BRN Discussion Ongoing

Lex555

Regular
Last Tuesday 12:52, @Fullmoonfever brought up something by Maxim
Hi D,
Appreciate your thoughts on the following new recent release. Couple things caught my eye particularly the weights but not tech enough to be comfortable this is similar to Akida?

Edit. Not saying Akida involved but curious to any similarities
.

https://au.mouser.com/ProductDetail/Maxim-Integrated/MAX78000EXG+?qs=yqaQSyyJnNigS5t/Kz0nhQ==

MAX78000 Artificial Intelligence Microcontroller with Ultra-Low-Power Convolutional Neural Network Accelerator

Artificial intelligence (AI) requires extreme computational horsepower, but Maxim is cutting the power cord from AI insights. The MAX78000 is a new breed of AI microcontroller built to enable neural networks to execute at ultra-low power and live at the edge.

www.maximintegrated.com

In my reply, I noted that Maxim referred to 1-, 2- , 4- , and 8-bit weights (They don't mention how many bits in the activations).

#19,425
While Maxim don't refer to spikes, they do refer to 1-bit weights. They also refer to 8-bit weights in the same breath, so they have gone for additional accuracy as an option cf Akida's optional 4-bit weights/actuations.

Maxim have several NN patents, mainly directed to CNN using the now fashionable in-memory-compute, eg
:

US2020110604A1 ENERGY-EFFICIENT MEMORY SYSTEMS AND METHODS
Priority: 20181003.
...

Now I'm not saying it's impossible for the Akida IP to stretch to 8-bits, but we have not been told that it does. Similar to the 4-bit Akida, an 8-bit Akida would have even greater accuracy than the initial 1-bit Akida at the expense of speed and power.

Maxim also dabbled in analog and Frankenstein (analog/digital) NNs.

An analog NN on its own would struggle with accuracy to compete with a multibit digital NN.

Interestingly, Maxim is now part of Analog Devices:

https://www.maximintegrated.com/en.html

Maxim have had an AI chip since 2020. It uses ARM Cortex M4

https://www.maximintegrated.com/en/products/microcontrollers/artificial-intelligence.html

Artificial intelligence (AI) is opening up a whole new universe of possibilities, from virtual assistants to self-driving cars, automated factory equipment, and voice recognition in consumer devices. But the computational horsepower to enable these possibilities is extreme, and requires expensive, power-hungry, and big processors. In embedded devices, this functionality isn't really available–embedded microcontrollers are too slow to effectively process images and make decisions in real time.

Enter Maxim's new line of Artificial Intelligence microcontrollers. They run AI inferences hundreds of times faster and lower energy than other embedded solutions. Our built in neural network hardware accelerator practically eliminates the energy spent on audio and image AI inferences. Now small machines like thermostats, smart watches, and cameras can deliver the promise of AI—embedded devices can see and hear like never before.

Get started today with the MAX78000FTHR for only $25
.

https://datasheets.maximintegrated.com/en/ds/MAX78000.pdf

Artificial intelligence (AI) requires extreme computational horsepower, but Maxim is cutting the power cord from AI insights. The MAX78000 is a new breed of AI microcontroller built to enable neural networks to execute at ultra-low power and live at the edge of the IoT. This product combines the most energy-efficient AI processing with Maxim's proven ultra-low power microcontrollers. Our hardware-based convolutional neural network (CNN) accelerator enables battery-powered applications to execute AI inferences while spending only microjoules of energy. The MAX78000 is an advanced system-on-chip featuring an Arm® Cortex®-M4 with FPU CPU for efficient system control with an ultra-low-power deep neural network accelerator. The CNN engine has a weight storage memory of 442KB, and can support 1-, 2-, 4-, and 8-bit weights (supporting networks of up to 3.5 million weights). The CNN weight memory is SRAM-based, so AI network updates can be made on the fly. The CNN engine also has 512KB of data memory. The CNN architecture is highly flexible, allowing networks to be trained in conventional toolsets like PyTorch® and TensorFlow®, then converted for execution on the MAX78000 using tools provided by Maxim. In addition to the memory in the CNN engine, the MAX78000 has large on-chip system memory for the microcontroller core, with 512KB flash and up to 128KB SRAM. Multiple high-speed and low-power communications interfaces are supported, including I2S and a parallel camera interface (PCIF).

