Design of self-adaptive oscillating neurons using electrically reconfigurable skyrmion lattices

Taking inspiration from the brain, neuromorphic computing promises to actualize the transformative potential of Artificial Intelligence (AI) by providing a path for ultra-low power AI implementation. Moreover, mimicking the complex and advanced properties of the brain can deliver a more powerful form of computation than is currently available. Here, we design and simulate a novel artificial neuron that incorporates two advanced neural behaviors: oscillatory dynamics and neuromodulation. Neuromodulation is the self-adaptive ability of a neuron to regulate its dynamics in response to its environment and contextual cues. The artificial neuron is implemented with a lattice of five magnetic skyrmions in a bilayer of insulating thulium iron garnet (TmIG) and platinum (Pt). The oscillatory dynamics of the coupled skyrmions has a multi-frequent spectrum which provides the neuron with a rich basis for information representation. Neuromodulation is enabled by the reconfigurability of the skyrmion lattice: individual skyrmions can be manipulated by electrical currents to change their arrangement in the lattice, which shifts the resonant frequencies and modulates the amplitudes of oscillatory outputs of the neuron in response to the same external excitation. Bio-mimicking dynamics such as bursting are shown. The results can be used to implement advanced neuromorphic applications including burst-coding, motion detection, cognition, brain-machine interfaces and attention-based learning.