2017 AFOSR MURIPO: Dr. Ali Sayir, Aerospace Materials for Extreme Environments PI: Dr. William Doolittle, Georgia Institute of Technology School of Electrical and Computer EngineeringMURI Website
We propose to implement a biomimetic, non-silicon, low power multi-layer 3D, adaptive neuristor with thermally engineered broadened timing windows useful for stochastic learning enhancements that combines Neuromorphic and Boolean computation for vision analysis. To do this, significant advancement in materials discovery relating to stoichiometry, correlation and disorder effects are proposed using state of the art toolsets providing discernment into the hidden quantum state density and charge occupancy internal state variables of adaptive oxides. This fundamental materials research will enable key device innovations including two never realized devices, and a new “Adaptive” building block akin to a biological neuron/axon. Using the novel quantum properties of these adaptive oxide materials, an adaptive time window for Spike Timing Dependent Plasticity (STDP), and new approaches to stochastic circuitry, a 3D scalable, ultra-low voltage (mV range) artificial retina will be demonstrated that can finally deliver the low power operation promised by Neuromorphic circuits. These innovations will leverage the adaptive oxide materials quantum properties into new devices and circuits to (1) exploit device noise to provide stochasticity and (2) control the programmable spike time window facilitating on-line, unsupervised learning.