HOW ADVANCED IS AKIDA TECHNOLOGY?
I often think about this question and then today an article was posted regarding Intel entering an agreement with Sandia National Laboratories to explore the value of neuromorphic computing for scaled-up Ai problems.
https://venturebeat.com/2020/10/02/intel-inks-agreement-with-sandia-national-laboratories-to-explore-neuromorphic-computing/
It is a reasonably large article and I skimmed over it and in so doing saw a reference to Huawei which caught my attention. Reading more closely I discovered that Intel had won a license to supply Hauwei with a 14 nanometer Loihi chip containing over 2 billion transistors, 130,000 artificial neurons and 130 million synapses.
The first thing that stood out was the number of neurons and synapses against AKD1000.
AKD1000 has 1.2 million neurons while this Loihi Chip has 130,000 artificial neurons.
AKD1000 has 10 billion synapses this Loihi Chip has 130 million synapses.
The second thing that stands out is that this Loihi Chip is to be implemented in 14nm whereas AKD1000 is implemented on 28nm.
The CEO of Brainchip was asked recently by a shareholder whether it was possible to implement AKD1000 on smaller chips and he stated that as it was fully digital it was directly scalable down to 14nm and 7nm with the usual reductions in energy requirements and improvements in performance.
Using simple logic it cannot be that by implementing this Loihi Chip on 14nm accounts for why there is a much reduced neuron and synapses count over AKD1000. Even if you gang two Loihi Chips together you only get 260,000 neurons and 260 million synapses.
This caused me to ponder the fact that Intel has received a licence to supply their Loihi Chip to Huawei. How could this be?
Huawei is a Chinese owned company. Intel is working with DARPA and NASA in defence and space projects. In the USA and around the world Huawei has seen ever increasing concerns across various governments and their agencies about the security implications of engagements with Huawei particularly in the area of communications.
Then it hit me Intel some time ago released its Loihi technology to the world at large. Its’ stated intention being to bring the academic and scientific communities from around the world to together in a joint effort to make it work.
This means there is absolutely nothing secret about Loihi and the underlying technology which needs protecting from China.
On the other side of this coin I recently posted an extract from a presentation by the CEO of Brainchip from 2018 where he stated that they were exploring the use of AKIDA technology in mobile phones. He stated that they were in discussions with a mobile phone manufacturer and had meetings and he threw out three names one of which was Huawei.
Those who have held their BRN shares for some time or those who have actually done some research will remember that the CEO last year before Covid-19 spoke lovingly about his trips to China in particular Shanghai and how much was going on over there and how Brainchip’s strategic plan involved an Innovation Centre in Shanghai and that steps in this direction were well advanced.
Then Covid-19 struck and the CEO earlier this year complained that if not for Covid-19 he would have been in China, Korea and Japan for various meetings and would have been there for about 6 weeks but that this had all had to be shelved but that things were still progressing electronically.
Subsequently in a further CEO presentation to shareholders he was asked about how China/Shanghai was progressing and he deflected this question stating that it had been handed off to Roger Levinson.
Then we had the great uiux reveal of his NASA Tensor Innovation Partners discovery that they had successfully won a Phase 1 grant from NASA based upon the implementation of AKIDA technology in a CubeSat which as we know has moved on to Phase 2 back in August, 2020.
Various posters then put up screen shots from Tensor Innovation Partners web site where they were clearly boasting about the use of AKIDA technology in their CubeSat design and about all of the other uses they proposed across space and defence applications. Various shareholders bombarded Brainchip with questions about this relationship and the next thing we know is that Tensor Innovation Partners web site was amended by removing the AKIDA and Brainchip references but it still referred to the use of this type of neuromorphic technology and no longer referred to the other space and defence applications.
