Spin–Orbit Torque‐Induced Domain Nucleation for Neuromorphic Computing

Abstract

Neuromorphic computing has become an increasingly popular approach for artificial intelligence because it can perform cognitive tasks more efficiently than conventional computers. However, it remains challenging to develop dedicated hardware for artificial neural networks. Here, a simple bilayer spintronic device for hardware implementation of neuromorphic computing is demonstrated. In L11‐CuPt/CoPt bilayer, current‐inducted field‐free magnetization switching by symmetry‐dependent spin–orbit torques shows a unique domain nucleation‐dominated magnetization reversal, which is not accessible in conventional bilayers. Gradual domain nucleation creates multiple intermediate magnetization states which form the basis of a sigmoidal neuron. Using the L11‐CuPt/CoPt bilayer as a sigmoidal neuron, the training of a deep learning network to recognize written digits, with a high recognition rate (87.5%) comparable to simulation (87.8%) is further demonstrated. This work offers a new scheme of implementing artificial neural networks by magnetic domain nucleation.

Bio‐Inspired In‐Sensor Compression and Computing Based on Phototransistors

“…In addition, traditional neural networks still require the entire signal to complete the recognition tasks, hence the speed and energy efficiency of devices that use these sensors are dominated not by the imagecapturing-compression process, but by the reconstructed data transmission-recognition process. [22][23][24][25][26][27][28][29] To solve such problem, the in-sensor computing, where sensors can collaborate to perform the information processing, data aggregation and compression, is a possible way for achieving the efficient hardware to reduce data redundancy and movement. [1,[30][31][32][33][34] Furthermore, the reservoir computing (RC) can not only directly respond to stimuli without extra processors to implement the in-sensor computing, but also nonlinearly extract the high dimensional features from temporal data, thus reducing the network size and redundant data.…”

Field-Free Magnetization Switching Induced by Bulk Spin–Orbit Torque in a (111)-Oriented CoPt Single Layer with In-Plane Remanent Magnetization

“…For the other field-free strategies, a complex fabrication process, an asymmetric structure, and additional functional layers are essential, which would destroy the simple single-layered structure and is complex for application. 30,32 In this work, field-free perpendicular magnetization switching in a CoPt single layer induced by the bulk SOT effect is implemented at room temperature. A (111)-oriented CoPt single layer with tilted perpendicular magnetic anisotropy (PMA) is directly deposited on the Si/SiO 2 substrates, followed by vacuum annealing treatment.…”

“…However, the additional layers impair the inherent advantages of a single layer structure, for example, process compatibility and fabrication costs. For the other field-free strategies, a complex fabrication process, an asymmetric structure, and additional functional layers are essential, which would destroy the simple single-layered structure and is complex for application. , …”

Emerging Low‐Dimensional Heterostructure Devices for Neuromorphic Computing

“…Spintronic devices based on the low‐dimensional heterostructures [ 19,41,42 ] are also promising for neuromorphic computing. [ 22,43–48 ] Magnetic tunnel junction based on ferromagnetic metal/oxide/ferromagnetic metal heterostructure has been regarded as one of promising candidates for nonvolatile memory and computing applications. [ 41,49–51 ] Magnetization dynamics in the spin–torque oscillators based on magnetic tunnel junction gives rise to nonlinear I–V characteristics and finite magnetization relaxation time and have been employed to implement reservoir computing.…”

“…[ 50 ] Spin–orbit torque devices have separate writing and reading paths and can be exploited for emulating the biological synapse and neuron. [ 22,43,44,48 ] However, an external field is usually required to achieve magnetization switching in the spin–orbit torque neuromorphic devices, which not only increases power consumption but also limits device scaling. [ 50 ] The exchange bias field generated at the interface of antiferromagnet/ferromagnet heterostructures allows for eliminating the external magnetic field [ 43,55 ] and thus enables realization of field‐free artificial neurons and synapse with same‐architecture device.…”