BRN Discussion Ongoing

[Dhm]

I have also been looking at Tesla's significant inhouse investment in DOJO and it seems Tesla are very comfortable with what they have. Yet it seems extremely high in power consumption. From what I can see Tesla won't be looking in our direction anytime soon.

 

[Slade]

Rob and Todd talking about vibration and sound analysis got me thinking and searching. I think this company is very interesting and since Renesas is one of their partners I feel that there might be a connection to Akida, particularly in the area of predictive maintenance. Forgive me if previously discussed.

MicroAI™, an Edge AI-enablement company based in Dallas

MicroAI is an AI-driven technology company that designs, develops & deploys next-generation IoT solutions for OEMs, network operators & enterprises.

www.micro.ai

Company Overview

MicroAI™, an Edge AI-enablement company based in Dallas, TX., delivers Intelligent Asset Management solutions to companies within the Manufacturing, O&G, Automotive, Telecom, and Semiconductor sectors.​

MicroAI’s award-winning, proprietary, AI/ML solution—MicroAI AtomML™–enables a transformational approach to using Embedded AI to improve the performance and security of IIoT devices and machines. MicroAI is a leader in Industry 4.0 initiatives, providing Edge and Endpoint AI/ML solutions that deliver business value in the areas of Asset Optimization, Predictive Maintenance, and advanced Cyber-Security.
Making Predictive Manufacturing a Reality – MicroAI Factory™

Official Partners​

• 
• 
• 
• 
• 

MicroAI Factory™ and the Power of Predictive Maintenance​

04 Feb 22

Machine output and uptime are critical KPIs (key performance indicators) for any machine-intensive enterprise. Lack of real-time insights, unscheduled downtimes, and static maintenance routines can all combine to create a production ecosystem that performs well below its optimum capability.
Preventive maintenance is the legacy approach to machine optimization. For several decades that approach was sufficient to maintain decent levels (i.e. ~ 70%) of OEE (overall equipment effectiveness). To gain and maintain competitive advantage “decent” performance is no longer enough. Manufactures across every industry segment are now looking at Edge AI (artificial intelligence) and ML (machine learning) technologies to power a shift from preventive to predictive maintenance.
What is predictive maintenance? How does it work? Why is it better that the old preventive approach?

Predictive Maintenance Defined​

An intelligent predictive maintenance solution utilizes Endpoint AI and Edge-native AI technologies to enable equipment operators and stakeholders to gain deeper insights into the real-time status and heath of their machine assets. Asset-specific data is analyzed to predict when that asset will require maintenance. In this method, maintenance is performed based on actual machine-generated information instead of the old time-based approach.
The primary differences between predictive and preventive maintenance are summarized below.

	Predictive	Preventive
AI and ML enabled?	Yes	No
Maintenance triggers	Based on analysis of real-time and historical asset performance data. Supported by device and machine AI-enabled endpoint analytics.	Based on static, manual, schedules and routines. Often hindered by lack of insight into current asset health.
Operational impact	Maintenance performed when needed to maintain asset health, minimizing production impact.	Maintenance often performed too early or too late, resulting in non-optimized performance and/or unnecessary downtimes.
Business impact	Reduced maintenance costs, increased output, better resource utilization, and optimized OEE.	Higher maintenance costs, non-productive maintenance routines, shorter asset lifespans, and sub-par OEE.
Competitive impact	Higher production rates + longer asset lifespans + reduced maintenance costs = lower production costs and enhanced competitive position.	Non-optimized production capacity + degraded asset health + reliance on human-dependent, non-automated, processes = higher production costs, higher prices, and competitive weakness.

