Spintronic Computing Architecture and Method

A method of performing unsupervised learning within a neural network, the method comprising:

• applying input currents and/or voltages as inputs to the neural network, wherein the neural network comprises:
• a crossbar array of domain wall magnetic tunnel junction artificial synapses, each comprising a ferromagnetic domain wall track within which is a magnetic domain wall, wherein the domain wall track has a first fixed magnetization region at a first end and a second fixed magnetization region at a second end, and wherein the second fixed magnetization region has a magnetic direction opposite to the first fixed magnetization region;
• a magnetic tunnel junction between the first and second ends of the domain wall track, wherein the magnetic tunnel junction comprises a fixed ferromagnet;
• a tunnel barrier beneath the fixed ferromagnet;
• and the domain wall track beneath the tunnel barrier;
• a plurality of domain wall magnetic tunnel junction artificial neurons, each comprising a ferromagnetic domain wall track within which is a magnetic domain wall, wherein the domain wall track has a first fixed magnetization region at a first end and a second fixed magnetization region at a second end, wherein the second fixed magnetization region has a magnetic direction opposite to the first fixed magnetization region, and wherein the domain wall track comprises an energy gradient between the first and second ends;
• a magnetic tunnel junction between the first and second ends of the domain wall track, wherein the magnetic tunnel junction comprises a fixed ferromagnet;
• a tunnel barrier beneath the fixed ferromagnet;
• a free ferromagnet beneath the tunnel barrier;
• and
• an electrically insulating, magnetic coupling layer between the free ferromagnet and the domain wall track, wherein the free ferromagnet is electrically isolated from, and magnetically coupled to, the domain wall track;
• input signals to the crossbar synapse array provided to the first ends of the domain wall tracks of the synapses;
• connections between the fixed ferromagnets of the synapses and the first ends of the domain wall tracks of the neurons;
• and
• connections between the second ends of the domain wall tracks of the synapses and the free ferromagnets within the neurons;
• wherein the domain walls within the synapse domain wall tracks move in response to the voltages and/or currents applied by input signals to the first ends of the synapse domain wall tracks relative to the firing signals provided by the neurons to the second ends of the synapse domain wall tracks, resulting in unsupervised learning.