Neuromorphic spintronics

Abstract

Neuromorphic computing uses basic principles inspired by the brain to design circuits that perform artificial intelligence tasks with superior energy efficiency. Traditional approaches have been limited by the energy area of artificial neurons and synapses realized with conventional electronic devices. In recent years, multiple groups have demonstrated that spintronic nanodevices, which exploit the magnetic as well as electrical properties of electrons, can increase the energy efficiency and decrease the area of these circuits. Among the variety of spintronic devices that have been used, magnetic tunnel junctions play a prominent role because of their established compatibility with standard integrated circuits and their multifunctionality. Magnetic tunnel junctions can serve as synapses, storing connection weights, functioning as local, nonvolatile digital memory or as continuously varying resistances. As nano-oscillators, they can serve as neurons, emulating the oscillatory behavior of sets of biological neurons. As superparamagnets, they can do so by emulating the random spiking of biological neurons. Magnetic textures like domain walls or skyrmions can be configured to function as neurons through their non-linear dynamics. Several implementations of neuromorphic computing with spintronic devices demonstrate their promise in this context. Used as variable resistance synapses, magnetic tunnel junctions perform pattern recognition in an associative memory. As oscillators, they perform spoken digit recognition in reservoir computing and when coupled together, classification of signals. As superparamagnets, they perform population coding and probabilistic computing. Simulations demonstrate that arrays of nanomagnets and films of skyrmions can operate as components of neuromorphic computers. While these examples show the unique promise of spintronics in this field, there are several challenges to scaling up, including the efficiency of coupling between devices and the relatively low ratio of maximum to minimum resistances in the individual devices. I-Neuromorphic computing is the path to low energy Artificial Intelligence Artificial Intelligence has experienced unprecedented progress in recent years, promising to transform multiples areas of how we live and how we work. However, this development comes with a considerable challenge: the energy consumption associated with existing approaches 1 , making it

Atomistic origin of exchange anisotropy in noncollinear γ−IrMn3 –CoFe bilayers

“…Our findings provide new insight into the physical origin of exchange anisotropy by accounting for the correct nature of the antiferromagnetic spin structure and crystallography, finally resolving one of the most complex and outstanding challenges in the field. The enhanced understanding will provide new routes for optimisation of nanoscale exchange biased systems with relevance to upcoming neuromorphic [35] and antiferromagnetic spintronic [1,2] devices.…”

Current-controlled nanomagnetic writing for reconfigurable magnonic crystals

“…ersatile, low-power means to selectively control magnetization states at the nanoscale are critical across a host of applications, both in fundamental science and deviceoriented systems. Alongside mature technologies such as data storage, nanomagnetic arrays support a host of more recent applications including neuromorphic computation [1][2][3][4][5] , superconducting vortex control [6][7][8][9] and reconfigurable magnonic crystals [10][11][12][13][14]. RMCs are nanopatterned metamaterials harnessing varying magnetic configurations to manipulate and store information by tuning magnonic (spin-wave) dynamics [15][16][17][18][19][20][21][22][23][24][25][26] .…”

Prospect of Spin-Orbitronic Devices and Their Applications

“…In particular, an artificial synapse requires multilevel differentiated nonvolatile states, e.g., a memristive device, to store synaptic weights via implementing the analog of a variable resistor. Although spin-orbitronic devices show better performance indexes over other memristor candidates, however, the characteristic binary resistance nature of generic MTJ makes it difficult to proceed implementing as a multilevel synapse rather than a stochastic binary synapse ( Grollier et al., 2020 ). One representative solution is to utilize the current-induced domain wall motion within the FM free layer ( Sengupta et al., 2015a ; Lequeux et al., 2016 ; Yue et al., 2019 ; Yang et al., 2019c ; Siddiqui et al., 2019 ; Azam et al., 2020 ; Zhang et al., 2019b ), for instance, by tuning the pinning potential of FM domain wall motions, SOT-induced multilevel magnetization switching as well as the typical synaptic functionality of spike-timing dependent plasticity (STDP) have been experimentally demonstrated ( Cao et al., 2019 ).…”

“…Connecting the classical digital computing based on deterministic bits of 0 or 1 and the quantum computing based on quantum bits (q-bits) with coherent superpositions of 0 and 1, there is an intermediate approach of probabilistic computing that relies on sampling probabilistic bits (p-bits) with spontaneously fluctuated states of either 0 or 1 ( Chowdhury et al., 2019 ). In the current new era of noisy intermediate-scale quantum (NISQ) technology, a period that the noise in quantum gates limits the size of quantum circuits ( Preskill, 2018 ), which is far from the eventually accurate, fully fault-tolerant quantum technologies in the future, the probabilistic computing with interconnected p-bits can be practically useful in a host of quantum computing applications, such as solving satisfaction, optimization (e.g., the traveling salesman problem, the integer factorization, and the invertible logics), and sampling problems ( Sutton et al., 2017 ; Grollier et al., 2020 ). The hardware construction of p-bit can be exactly realized by utilizing the low-barrier nanomagnet (LBNM), a single domain magnet or superparamagnetic particle whose magnetization reverses randomly when the barrier between opposite magnetic states is comparable with the thermal energy ( H K M s V/2 ≈ k B T , where the five quantities are the anisotropy field, the saturation magnetization, the volume, the Boltzmann constant, and the absolute temperature, respectively) ( Camsari et al., 2017 ).…”