Predictive Maintenance at the Endpoint​

MicroAI Factory is an Edge-native AI solution that provides manufacturers with predictive maintenance capabilities that produce advantages in production output, machine utilization, operator efficiency, production costs and OEE. Features of MicroAI Factory include the following:

• Microcontroller-based intelligence: MicroAI Factory is unique in that it embeds predictive maintenance intelligence directly into the microcontroller (MCU) of the manufacturing device or machine. This approach offers distinct advantages when compared to cloud-based solutions.
  • Completely self-contained
  • Local collection and analysis of asset data
  • Reduction in the amount of data transferred to the cloud
  • More secure from cyber-attack
  • Lower cost
• Customizable algorithms: Ability to customize AI-enabled algorithms on an asset-by-asset level to accommodate specific operational or environmental conditions for the device or machine. Predictive maintenance routines are based on real-time analytics that provide insights into current and historical performance trends.
• Data aggregation and dashboarding: Most legacy preventive maintenance routines rely on the manual collection of asset data and subsequent siloed analysis of that data. MicroAI Factory provides automatic aggregation, analysis, and presentation of asset data. Asset owners can quickly customize asset analytics to best meet their operational needs.
• Workflow optimization: MicroAI Factory embeds workflows that learn, train, and evolve. Predictive maintenance is supported by workflows that are automated and intelligent. This reduces or eliminates the need for human intervention in the maintenance scheduling process.

Predictive Maintenance – Operational and Business Value​

Any technological innovation is only as good as the tangible value that it provides to its adopters. Preventive maintenance powered by MicroAI Factory’s Edge-native AI technology delivers both operational and business value. Just a few examples would include:

• Holistic and self-contained ecosystem: Non-siloed, at-a-glance, perspective of real-time performance and events. A more comprehensive and efficient predictive maintenance solution.
• Deeper visibility and insights: Ability to fast-track issue identification and corrective action and to identify recurring problems based on historical analytics.
• Increased machine and device uptime: Elimination of downtimes due to unnecessary maintenance activities and/or unforeseen malfunctions.
• Optimization of production capacity: Reduction in machine downtime equates to increased output and higher OEE scores.
• Improved product quality: Predictive maintenance helps detect those machine performance issues that can negatively impact the quality of the product(s) being produced.
• Reduced maintenance costs: Cost is reduced via elimination of unnecessary maintenance activities, less reliance on human interaction, and automation of processes.
• Extension of capital-intensive asset lifespans: Real-time asset health monitoring, process-driven mitigation actions, and predictive maintenance capability all combine to extend the lifespan of expensive assets.
• More competitive operational overhead: Improved asset health and performance reduces operational costs and has a beneficial trickle effect across other business costs.

MicroAI Factory is bringing Edge-native AI predictive maintenance to companies within the manufacturing, telecom, energy, automotive, and semiconductor sectors. Factory floors are being optimized to reach new levels of operational excellence and OEE.

 

[Diogenese]

Now as for Neon it is a bit like pulling rabbits out of hats. You make it out of nothing or at least what you see when you look in the magicians top hat which as we know is actually full of air. Clearly air is equally distributed around the world so it would not be a long term shortage:

Chemistry for Kids: Elements - Neon

Kids learn about the element neon and its chemistry including atomic weight, atom, uses, sources, name, and discovery. Plus properties and characteristics of neon.

www.ducksters.com

The above is science for children but it gets the idea across. Neon is one of those dirty gases that you have to be careful with in production so that is why we allowed Ukrainian workers to take the risk.

I have no doubt the tech giants if pushed would allow their employees to take the same risk in Bangladesh, Africa or India.

Did I just say that?

My opinion only DYOR
FF

AKIDA BALLISTA

50 years ago, you would have had to go to uni to learn that stuff. Now they're giving it to 5 year olds for free

 

[Fact Finder]

I have also been looking at Tesla's significant inhouse investment in DOJO and it seems Tesla are very comfortable with what they have. Yet it seems extremely high in power consumption. From what I can see Tesla won't be looking in our direction anytime soon.

