The von Neumann architecture was also inspired by the working of the human brain and that is why the specific terminology was used (e.g. memory.) Apparently the perception of brain operation has changed since, as it was much more mechanistic back then (i.e. processor.) I wonder how this compares to the IBM's cognitive computing chips.
While I don't doubt the work is a "massively parallel computing platform", like the IBM chips, but it's similarity or structure is at best only tangentially reminiscent of the brain.
The little we do know about the brain, or any other part of our systems, it functions in far more complex ways than "massively parallel" describes. In a very sketchy way, brains work as distributed circuits of distributed circuits built on sets of massively parallel units whose subunits are variably interacting and in parallel with other similarly, but not identically constructed massively parallel units.
I don't know what I said is true, but comparing human-invented devices to even a "simple" brain is much more marketing buzzword than anything like reality.
I guess the system is programmable. The point is it distributes memory accesses and computations, which should allow things which usually run slow to run in real-time.
I'm still waiting for an architecture like a distributed programmable memory rather than a distributed programmable computer. I think that's what we're aiming for, really. AI doesn't consume CPU it consumes memory cycles.
It would be really interesting to see what could be done by coupling memory and some SIMD capability very closely. Would probably be some crazy expensive memory, but it could be memory that offloads a lot out of the CPU.
" Three of the principle axioms of parallel machine design - memory coherence, synchronicity and determinism - have been discarded in the design without, surprisingly, compromising the ability to perform meaningful computations. "
