Electrical manipulation of magnetic domain wall motion and its applications constitute a long-standing central topic of spintronics. Recent enthusiasm for magnetically tunable diffractive optics calls for a fast and efficient method of device reconfiguration. Here we develop a chiral domain-wall racetrack grating device whose periodic magnetization structures can be electrically reconfigured by spin-orbit torques without the assistance of external magnetic fields, demonstrating a fast, efficient, and continuous tuning of optical diffraction. We further show that these unique features can be utilized to emulate the synaptic potentiation and depression and achieve a pattern recognition accuracy of 92% when integrated into an artificial neural network. The continuously tunable domain wall position in combination with the inherent dispersion ability of grating provides an analog and multichannel solution to the network, with the potential of up to 252 synapse levels and a parallel capacity of several thousand. These results are expected not only to establish a transformative pathway to fast and efficient light field manipulation, particularly with applications in high-precision and massive-parallelism optoelectronic neuromorphic computing, but also to pioneer unprecedented optoelectronic applications of magnetic domain-wall racetrack devices beyond their traditional memory and logic functionalities.
Huang et al. (Thu,) studied this question.