15KW is 15,000 watts for one DOJO producing 400watts of heat requiring dissipation. Tesla must not have got the memo about problems with power generation, base load power or climate change. They might be processing quickly but the information/data still has to come from the vehicle to the DOJO and back to the car to take action. What happens if there is a connection problem because Russia explodes a satellite and takes out a number of Tesla's 10,000 satellites. Even a cyber attack creating mass loss of connection. I cannot see how this actually solves problems for Tesla. Maybe Dio can show me the path to righteousness.

As Tesla was speaking about its robots maybe their end game is a closed system for autonomous taxis in cities running like large robots within a defined geographic location???

My opinion only DYOR
FF

AKIDA BALLISTA

 

[Deleted member 118]

Hi DHM,

Please see Gex reply to explain Pantene

Rob or Todd said half our customer engagements are Semiconductor IP Customers, we have 15 EAP's so half of that is either 7 or 8, we already know about 2 which are Renesas & Megachips so there are 5 or 6 Semiconductor IP Customers remaining that we don't know about.

I can’t see why they can’t just disclose to the shareholders the precise amount of NDA they have in place currently. What’s the harm in that? Does anyone want to ask @ fact finder

 

[Dhm]

Another thing that many people I speak with don't understand is that neuromorphic chips are analog based. This is partly because that the maximum number of transistors that can be put on a chip seems to have been reached. Moore's law is no longer relevant. Here is a video on analog computers, part of a fascinating series called Veritasium. All videos produced under this banner are so worth watching. The presenter is an Aussie too!



The amount of views that the Veratasium series gets ventures into many millions of views for each video. It is astounding. A very interesting series.

 

[Fact Finder]

My key points from this presentation which repeat some of those already made by others however it is a must watch:

Brainchip Headcount: Currently 71 and will be over 100 by years end - actively growing the business.

Brainchip Patents: Current 21 patents either granted or filed.

Brainchip in cabin demonstration: When this video was first released there was debate in another place about how many AKD1000's would be used to provide the demonstration. Well now we know. It is one AKD1000 doing all three functions being Facial Recognition at 22.6mw's, Keyword Spotting at 600uw's and Visual Wake at 6mw's.

Brainchip AKIDA accuracy: Once converted to 4 bit AKIDA is returning the same accuracy as the others running at 32 bit, 16 bit & 8 bit.

Brainchip Meta TF: In 2021 Brainchip Meta TF welcomed over 5,000 new users - there was debate over in another place about whether this statistic was new users or not. Well now we know. No wonder the small band at Brainchip spoke of explosive sales and insatiable demand.

Brainchip high speed recognition: May mean nothing but the Formula Racing Car has BMW printed large on its rear foil.

Brainchip & Mega Chips: As above but noting that to my knowledge the visual of the person using ARVR and holding a hand controller has not previously appeared in earlier presentations. Someone who is into gaming may be able to identify the hand controller.

Brainchip doing vibration & sound for defect detection: Rob and Tod were both excited by AKIDA processing both sound and vibration on chip and they revealed for the first time that I am aware the room fan demonstration.

Brainchip doing beer tasting: Slade would have been worried by this and would be imagining a world where machines take over drinking his beer, but for me it was the definitive statement by Rob and Tod that AKIDA can detect any chemical compound regardless of whether it is presented as a liquid or a gas. This opens up the full range of domestic, industrial, automotive, aerospace and defence applications where air quality or chemical infiltration is a potential issue. In motor vehicles AKIDA could be monitoring carbon monoxide infiltration in the passenger cabin. In passenger jets it has been hypothesised that MH370 may have been due to a gas escape into the air filtration system causing unconsciousness. Not to mention chemical warfare. In the home particularly in the UK and EU gas detectors as well as smoke detectors are legislated. If the detectors were made smart and could say what the chemical involved was this would reduce the hazard faced by emergency personnel and allow for the correct medical treatment at first instance. As a by the way one of the things I was taught as a first responder to a gas poisoning was not to do mouth to mouth as you would potentially put yourself at risk. Knowing what the gas inhaled was is critical.

