Antiferromagnet-based Neuromorphics using dynamics of topological charges
ORAL
Abstract
Compared with conventional computers, the human brain is extremely energy efficient and can perform complicated cognitive tasks. Spintronics-based hardware implementations of neuromorphic computing has several advantages, such as low energy dissipation, consistent material platform, and nonvolatile memory. In this work, we propose a new candidate for neuromorphic hardware based on the dynamics of topological charges in an antiferromagnet. The two basic elements—neuron and synapse—are realized and we demonstrate their functionalities that are crucial to form a spiking neural network. A single magnetic domain exhibits binary switch and performs leaky integrate-and-fire as a neuron. A synapse with spike-timing-dependent plasticity is realized by a one-dimensional interacting gas of domain walls, where the synaptic weight is simply the degree of saturation of domain walls. Due to the full compatibility of our artificial neurons and synapses, Hebbian learning can be achieved by simple action rules in the connected system.
–
Presenters
-
Shu Zhang
University of California, Los Angeles, Physics, University of California, Los Angeles, Johns Hopkins University
Authors
-
Shu Zhang
University of California, Los Angeles, Physics, University of California, Los Angeles, Johns Hopkins University
-
Yaroslav Tserkovnyak
Physics and Astronomy, University of California, Los Angeles, University of California, Los Angeles, Physics, University of California, Los Angeles, Physics & Astronomy, University of California, Los Angeles