Intel Labs proposes Spintronic and Memristor based Neuromorphic chip architecture

A new paper titled “Proposal For Neuromorphic Hardware Using Spin Devices” put out by the researchers at the Circuits Research Lab at Intel and Purdue University shows promise in simulations for modeling bipolar spin neurons utilizing memristive cross-bar neural network (NN) architectures as synapses:

[...] Emergence of programmable conductive elements, like TiO2 memristor and phase change memory (PCM), have provided a compact model for neurons. However current-mode signal processing with CMOS-neurons in such circuits leads to power hungry designs. Application of the proposed bipolar spin-neuron in such a design, however, can lead to ultra low power NN hardware. [1]

Cross-bar neural network architecture using memristor (/PCM) synapses and spin neurons:

Differential MTJ latch for inter-neuron current-mode signaling using deep triode current source (DTCS) transistor:


From the paper abstract of ["Proposal For Neuromorphic Hardware Using Spin Devices"]