Magnetic Straintronics for Ultra-Energy-Efficient Unconventional Computing: A Review

• Published in: IEEE transactions on magnetics Vol. 60; no. 9; pp. 1 - 10
• Main Author: Bandyopadhyay, Supriyo
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.09.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Analog and binary stochastic neurons; Bayesian inference engines; Belief networks; Boltzmann machine (BM) and Ising machine (IM); Cybersecurity; Energy efficiency; Hardware; Ising model; magnetic straintronics; Magnetic tunneling; Magnetization; Magnetostriction; matrix multiplication hardware; neuromorphic architectures; Neurons; Strain; Switches
• ISSN: 0018-9464; 1941-0069
• DOI: 10.1109/TMAG.2024.3403973

With rapidly increasing edge intelligence, domain-specific computers in heterogeneous fabrics are likely to rule the roost. Judicious choice of device technology and computational paradigms can drastically reduce the size, weight, and power (SWaP) of such computers, while also making them fully autonomous (clockless) and resilient against malicious attacks. Here, we review the promise of an emerging device technology-magnetic straintronics-in implementing extremely energy-efficient hardware for a wide variety of computing paradigms: neuromorphic, probabilistic, Bayesian belief networks, Boltzmann (BM) and Ising machines (IMs), matrix multipliers for deep learning networks, and reconfigurable stochastic neurons for p-computing. Magnetic straintronics has two important features-non-volatility and very low energy expenditure-which are conducive to edge processing and hardware cybersecurity. We discuss some unconventional computing paradigms implemented with magnetic straintronics while pointing out the remarkable energy efficiency in all cases.