Typical applications of high-performance magnetic tunnel junctions in magnetic random access memory and artificial intelligence

Spintronic devices, with the magnetic tunnel junction (MTJ) as a typical representative, due to their excellent advantages such as nonvolatility, high speed, low energy consumption, long data retention time, long endurance, small size, CMOS compatibility, radiation resistance, etc., have been successfully applied in the fields of magnetic sensors, hard disk magnetic read heads, and magnetic random access memories (MRAMs) since their birth, profoundly reshaping the landscape of the magnetic storage industry. This paper reviews the evolution process of MTJ devices, and then introduces the latest research and development progress in spin transfer torque (STT), especially in spin orbit torque-MTJ (SOT-MTJ) and SOT-MRAM. Finally, it elaborates on a new category of application scenarios of MTJ, namely true random number generators (TRNGs) which act as probabilistic bits for probabilistic neural networks and artificial intelligence. It mainly focuses on two aspects: Boltzmann machines and generative artificial intelligence, and random number generators for arbitrary probability distribution functions. This direction is expected to open up new application fields for spintronic devices that differentiate them from other new types of nonvolatile memories, and bring new potentials for the sustainable development of spintronics.