Neural Network Accelerator
• Highly Optimized for Deep Convolutional Neural Networks • 442k 8-Bit Weight Capacity with 1,2,4,8-Bit Weights
• Programmable Input Image Size up to 1024 x 1024 pixels
• Programmable Network Depth up to 64 Layers
• Programmable per Layer Network Channel Widths up to 1024 Channels
• 1 and 2 Dimensional Convolution Processing
• Streaming Mode
• Flexibility to Support Other Network Types, Including MLP and Recurrent Neural Networks



View attachment 10648


This is a one-to-one fit for the first highlighted paragraph.
The MCU with an embedded CNN accelerator is a system on chip combining an Arm Cortex-M4 with a RISC-V core that can execute application and control codes as well as drive the CNN accelerator. The CNN engine has a weight storage memory of 442KB and can support 1-, 2-, 4-, and 8-bit weights (supporting networks of up to 3.5 million weights). On the fly, AI network updates are supported by the SRAM-based CNN weight memory structure. The architecture is flexible and allows CNNs to be trained using conventional toolsets such as PyTorch and TensorFlow.


View attachment 10651
Hi D, appreciate your expertise and insights, though sometimes don’t really follow what your trying to say. It could be helpful to post a summary at top so slow-wits like myself can try to grasp your posts
 
  • Like
  • Haha
  • Fire
Reactions: 15 users
D

Deleted member 118

Guest
Brought some more BRN shares today after I decided it was to risky getting a mortgage for a house.

 
  • Like
  • Haha
  • Fire
Reactions: 46 users

Rskiff

Regular
how good is your brain, I bet Akida would do well
 
  • Haha
  • Like
  • Love
Reactions: 7 users
Brought some more BRN shares today after I decided it was to risky getting a mortgage for a house.


Yeah cause you won’t need a mortgage in anout 2 years time 😉
 
  • Like
  • Fire
  • Love
Reactions: 17 users

equanimous

Norse clairvoyant shapeshifter goddess
Just checked my emails and nothing at this point in time. I sent it to Perth and to the US. The time difference can come into play but I suspect it has taken them by surprise. 😂

Will report when and if I hear anything. They may choose to ignore me for a little while. 😎

Regards
FF

AKIDA BALLISTA

Still all quiet on the western and US front. FF

AKIDA BALLISTA
I think people are scrambling to put the worms back into the can....
 
  • Like
  • Haha
  • Fire
Reactions: 13 users
Hi 👋 FF , What is communications EAP ? I crossed to the dark side momentarily and a poster mentioned that . With Napoleon and Hitler failing to defeat Russia via the Ukraine ( the flat lands ) what on earth is the Biden administration thinking by taking on a similar strategy. All I see as a result is petrol and gas going through the roof and creating disharmony and the need for our ultra low energy low cost chip to be in greater demand .Am I correct in stating that there is an opening for a government to take from a company an asset ( our Brainchip) if it is deemed of utmost importance to the nation if that company has not been moving fast enough to get it into use pronto . If so and BRN is deemed to be in that category, I reckon we would be given an exemption due to most governments around the world at the moment being unable to show that they have anyone smart in them .
Just a bit of law from ‘the Castle’ it would have to be ‘on just terms’ 🤣😂🤣😂

‘What is communications EAP?’

The only thing that comes to mind is in the video interview involving Sean Hehir and Tony Dawe Sean Hehir said that they had been approached by a large communications company to discuss using AKIDA for a purpose Brainchip had not considered.

If it is not this then it could be Intellisense as they are looking at cognitive communications using AKIDA on an SBIR in the USA.

Cannot think of anything else.

My opinion only DYOR
FF

AKIDA BALLISTA
 
  • Like
  • Love
  • Fire
Reactions: 24 users
Just a bit of law from ‘the Castle’ it would have to be ‘on just terms’ 🤣😂🤣😂

‘What is communications EAP?’

