Hi Bravo,Well, seeing I'm as good at engineering as I was at math - the only thing I took out of that was that they need some sort of electronic component. Is that worth a least a quarter of a star?
View attachment 27287
Well, they've been footling around with MemRistors for 20 years and:-
"In a presentation titled "Resistive Memories-Based Concepts for Neuromorphic Computing", Elisa Vianello, CEA-Leti's edge AI program manager, said RRAMs, aka memristors, offer advantages in energy efficiency and computing power when processing AI workloads. She noted, however, scientists must overcome device issues, especially variability, quantization error and limited endurance to achieve commercialization of this approach."
Many AI researchers have been beguiled by the close similarity between synapses and ReRAM in that ReRAM theoretically can be arranged to directly add the strengths of incoming signals (analog), whereas digital synapses such as Akida counts binary bits to arrive at the sum (digital). But, as Elias points out, there are practical difficulties in making the ReRAM devices sufficiently precise to produce consistent results, remembering that we are talking about millions of ReRAM devices so cumulative errors can be significant.
So, in theory, ReRAM has advantages over digital, but, in practice, digital works.
Like Renesas, they have sunk costs in their in-house tech.
Last edited: