Brain-Inspired Computing Using Magnetic Domain Wall Devices

Neuromorphic computing or brain-inspired computing is considered as a potential solution to overcome the energy inefficiency of the von Neumann architecture for artificial intelligence applications [1]-[4]. In order to realize spin-based neuromorphic computing practically, it is essential to design and fabricate electronic analogues of neurons and synapses. An electronic analogue of a synaptic device should provide multiple resistance states. A neuron device should receive multiple inputs and should provide a pulse output when the summation of the multiple inputs exceeds a threshold.

We have been carrying out investigations on the design and development of various synaptic and neuron devices in our laboratory. Domain wall (DW) devices based on magnetic tunnel junctions (MTJs), where the DW can be moved by spin-orbit torque, are suitable candidates for the fabrication of synaptic and neuron devices [2]. Spin-orbit torque helps in achieving DW motion at low energies whereas the use of MTJs helps in translating DW position information into resistance levels (or voltage pulses) [3]. This talk will summarize various designs of synthetic neurons synaptic elements and materials [4]. The first half of the talk will be at an introductory level, aimed at first-year graduate students. The second half will provide details of the latest research.

[1] K. Roy, A Jaiswal, and P Panda, “Towards Spike-Based Machine Intelligence With Neuromorphic Computing,” Nature 575, 607-617 (2019).

[2] W. L. W. Mah, J. P. Chan, K. R. Ganesh, V. B. Naik, S. N. Piramanayagam, “Leakage Function in Magnetic Domain Wall Based Artificial Neuron Using Stray Field,” Appl. Phys. Lett. 123, 092401 (2023).

[3] D. Kumar, H. J. Chung, J. P. Chan, T. L. Jin, S. T. Lim, S. S. P. Parkin, R. Sbiaa, S. N. Piramanayagam, “Ultralow Energy Domain Wall Device for Spin-Based Neuromorphic Computing,” ACS Nano 17, 6261-6274 (2023).

[4] R. Maddu, D. Kumar, S. Bhatti, S. N. Piramanayagam, “Spintronic Heterostructures for Artificial Intelligence: A Materials Perspective,” Phys. Stat. Sol. RRL 17, 2200493 (2023).