OBSERVATIONS ON MAGIK-EYE & BRN

Some Observations about the partnership of BrainChipTechnologies and Magik-Eye Inc

This partnership is about vision and this is important because in formulating his theory of how to build artificial general intelligence the inventor of AKD1000 and joint founder of BrainChip Technologies Peter van der Made had as his very early goal the use of AKIDA technologies in this field.

When BrainChip Technologies commenced trading on the ASX in 2015 he was joined in this journey by Adam Osserian who was a Non Executive Director and Anil Mankar who was Chief Operating Officer.

It is important to note that both this distinguished gentleman are still on the journey with him. Adam Osserian currently is chairman of the Scientific Advisory Board and of course Anil Mankar is and has had the responsibility for overseeing the implementation of AKD1000 as an FPGA and designing in conjunction with Socionext the Chip called AKIDA1000.

The first incantation of Peter van der Made’s technology was released as licenced intellectual property in October, 2015 and it was called SNAP (Spiking Neuron Adaptive Processor).

It was revolutionary even by todays standards and failed to gain traction. In an effort to give it greater traction BrainChip on the 1^st September, 2016 acquired a French based company called Spikenet Technology a specialist Ai computer vision technology firm with a view to reintroduce SNAP technology to the market as a video processing solution.

In consequence of taking over Spikenet a relationship developed between BrainChip and Cerco Research Centre in France which saw two things occur. The first an enduring relationship with Simon Thorpe who knows sits on the BrainChip Scientific Advisory Board and the entering of an agreement on 20 MARCH, 2017 giving BrainChip exclusive rights to the JAST learning rules which Simon Thorpe had jointly developed.

The acquisition of Spikenet gave BrainChip a highly skilled team of computer vision experts and with the JAST learning rules on board a new upgraded state of the art video processing product was developed called BrainChip Studio. The brilliance of this product was recognised when on 21 August, 2017 it was awarded New Product of the Year for 2017 for Video Analytics by Security Today’s panel of independent experts.

Not content on 12 September, 2017 BrainChip Accelerator an 8-lane PCI Express add-in card that increased the BrainChip Studio’s speed an accuracy by a multiple of 6 making it 7 times more efficient than GPU-accelerated deep learning systems while being able to run 16 channels of simultaneous video processing.

There is more to this story but the point of including it here in a discussion of the partnership with Magik-Eye is to let new investors know that BrainChip is well versed in vision technologies and how it might be implemented with AKIDA SCNN (Spiking Convolutional Neural Network).

ABOUT MAGIK-EYE

It is a Japanese early stage company with a first of its kind patent protected 3D vision sensor. The patent to which I am about to refer is US20160117561A1 granted on 23 April, 2019. Those who are interested should read the entire document but I have included firstly the title of the invention and Abstract but from the perspective of my observations paragraph 54 is relevant to BrainChip’s partnership and how developed it might be with respect to the actual commercialisation of the Magik-Eye Akida 3D solution:

Distancesensor

Abstract

In one embodiment,a distance sensor includes an image capturing device positioned to capture animage of a field of view and a first plurality of projection points arrangedaround a first lens of the image capturing device, wherein each projectionpoint of the first plurality of projection points is configured to emit aplurality of projection beams in different directions within the field of view.

[0054] It should be noted that the present disclosurecan be implemented in software and/or in a combination of software andhardware, e.g., using application specific integrated circuits (ASIC), a programmablelogic array (PLA), including a field-programmable gate array (FPGA), ora state machine deployed on a hardware device, a general purpose computer orany other hardware equivalents, e.g., computer readable instructions pertainingto the method(s) discussed above can be used to configure a hardware processorto perform the steps, functions and/or operations of the above disclosedmethods. In one embodiment, instructions and data for the present moduleor process 605 for calculating distance (e.g., a software programcomprising computer-executable instructions) can be loaded into memory 604 and executedby hardware processor element 602 to implement the steps, functions or operations asdiscussed above in connection with the example method 400. Furthermore,when a hardware processor executes instructions to perform “operations”, thiscould include the hardware processor performing the operations directly and/orfacilitating, directing, or cooperating with another hardware device orcomponent (e.g., a co-processor and the like) to perform the operations

1. The first thing to note is that Magik-Eye canbe implemented in software.

AKIDA1000 ADE (AkidaDevelopment Environment) has been available to selected parties since 18 June2019 approximately two months after the grant of Magik-Eye’s patent and so ithas been possible for the implementation and trialling of the Magik-EyeAkida1000 3D Vision Sensor (hereafter MEA1000 3DVS) to have been engaged andcompleted.

