Purely Spintronic Perceptron with Unsupervised Learning Enabled by Four-Terminal Domain Wall-Magnetic Tunnel Junction Neuron*

We propose a four-terminal domain wall-magnetic tunnel junction (DW-MTJ) neuron that enables the first-ever CMOS-free purely spintronic perceptron with unsupervised learning. The proposed leaky integrate-and-fire neuron has a ferromagnetic DW track coupled to a free layer of a binary MTJ by an electrically insulated layer. Current through the ferromagnetic track performs integration by moving the DW. Intrinsic leaking occurs by moving the DW in the opposite direction of integration due to either dipolar magnetic field, anisotropy gradient, or shape variation. When the DW passes underneath the electrically-isolated MTJ, it fires by switching between the anti-parallel resistive state and parallel conductive state. Additionally, by exploiting stray magnetic field interactions, these neurons perform lateral inhibition.
In a crossbar perceptron, the DW track of each four-terminal neuron is connected to the analog three-terminal DW-MTJ synapses. Finally, an unsupervised learning algorithm results from the feedback between the neuron MTJ and the third terminals of the analog synapses, providing best results of 98.11% accuracy on the Wisconsin breast cancer clustering task.