My opinion only DYOR
FF

AKIDA BALLISTA in 2022 and beyond.

Being Friday it is nice to finish on a beer note so read the last paragraph headed Brainchip doing beer tasting then read the following and the full linked paper:

• Classifying Carbon Nanotube Sensor Data
  Working with the Chemical Gas Sensor team at NASA Ames, HECC data science experts built a classification system for carbon nanotube sensors for aerospace applications, such as cabin air quality monitoring. The Ames team got significant initial results— accuracy was better than 94%—allowing them to move forward with the pilot project. See: Carbon Nanotube Gas Sensor Using Neural Networks (PDF, 467KB)

I am not saying it is definitely AKIDA technology but AKIDA is being tested at Ames, Rob Telson said NASA was doing vision and other things they cannot talk about and Tod Vierra said "AKIDA can detect any chemical compound regardless of whether it is presented as a liquid or a gas and the AERO system using the ADE back in 2019/20 the authors including Peter van der Made wrote that using only 200 neurons AERO identified the 20 gas data set each set containing 10 gasses with state of the art accuracy of 95% with a latency of 3 seconds. The accuracy and speed of any chip that has been hardened for space flight is affected so the disparity between the results reported of 1% fits and of course AERO was not using carbon nanotube technology.

My opinion only but I suspect I might have got one right but DYOR
FF

AKIDA BALLISTA

 

[Diogenese]

15KW is 15,000 watts for one DOJO producing 400watts of heat requiring dissipation. Tesla must not have got the memo about problems with power generation, base load power or climate change. They might be processing quickly but the information/data still has to come from the vehicle to the DOJO and back to the car to take action. What happens if there is a connection problem because Russia explodes a satellite and takes out a number of Tesla's 10,000 satellites. Even a cyber attack creating mass loss of connection. I cannot see how this actually solves problems for Tesla. Maybe Dio can show me the path to righteousness.

As Tesla was speaking about its robots maybe their end game is a closed system for autonomous taxis in cities running like large robots within a defined geographic location???

My opinion only DYOR
FF

AKIDA BALLISTA

But surely Valdimir would not do anything as irrational as that!?

I was going to comment on that brassy thing near the bottom (definitely not talking about Anastasi) being full of cooling liquid (much like myself and @Slade) but I thought better of it as Anastasi explained the heat thing in a much more interesting way. 15 kW is enough to power your house and both your neighbours'.

 

[Diogenese]

Being Friday it is nice to finish on a beer note so read the last paragraph headed Brainchip doing beer tasting then read the following and the full linked paper:

• Classifying Carbon Nanotube Sensor Data
  Working with the Chemical Gas Sensor team at NASA Ames, HECC data science experts built a classification system for carbon nanotube sensors for aerospace applications, such as cabin air quality monitoring. The Ames team got significant initial results— accuracy was better than 94%—allowing them to move forward with the pilot project. See: Carbon Nanotube Gas Sensor Using Neural Networks (PDF, 467KB)

I am not saying it is definitely AKIDA technology but AKIDA is being tested at Ames, Rob Telson said NASA was doing vision and other things they cannot talk about and Tod Vierra said "AKIDA can detect any chemical compound regardless of whether it is presented as a liquid or a gas and the AERO system using the ADE back in 2019/20 the authors including Peter van der Made wrote that using only 200 neurons AERO identified the 20 gas data set each set containing 10 gasses with state of the art accuracy of 95% with a latency of 3 seconds. The accuracy and speed of any chip that has been hardened for space flight is affected so the disparity between the results reported of 1% fits and of course AERO was not using carbon nanotube technology.

My opinion only but I suspect I might have got one right but DYOR
FF

AKIDA BALLISTA

This is something that's always puzzled me. On a Qantas flight to Europe, you can't smell the shiraz. I don't know whether it's because:
a) your nose stops working at altitude;
b) the wine does not emit aroma carrying molecules;
c) the wine has been in the fridge;
d) other;

but it is certainly worth a research paper.

