 clauses, each controlled by 20 Tsetlin Automata. The accelerator features random patch selection, in parallel for all clauses, based on reservoir sampling. The design is implemented in a Xilinx Zync XC7Z020 FPGA. With an operating clock speed of 30 MHz, the accelerator is capable of inferring at the rate of 3.3 million images per second with an additional power consumption of 20 mW from idle mode. The average test accuracy is 96.7% when trained on data with 10% noise. A training session with 100 epochs and 8192 examples takes 1.5 seconds. Due to the limited hardware resources required, the CTM accelerator represents a promising concept for online learning in energy-frugal systems. The solution can be scaled to multi-class systems and larger images.




From their website:


Literal Labs - Cambridge Future Tech (camfuturetech.com)

Literal Labs | Cambridge Future Tech | DeepTech Venture Builder

A new world of AI-enabled devices that are 10,000 times more energy efficient.

camfuturetech.com

Artifical Intelligence Redefined

Similar to NNs in that customers train a dataset, Literal Labs trains Tsetlin machine models specific to customer datasets. This approach results in an optimised machine model that is then deployed onto the target hardware.

The Tsetlin machine model can be deployed as software only or can be accelerated using Literal Labs accelerators. Literal Labs benchmarking shows it can achieve 250X faster inferencing than XG Boost using software only, and up to 1,000X faster and up to 10,000X less energy consumption when using hardware acceleration.

Value Proposition

One of the major challenges with traditional neural network-based models is their resource-intensive nature. As models become more complex, so too does the resource requirement. Literal Labs’ architecture, based on propositional logic, requires significantly fewer resources to solve AI problems, meaning that Literal Labs can deliver intelligent compute on devices with minimal energy usage and with little or no internet coverage.

The company was spun out of Newcastle University by world leaders in Tsetlin machine Dr. Alex Yakelov and Dr. Rishad Shafik, and led by former Arm CPU division VP and semiconductor startup founder Noel Hurley.

There is a video which shows that the processor is involved in the calculations:

This would add to latency, so it is probably slower than Akida.

No wonder Jem was a bit cryptic on the podcast - buyer's remorse. I think he's Tsetlin for second best.

 

[Smoothsailing]

Can someone please advise me when the next financial is due by is it this week ?.