2020 BRN Discussion, page-16988

Yes you are correct.

As you know an EAP customer has had access to the ADE and as such should have built the model of what it would like to achieve running it on the ADE.

AERO is a perfect example as it used the ADE to test its efficiency on the 20 gas data set. Running on the ADE utilising only 200 neurons it achieved state of the art performance at identifying the various gases with a latency of 3 seconds.

The Boards incorporating AKD1000 are designed to swap over with the ADE so that you can just switch it on and see what the performance is running AERO for example on 1.2 million neurons in lieu of only 200.

Then you have the EAP customers who have had access to the IP with a view to implementing it on their own chip in an existing product. These customers are waiting for the Boards to verify that if they implement the IP that it will actually work.

You will also recall that the Boards have the ability to link with two other Boards so a customer such as NASA can also test to see what more than one AKD1000 chip ie three can do Performance wise and also how effective having three AKD1000 chips will be for preserv ing functionality on deep space missions.
Even in autonomous driving applications having a back up chip ready to takeover should a chip fail is critical so testing whether the AKD1000 shares its incrementally learned knowledge with the other chips so they are completely up to date when and if they have to take over is important.

Given what the CEO has told us about how successful validation has been and that all blocks are working exceptionally well there should not be any hiccups.

My opinion only.