2. The second thing to note is that Magic-Eye canbe implemented in hardware being a field – programmable gate array (FPGA)

AKIDA1000 was implemented asan FPGA and available form early 2020. It is well known that if you can successfully implement on FPGA you will be well positioned to fully understand what the performance of your technology will be when produced as a chip. If as I would like to think BrainChip made available the FPGA for testing by Magik-Eye with its Distance sensor and it worked as expected then the confidence which the recent announcement appears to evidence would from a scientific point of very be entirely justified.

3. Finally theDistance Sensor has a module which is so configured as to allow the function itperforms to be downloaded into the memory of a hardware processor and suchhardware processor could include the ability to perform once so trained toperform operations directly (read here independently) or in cooperation withanother hardware device.

[A sidenote by way of explanation taken from elsewhere in the patent. One Magik-Eye 3Dvision sensor has a field of vision up to 180 degrees. By placing two Magik-Eye 3D vision sensorsback to back this 180 degree vision becomes 360 degrees. In doing so a simplified form would involve havingtwo hardware devices one on each sensor capable of sharing information witheach other and acting cooperatively.]

AKIDA1000 has on chip memoryand a library for the purpose of storing the instructions downloaded from themodule. It has the capacity on chip to process these instructions and performoperations directly/independently or in cooperation with another AKIDA1000 processor.

Now of course it is possiblethat BrainChip and Magik-Eye only bumped into each other last week but thisdoes seem highly unlikely and so I am confident in opining that this is a welldeveloped relationship and that both BrainChip which has long history in visionprocessing and Magik-Eye which has developed a product which from its veryinception envisaged it being incorporated with an advanced system on a chipprocessor have been secretly working away at creating a market ready product.

TWO FINALOBSERVATIONS:

A.

I have tried very hard tofind a statement by Magik-Eye or anyone else as to what power their 3D VisionSensor draws. I have failed in this endeavour. The best I can get is a statement which occurs multiple times on Magik-Eye’s web site to the effect that it uses very little power by comparison with other sensors. When you look however at everything else that is available this seems very unusual until I realised this simple fact. Until they plug it into AKD1000 they do not have the ability to measure exactly how much power it will draw. This of course may not be a correct conclusion but there is no reason for the secrecy given their patent has been granted and published and they have told their competitors they use far less power than they do.

B.

I have been reading a greatdeal about Mobileye which is owned by Intel and its latest version of itsvision sensor due for release in 2021 called Q5. It is impressive however itfunctions by utilising an accelerator not a stand alone processor such as AKD1000(oops there is no other processor like AKIDA) and in operation uses 10 watts ofpower and will when utilised in autonomous vehicles involve 9 units beingdeployed. in fairness three of these units are for built in redundancy. However it does not HAVE THE ABILITY TOMEASURE DISTANCE TO OBJECTS THAT IT SEES.

WHEREAS, IN THEORY MAGIK-EYEWITH AKD1000, CAN COMPLETELY COVER THE VISION NEEDS OF AN AUTONOMOUS VEHICLEWITH JUST TWO 3D VISION SENSORS AT A SIGNIFICANTLY LOWER POWER REQUIREMENT WHILEMEASURING THE DISTANCE TO ALL OBJECTS IT ENCOUNTERS MOVING OR STATIONARY.

For anyone whomight have missed it the following is an article posted first by uiux regardingwhat has become known as electric vehicle range-anxiety and what Bosch istrying to do in this area:https://www.reuters.com/article/us-tech-bosch/bosch-to-make-silicon-carbide-chips-in-electric-vehicle-range-anxiety-play-idUSKBN1WM0YD Suffice to say that low power is veryattractive to the electric vehicle industry.

THIS IS MYOPINION ONLY SO DYOR.