The taste (gustatory) is fine, but the olfactory is missing.

So PvdM/Akida have a future project next time they jet off to France.

 

[Diogenese]

Another thing that many people I speak with don't understand is that neuromorphic chips are analog based. This is partly because that the maximum number of transistors that can be put on a chip seems to have been reached. Moore's law is no longer relevant. Here is a video on analog computers, part of a fascinating series called Veritasium. All videos produced under this banner are so worth watching. The presenter is an Aussie too!



The amount of views that the Veratasium series gets ventures into many millions of views for each video. It is astounding. A very interesting series.

Hi Dhm,

What a magnificent discussion of the development of computer. We worked with analog computers during the early years at uni.

I had the priviledge, as a very junior engineer, of working with the engineers who, many years earlier, installed the first digital computer at Sydney uni in 1949 (Silliac, based on Ennica described in the video from a little after 14 minutes on). It was all valves and hand wired and took up a whole room, and would have had much less computing power than your mobile phone while generating enough heat to roast a bullock.

The problem of repeatablity of results due to manufacturing variations persists with silicon analog computers today.

That is why PvdM's inspired decision to build a digital neural network is so remarkable.

Digital circuits are much more immune to manufacturing variation than analog components.

This is the basic neural processing unit (NPU) which PvdM invented:

WO2020092691A1 AN IMPROVED SPIKING NEURAL NETWORK

It is purely digital. This is what makes it so malleable.

[some may recognize this from the other place as my favourite (Zeebot, can we have an English spell checker?) Akida picture].

Many university and competing commercial projects are attempting to use analog neurons based on ReRAM/MemRistors, probably drawn by the close analogy between real neuron spikes and analog silicon spikes, but the problems such as manufacturing variability are hampering their progress.

Theoretically analog neuron more closely approximate digital neurons, but the practical difficulties of building reliable analog neurons have, to a large extent, stymied the progress of analog neurons.

In addition, SNNs are far more efficient than CNNs, the current vogue in neural networks, although Akida has been modified to convert CNN data to SNN inputs (digital "spikes", being individual binary digital bits, or in the latest embodiment Akida 1000 can accommodate up to 4-bit inputs and internal weights.

Thus existing CNN based systems, of which there are many, can be adapted to run much more efficiently on Akida.

 

[Ian]

Valeo's strategic outlook

 

[hamilton66]

Great Summary, thank you FF.
Akida test tasting beer is great and shows just how healthy beer can be. From my analysis of the screenshot below it is clear to me that Akida has identified Pale Ale as a healthy drink, a 'winning' formula, providing essential vitamins and minerals through wheat and barley and giving the drinker a power ratio of 6.3. I am yet to work out what the 816 FPS is but I am sure it is good. I take it that Akida can spit the testing sample back into the glass, so really it's a win win. Pale Ale will be on the menu this weekend.

View attachment 1770

Slade, a bit of useless trivia information, which I'm sure will be music to ur ears. Beer is a registered food.
GLTA

 

[Baisyet]

Anybody watching Valeo live stream

 

[thelittleshort]

Hi Dhm,

What a magnificent discussion of the development of computer. We worked with analog computers during the early years at uni.

I had the priviledge, as a very junior engineer, of working with the engineers who, many years earlier, installed the first digital computer at Sydney uni in 1949 (Silliac, based on Ennica described in the video from a little after 14 minutes on). It was all valves and hand wired and took up a whole room, and would have had much less computing power than your mobile phone while generating enough heat to roast a bullock.

The problem of repeatablity of results due to manufacturing variations persists with silicon analog computers today.

That is why PvdM's inspired decision to build a digital neural network is so remarkable.

Digital circuits are much more immune to manufacturing variation than analog components.