The only thing that comes to mind is in the video interview involving Sean Hehir and Tony Dawe Sean Hehir said that they had been approached by a large communications company to discuss using AKIDA for a purpose Brainchip had not considered.

If it is not this then it could be Intellisense as they are looking at cognitive communications using AKIDA on an SBIR in the USA.

Cannot think of anything else.

My opinion only DYOR
FF

AKIDA BALLISTA
Thankyou . You are the best .
 
  • Like
  • Love
Reactions: 6 users
how good is your brain, I bet Akida would do well

No this proved why AKIDA uses so little energy just like the human brain.

If you told AKIDA to watch the Skoda that is what it will do it will only process the car.

This is a cheap trick used by magicians they tell you to watch something and that is what you do allowing things outside the subject of your focus to occur without being seen.

That is why when AKIDA is the brain a of a security camera alarm it does not give false alerts for birds, cats, dogs it only reacts to unwanted relatives and debt collectors.

My opinion only but it is a very important distinction between AKIDA and everything else on the market.
FF

AKIDA BALLISTA
 
  • Like
  • Fire
Reactions: 13 users
No this proved why AKIDA uses so little energy just like the human brain.

If you told AKIDA to watch the Skoda that is what it will do it will only process the car.

This is a cheap trick used by magicians they tell you to watch something and that is what you do allowing things outside the subject of your focus to occur without being seen.

That is why when AKIDA is the brain a of a security camera alarm it does not give false alerts for birds, cats, dogs it only reacts to unwanted relatives and debt collectors.

My opinion only but it is a very important distinction between AKIDA and everything else on the market.
FF

AKIDA BALLISTA
Keeping in mind I am a technophobe, old and a retired lawyer and so have no idea but this AKIDA technology differentiator could provide a way for ground to air and air to air missiles to remain locked on the enemy plane and not be confused by chaff or flares.

My opinion only DYOR
FF

AKIDA BALLISTA
 
  • Like
  • Fire
  • Wow
Reactions: 22 users
D

Deleted member 118

Guest
Yeah cause you won’t need a mortgage in anout 2 years time 😉
I never brought that much but might be able to get a caravan, maybe if I buy a few more I might be able to afford a car as well

 
  • Haha
  • Like
Reactions: 15 users

Tuliptrader

Regular
  • Like
  • Fire
  • Love
Reactions: 19 users
  • Like
  • Fire
Reactions: 33 users

Slade

Top 20
Thank you @Toto111 . All eyes on TDK
 
  • Like
  • Love
Reactions: 14 users

Pappagallo

Regular
Very interesting wrt TDK especially when considering the preventative maintenance video published on our website. Echoing The Castle once again it’s the vibe-rational analysis one.
 
  • Like
  • Love
Reactions: 13 users
Hi @Toto111
This has all the signs of being significant. We need @Diogenese to do a deep patent dive into TDK to see what Ai they are using in this Edge sensor.

My opinion only DYOR
FF

AKIDA BALLISTA

I found this patent for TDK involving NN however I don’t have the expertise to interrogate it. I’ll leave that to DIO.


AppFThdr.gif
[Help] [Home] [Boolean Search] [Manual] [Number Search] [PTDLs] [CURR_LIST]
[Shopping Cart] [Order Copy] [Image]


( 1 of 1 )​



United States Patent Application
20200371076
Kind Code
A1
Koenig; Matthias
November 26, 2020

Method and Apparatus for Operating a Multi-Gas Sensor

Abstract
A method and apparatus for operating a multi-gas sensor are disclosed. In an embodiment, a method includes providing at least one calibration input comprising sensor design data of the multi-gas sensor, which varies dependent on production process parameters, and/or sensor production process parameter data of the multi-gas sensor, and/or measurement results of the multi-gas sensor captured when the multi-gas sensor is exposed to one of the gases or a gas mixture to be detected and/or sensed by the multi-gas sensor; providing a trained neural network including an input layer with K input nodes, an output layer with L output nodes and at least one hidden layer; storing each calibration input as a fixed input to a corresponding input node of the trained neural network; and providing a multi-gas sensor output for at least a part of the gases to be detected and/or sensed by the multi-gas sensor dependent on the trained neural network and actual measured sensor values from the sensor elements.


