|
Big data generated by Internet-of-Things (IoT) sensors and devices along with AI is changing the society we live in, allowing us to better understand and manage the world. Both IoT and AI require low energy information handling. For example, IoT devices are much better if it does not require a battery (i.e. energized by energy harvesting) and AI is known to be power-hungry calling for a means to drastically reduce the power for its ubiquitous use. Spintronics nonvolatile device, magnetic tunnel junction (MTJ), has been shown to reduce the operation power of a CMOS-based microprocessor by up to two orders of magnitude, suitable for IoT purposes and other applications. MTJs have also been shown to be scalable down to 2.3 nm without resorting to new materials, a significant advantage along with their high thermal stability, high endurance, and low voltage operation. Moreover, new computing schemes are emerging, where one expects to have higher performance than relying solely on silicon-based digital processing. One of the approaches is neuromorphic computing; we have made proof-of-concept spintronics devices for artificial synapses as well as neurons for neuromorphic applications. Another approach utilizes less stable MTJs for a novel form of computing, probabilistic computing, to address optimization problems: One can formulate integer factorization as an optimization problem in such a way that the most preferred state in terms of energy gives the factorized result. The time scale involved in these probabilistic MTJs is also discussed. If time allows, I will touch upon synthetic antiferromagnetic skyrmions for information carriers
|
