We have an exciting paper just published on leaky, integrate, and fire (LIF) artificial neurons using magnetic domain walls, read out by magnetic tunnel junctions (MTJs). As predicted by simulation e.g. by Wesley “Gunther” Brigner and Joseph S. Friedman, the neurons naturally leak due to domain wall movement to the narrow part of the track, allowing implementation of the bio-mimetic LIF neuron model without energy cost to leaking. By engineering an asymmetric notch in the tapered domain wall track, we control the leaking and integrating behavior via relaxation time between integration pulses. Given MTJs are compatible with CMOS+X timescales and fabrication, this is an exciting direction for implementing efficient neuron function at the device level for neuromorphic computing. 

The paper was co-led by Dr. Thomas Leonard and Nicholas Zogbi and a collaboration with Christopher H. Bennett supported by Sandia National Labs. 

Click here to read it!