Inventors:​
Koenig; Matthias; (Muenchen, DE)
Applicant:​
Name​
City​
State​
Country​
Type​

TDK Electronics AG

Munich

DE​
Family ID:​
73052967
Appl. No.:​
16/878167
Filed:​
May 19, 2020
 
  • Like
  • Fire
Reactions: 21 users
I found this patent for TDK involving NN however I don’t have the expertise to interrogate it. I’ll leave that to DIO.


AppFThdr.gif
[Help] [Home] [Boolean Search] [Manual] [Number Search] [PTDLs] [CURR_LIST]
[Shopping Cart] [Order Copy] [Image]


( 1 of 1 )​



United States Patent Application
20200371076
Kind Code
A1
Koenig; Matthias
November 26, 2020

Method and Apparatus for Operating a Multi-Gas Sensor

Abstract
A method and apparatus for operating a multi-gas sensor are disclosed. In an embodiment, a method includes providing at least one calibration input comprising sensor design data of the multi-gas sensor, which varies dependent on production process parameters, and/or sensor production process parameter data of the multi-gas sensor, and/or measurement results of the multi-gas sensor captured when the multi-gas sensor is exposed to one of the gases or a gas mixture to be detected and/or sensed by the multi-gas sensor; providing a trained neural network including an input layer with K input nodes, an output layer with L output nodes and at least one hidden layer; storing each calibration input as a fixed input to a corresponding input node of the trained neural network; and providing a multi-gas sensor output for at least a part of the gases to be detected and/or sensed by the multi-gas sensor dependent on the trained neural network and actual measured sensor values from the sensor elements.


Inventors:​
Koenig; Matthias; (Muenchen, DE)
Applicant:​
Name​
City​
State​
Country​
Type​

TDK Electronics AG

Munich

DE​
Family ID:​
73052967
Appl. No.:​
16/878167
Filed:​
May 19, 2020
I’ll add this one in as well: Hopefully it’s useful.


CONTROLLER OF ARRAY INCLUDING NEUROMORPHIC ELEMENT, METHOD OF ARITHMETICALLY OPERATING DISCRETIZATION STEP SIZE, AND PROGRAM​

Feb 19, 2018 - TDK CORPORATION
A controller is a controller of an array including a neuromorphic element that multiplies a weight based on a value of a variable characteristic by a signal, and includes a control unit that controls the characteristic of the neuromorphic element by using a discretization step size obtained so that a predetermined condition for reducing an error or a predetermined condition for improving accuracy is satisfied on the basis of a case where a true value of the weight obtained with a higher accuracy than a resolution of the characteristic of the neuromorphic element is used and a case where a discretization step size which is set for the characteristic of the neuromorphic element is used.

:)
 
  • Like
  • Fire
Reactions: 20 users
  • Like
  • Love
  • Fire
Reactions: 29 users

Diogenese

Top 20
I found this patent for TDK involving NN however I don’t have the expertise to interrogate it. I’ll leave that to DIO.


AppFThdr.gif
[Help] [Home] [Boolean Search] [Manual] [Number Search] [PTDLs] [CURR_LIST]
[Shopping Cart] [Order Copy] [Image]


( 1 of 1 )​



United States Patent Application
20200371076
Kind Code
A1
Koenig; Matthias
November 26, 2020

Method and Apparatus for Operating a Multi-Gas Sensor

Abstract
A method and apparatus for operating a multi-gas sensor are disclosed. In an embodiment, a method includes providing at least one calibration input comprising sensor design data of the multi-gas sensor, which varies dependent on production process parameters, and/or sensor production process parameter data of the multi-gas sensor, and/or measurement results of the multi-gas sensor captured when the multi-gas sensor is exposed to one of the gases or a gas mixture to be detected and/or sensed by the multi-gas sensor; providing a trained neural network including an input layer with K input nodes, an output layer with L output nodes and at least one hidden layer; storing each calibration input as a fixed input to a corresponding input node of the trained neural network; and providing a multi-gas sensor output for at least a part of the gases to be detected and/or sensed by the multi-gas sensor dependent on the trained neural network and actual measured sensor values from the sensor elements.