As luck would have it not long after there was another presentation to shareholders scheduled where the CEO addressed a question from a shareholder to the effect did ‘Brainchip have a contract with NASA’ which he denied but made no reference at all to Tensor. There is no doubt that the company received communications from shareholders about Tensor as one of the posters on HC kindly put up an email from Ken Scarince the CFO wherein he stated that they had certainly been speaking to Tensor but did not know where it was up too and would make enquiries. He never came back to the poster.
The denial by the CEO concerning NASA at that time was with hindsight obviously the truth as any agreement that existed would have been with Tensor. The fact remains as many other posters have continued to check the NASA web site it still contains the original Phase 1 application by Tensor where it is basing its application to build CubeSat on AKIDA technology.
In this same presentation the CEO made clear that Brainchip was no longer involved with China in any way shape or form and it was obvious that this was not a short term state of affairs as he stated that Brainchip did not need China. Some might consider a company saying it did not need China one of the largest markets in the world as rather extraordinary.
Not long after this we had the formal announcement of the Vorago and NASA agreement/relationship for the purpose of hardening the AKD1000 chip against radiation and extreme conditions encountered in space.
While I do not have access to top secret government and NASA documents it seems reasonable for me as a shareholder to assume that the abandoning of China as a market is in consequence of the relationship with NASA/DARPA and the potential security implications of AKIDA technology coming into the possession of the Chinese.
Yet Intel which is ostensibly developing the same technology has no such problem how can this be?
Surely if Intel has let the cat out of the neuromorphic bag then there can be nothing about the AKIDA technology that would be secret?
Well this is where the wild speculation comes into my thinking.
We know that the patents lodged for Brainchip’s AKIDA technology have been sited by Intel and others. We know that to some extent Intel and others have had to modify their patent applications so as not to infringe Brainchip’s patents.
The significance of these events is a strong indication that the patents of Intel and others and the claims made therein were considered by the US Patent office to be materially different from the technology patented by Brainchip as Intel and others did eventually obtain their patents.
Thus AKD1000 contains technology which is significantly different to Loihi Chip technology and others. This difference in the technology therefore means that the general release by Intel of the Loihi technology for research world wide has not given away that which makes AKD1000 revolutionary.
This being the case it makes complete sense as to why Intel can deal with Huawei and Brainchip has had to abandon any contact with China as part of its involvement with NASA/DARPA.
If this analysis is correct then in 2018 at the Brainchip AGM in Sydney when Peter van der Made in his presentation to shareholders regarding the AKIDA technology stated that his invention had a secret sauce which gave it a significant edge over the competition including Loihi he was not exaggerating. Again long term shareholders and those who have done their research will know that on more than one occasion the CEO in his presentations has referred to ‘Peter’s secret sauce’ as the reason AKIDA technology is ground breaking.
Now as I have said this is speculation based upon known facts so I would ask that you do your own research and decide for yourself just how advanced is AKIDA technology.
I will finish here with the following comparison:
AKD1000 can be scaled up by a factor of 64 chips which would be 76.8 million neurons and 640 billion synapses. If you accept as Brainchip claims AKD1000 is expected to use uni watts to milli watts of power depending on manner of operation then allowing on a but say basis for it to use a whole watt per chip when scaled to 64 chips it would use 64 watts of power. (1)
Intel’s Pohoiki Springs, which was announced in July 2019, issimilar in neural capacity to the brain of a small mammal, with 768 Loihi chipsand 100 million neurons spread across 24 Arria10 FPGA Nahuku expansion boards(containing 32 chips each) that operate at under 500 watts.
1. In the Intel article referenced above it states that scaling up of Loihi Chip as proposed with Scandia will increase overall power consumption by 30% then by using a but say power consumption of 1 watt for the purpose of these calculations is probably higher than would be the case for AKD1000. The number of AKD1000 chips that can be scaled is restricted as part of the design decided upon by Brainchip given that it is targeted at the edge. Brainchip has always stated that if a use case arose the design could be modified to allow more than 64 chips up to 1024.)
My opinion only DYOR.
(20min delay)