Hi Dr,
I'm think your comment that the company had not sufficiently engaged with customers can be viewed in the light that the Akda "engineering" version was distributed to EAPs and the design modified in response to feedback. As Sean said, Akida 1 had a limited field of application.
Akida 1 was produced starting in late 2021.
There was talk of LSTM, but then transformers, which only emerged in 2020, became more widely known after the commercial chip production. I haven't got a full understanding of transformers, but f\From what little I've been able to glean, I believe transformers are simpler and more efficient than LSTM. I don't know the details of Akida 2, but I understand that it is capable of performing time-based functions (speed, direction, speech, ...) on-chip, relieving the CPU of these tasks, hence further expediting the process.
And development is ongoing - we continue to work with customers to develop silicon which improves their products in a meaningful way.
So, from the meeting, we learned that in the real world Akida 1 was found to have a limited scope of application but is being incorporated in chips (Renesas, MegaChips), and, from the real world, we learned that the technological environment is not treading water, skipping from LSTMs to transformers in the blink of an eye.
The remarkable thing is that BrainChip has the agility to switch from Akida 1 to LSTM to transformer in such a short time.
This is a much harder journey then most of us, shareholders and company alike, anticipated, exacerbated by the global economic conditions and other factors.
There are major barriers to entry for the customers of IP licensing - licence fee and cost of designing and manufacturing chips. This excludes most smaller potential customers. They will need to obtain chips from MegaChips.
Otherwise we could monetize our model libraries, and possibly licence our Akida simulation software for commercial use.
