See I want to point out I am no expert and I am just taking a stab at doing my in-depth research. So there is possibly a lot left to imagination and maybe error in what I say.
So there are a couple of core problems worth consideration.
1) Do we need the processor to simulate the brain or an equivalent function is good enough ?
The problem with simulating the brain is that you need a deep understanding of the brain which is currently inadequate at best. The ANN models that we have to test these simulation are all "hush hush" and obviously without a deep understanding of human brain the ANN models are also inadequate.
Think about creating a dog. If simulating you would probably need to create legs , eyes and nose. However if the benefit equivalent is good enough then ensuring the safety of a subject (What a dog serves as a purpose) may work.
2) Why is DARPA and other tech companies interested in such technologies ? Because the problems that they face and the analysis of the datasets they require needs computational power in the order of the magnitude much higher than what the current technology can fulfill even if Moore's law was correct. Google may want to show you that your home is 40 mins away without utilizing 400 watts and do this in a manner that scales efficiently.
Think about these 2 things for a second. Now think if you had some instance of a processor which combined serves a purpose of both storing information and processing it (human brain) what you could do ?
Peter is certainly not the first one to invent it and the memristor based chips have existed for almost 30 years. However BRN will have a first mover advantage in this field if they take the project in a more open source way as opposed to keeping it closed. For example Google have captured exceptionally large market by open sourcing Android....
So keeping these in mind to answer your question
Point 0031 refers to a working mechanism of ANSC which he then goes on to explain
Point 0032 refers to how the information because being boolean (Synapses will either carry 0 or 1) can then be fed back and can be computer (perception)
From which point on gets a little complex and without spending adequate time in understanding the design I am not too qualified to comment on the matter.
The latest qualcomm processors have tried to address dendritic dynamics and have cited Peter's work which is encouraging
see this patent which refers to Peter's patent. However cell simulation is going to be an extremely challenging and probably beyond a 500 mil project.
Being a small sized and a start-up I would expect they will get a working chip which can do simple classification , perception and other cognitive tasks at which point mcap will reflect maybe 500 MIL. It is at this point and maybe 2 years from now when the actual value of this companies potential be reflected dependent upon what area they choose to work on.
DARPA's purpose to develop these chips was to be used in argus drones. Think about it. If the drone is in enemy territory and they need to keep track of multiple targets a simulation from a desktop would lag and possibly not be able to keep accurate measure of each target. Hence they need such a chip on board the drone which learns and acts based on the available dataset and also maintaining precision and knowledge of each target.
While IBM is in bed with DARPA , there exists a vast field for small players like BRN to come out with a working prototype which can be licenced out for a small royalty per chip and hence my enthusiasm.
(20min delay)