There seems to be an influx of people questioning the veracity of our founder Peter van der Made's claim that Brainchip is 3 years ahead of the competition.
Not being one to take things at face value, I did my research and am convinced that he's not only correct, but maybe also a little too conservative.
In this thread, we will present to you why we believe this to be so. I'm going to set the framework here and leave it to the respective domain experts to follow up.
1. Brainchip's IP and chips are in at least 3 commercial products to be made available shortly.
Renesas (model number not announced yet)
Socionext SynQuacer SC2A11
MagikEye DK-ILT001
No other neuromorphic chip companies that I know of can claim the same. A few have announced collaborations recently but do not have any commercially available products out.
2. Company growth stage.
Unlike the competition, which is mainly made up of startups in the research & development stage, Brainchip has already had at least 2 production runs of their Akida 1000 chips, including the Engineering Samples in the first batch and the latest batch of chips being tested / tested. In addition to this the Akida 500 & 2000 are nearing finalisation and work has already begun on Akida 3000.
3. Incumbents are flummoxed
The 2 neuromorphic chips that have received the most attention so far are computing giants IBM's TrueNorth and Intel's Loihi. Both chips have not had any commercial products using them (afaik), nor have volume production to date, and many of their features (number of neurons / synapses, power consumption, compatible neural networks / training models, etc) pale in comparison to Brainchip's Akida 1000.
4. Brainchip's patents are ahead of the competition.
There are at least 13 patents dating from 2015 (
SIX YEARS ago as of today) that have been assigned to Brainchip covering a whole range of neuromorphic technologies and applications. Many competitor's patent applications have had to be modified because they ran up against Brainchip's patents (
@BarrelSitter /
@uiux /
@Fact Finder and other resident patent meisters can fill you in on these). Furthermore, many "neuromorphic chip" companies merely are mathematical accelerators by design, they do not actually process information in ways that resemble the brain's neurons and synapses like Brainchip's Akida family of chips, but merely perform mathematical gymnastics to process information faster than conventional chips.
https://assignment.uspto.gov/patent/index.html#/patent/search/resultFilter?advSearchFilter=patAssigneeName:%22brainchip%22&qc=1Brainchip also has exclusive licence to the JAST learning rules which are considered by some to be essential for a true SNN to work at low power. See discussions below;
https://hotcopper.com.au/threads/jast-learning-rules.4419679/#post-35484559https://hotcopper.com.au/threads/all-roads-lead-to-jast.5683524/#post-48060693However the competition are only just waking up to the potential of Spiking Neural Networks (SNNs). Many of their patents are only in the application stages (not granted yet).
https://www.theengineer.co.uk/spiking-neural-network-ai-kaust/https://www.technologynetworks.com/neuroscience/news/why-a-brain-on-a-chip-would-need-little-training-347986https://semiengineering.com/spiking-neural-networks-research-projects-or-commercial-products/
(20min delay)