Paper Title: Applying Persistent Memory to Neuromorphic Computing

Paper Abstract: Artificial Intelligence (AI) or Machine Learning (ML) implementations typically rely on constant access to cache memory for t learning or subsequent execution. Each layer implements a digital representation of multiple neurons, starting with a blank slate to start. The human brain does not provide constant access to remote cache-like memory for individual neurons; rather each neuronretains minimal information internally; and reacts to brain wave triggers to process neuron inputs. A group of persistent, byte-addressable, NVM neurons using flash memory technology can process neuron inputs using a mathematical model; a discrete-event simulation compares the effect of a slightly-slower flash technology to high speed, remote cache technology. Computer architects always look for ways to minimize remote memory accesses, and flash memory technology offers such an opportunity for AI implementations.

Paper Author: Darryl Koivisto, Chief Technology Officer, Mirabilis Design

Author Bio: Darryl Koivisto is the CTO and visionary at Mirabilis Design. He has over 30 years of experience as an architect, program manager, and modeling expert. Mr. Koivisto previously was the Principal Modeling Consultant at Cadence Design Systems and worked at Ford Aerospace, Signetics, and Amdahl. Mr. Koivisto has a DBA from Golden Gate University and an MS from Santa Clara University.