This is the basic neural processing unit (NPU) which PvdM invented:

WO2020092691A1 AN IMPROVED SPIKING NEURAL NETWORK

It is purely digital. This is what makes it so malleable.

View attachment 1794

[some may recognize this from the other place as my favourite (Zeebot, can we have an English spell checker?) Akida picture].

Many university and competing commercial projects are attempting to use analog neurons, probably drawn by the close analogy between real neuron spikes and analog silicon spikes, but the problems such as manufacturing variability are hampering their progress.

Theoretically analog neuron more closely approximate digital neurons, but the practical difficulties of building reliable analog neurons have, to a large extent, stymied the progress of analog neurons.

In addition, SNNs are far more efficient than CNNs, the current vogue in neural networks, although Akida has been modified to convert CNN data to SNN inputs (digital "spikes", being individual binary digital bits, or in the latest embodiment Akida 1000 can accommodate up to 4-bit inputs and internal weights.

Thus existing CNN based systems, of which there are many, can be adapted to run much more efficiently on Akida.

Thanks @Diogenese - wow we should count ourselves lucky to have someone of your level of knowledge on here

 

[thelittleshort]

Anybody watching Valeo live stream

 

[thelittleshort]

View attachment 1796

Link

 

[Diogenese]

Hi Dhm,

What a magnificent discussion of the development of computer. We worked with analog computers during the early years at uni.

I had the priviledge, as a very junior engineer, of working with the engineers who, many years earlier, installed the first digital computer at Sydney uni in 1949 (Silliac, based on Ennica described in the video from a little after 14 minutes on). It was all valves and hand wired and took up a whole room, and would have had much less computing power than your mobile phone while generating enough heat to roast a bullock.

The problem of repeatablity of results due to manufacturing variations persists with silicon analog computers today.

That is why PvdM's inspired decision to build a digital neural network is so remarkable.

Digital circuits are much more immune to manufacturing variation than analog components.

This is the basic neural processing unit (NPU) which PvdM invented:

WO2020092691A1 AN IMPROVED SPIKING NEURAL NETWORK

It is purely digital. This is what makes it so malleable.

View attachment 1794

[some may recognize this from the other place as my favourite (Zeebot, can we have an English spell checker?) Akida picture].

Many university and competing commercial projects are attempting to use analog neurons, probably drawn by the close analogy between real neuron spikes and analog silicon spikes, but the problems such as manufacturing variability are hampering their progress.

Theoretically analog neuron more closely approximate digital neurons, but the practical difficulties of building reliable analog neurons have, to a large extent, stymied the progress of analog neurons.

In addition, SNNs are far more efficient than CNNs, the current vogue in neural networks, although Akida has been modified to convert CNN data to SNN inputs (digital "spikes", being individual binary digital bits, or in the latest embodiment Akida 1000 can accommodate up to 4-bit inputs and internal weights.

Thus existing CNN based systems, of which there are many, can be adapted to run much more efficiently on Akida.

I remember as a child, Donald Duck comics used to display computers consuming enormous amounts of electricity when calculating the latest stock market variations on Uncle Scrooges wealth - I'm glad that it does not happen now when my computer calculates my wealth ...


There may be more than one reason why that is the case ...

... but I do enjoy diving into my room full of dollar coins ...

 

[Ian]

Anybody watching Valeo live stream

I didn't realise it was live until you mentioned. Thanks!

 

[Diogenese]

Just an aside on Ukraine, for whom I have the greatest sympathy, but it takes a good war to have your capital city spelt correctly.

If the West was going to do anything, Crimea was the only real opportunity. Megalomaniacs must be nipped in the bud.

 

[HopalongPetrovski]

I remember as a child, Donald Duck comics used to display computers consuming enormous amounts of electricity when calculating the latest stock market variations on Uncle Scrooges wealth - I'm glad that it does not happen now when my computer calculates my wealth ...


There may be more than one reason why that is the case ...

... but I do enjoy diving into my room full of dollar coins ...