Inventors:​
Koenig; Matthias; (Muenchen, DE)
Applicant:​
Name​
City​
State​
Country​
Type​

TDK Electronics AG

Munich

DE​
Family ID:​
73052967
Appl. No.:​
16/878167
Filed:​
May 19, 2020
The input signals are the activations, the Calibration inputs are the weights.

The trick of this invention is to enable gas sensors to be calibrated from sampling only a few of the target gasses using a NN.

There is no description of an NPU, so the NN is one someone baked earlier.

US2020371076A1 Method and Apparatus for Operating a Multi-Gas Sensor

1656940325475.png


1656939481615.png


1 . A method for operating a multi-gas sensor comprising multiple sensor elements, wherein the multi-gas sensor is configured to detect and/or sense a predefined number M of different gases, the method comprising:

providing at least one calibration input comprising:

sensor design data of the multi-gas sensor, which varies dependent on production process parameters; and/or

sensor production process parameter data of the multi-gas sensor; and/or

measurement results of the multi-gas sensor captured when the multi-gas sensor is exposed to one of the gases or a gas mixture to be detected and/or sensed by the multi-gas sensor;

providing a trained neural network comprising an input layer with K input nodes, an output layer with L output nodes and at least one hidden layer, wherein L, M and K are natural numbers;

storing each calibration input as a fixed input to a corresponding input node of the trained neural network; and

providing a multi-gas sensor output for at least a part of the gases to be detected and/or sensed by the multi-gas sensor dependent on the trained neural network and actual measured sensor values from the sensor elements, which are provided to corresponding input nodes of the trained neural network.


[0030] The respective gas sensor element 10 shown in FIG. 1 comprises for example a sensing layer 11 of metal oxide. The gas sensor elements 10 are, for instance, integrated with a CMOS circuitry (not shown) on a single chip. A stack of layers 13 is arranged on a semiconductor substrate 14 required for the CMOS circuitry. The respective gas sensor element 10 comprises a membrane. A portion of the semiconductor substrate 14 is, for instance, etched away to form a cavity 12 at the location of the sensing layer 11 . Remaining layers 13 and possibly a remaining portion of the substrate 14 form a thin membrane to support the layer 11 .

[0031] The respective sensor element 10 comprises a heating element 15 . The heating element 15 is embedded within the layer 13 and comprises conducting elements. The heating element 15 is configured to provide a local source of heat to heat the metal oxide layer 11 e.g., during operation of the gas sensor element 10 . The temperature can rise rapidly around the metal oxide layer 11 on the membrane, while a thicker part of the gas sensor chip, i.e. the portion where the substrate 14 is not removed, reacts with a slower rise of temperature due to its thermal inertia. By controlling the heating element 15 accordingly, the metal oxide layer 11 can be activated for a measurement and be regenerated afterwards.

[0032] Each of the metal oxide layers 11 is contacted by two conductive electrodes and hence acts as a resistor. In the presence of a compound its resistance changes, thereby providing a measure of a concentration of the compound in the immediate vicinity of the metal oxide sensing layer 11 . The change of the resistance and/or impedance can be measured by a voltage measurement.

[0035] Gas sensors have to be calibrated. The output signals of the gas sensor elements 10 are generally in the form of a voltage value. Calibration is needed to implement a relation between the gas sensor element 10 signal and the concentration level of the corresponding gas.

[0038] Because of manufacturing tolerances it is not possible to produce exact copies of a gas sensor in a production process. There are always small fluctuations in the provided output signals. This is the reason why nearly all gas sensor products need to be calibrated after assembly. This means that calibration data is determined and used during operation of the gas sensor to adjust the sensor signals of the sensor elements 10 to provide accurate measurement output signals.

[0046] Embodiments provide relevant or most relevant calibration information and perform some gas exposure tests, i.e. not for all gases but only for some, and train the neural network N such that the neural network N can be used to provide the calibration also for other untested gases.
 
  • Like
  • Fire
Reactions: 19 users
  • Like
  • Haha
  • Love
Reactions: 16 users
Top Bottom