The article discusses the potential of spintronics, specifically magnetic tunnel junctions (MTJs), for neuromorphic computing. MTJs are promising because they can be used as controllable binary “weights” in neural networks, which are widely used in various information technologies.
The key features of MTJs that make them suitable for neuromorphic computing are their ability to be read and controlled using electronic circuits, and their compatibility with standard integrated circuits. These properties enable the implementation of fast, dense, and non-volatile memory in commercial applications.
The article also explores the use of MTJs in crossbar arrays, which can address energy efficiency issues in neural networks. Western Digital has collaborated with NIST to implement such arrays and solve neural network tasks.
The article also discusses the concept of superparamagnetism in MTJs, where the energy barrier separating the parallel and antiparallel states becomes comparable to ambient thermal energy. This property can be used to design computers that operate close to the thermal limit, potentially reducing power consumption.
Furthermore, the article explores the use of MTJs in stochastic computing, which represents numbers using random bitstreams generated with a probability corresponding to the number they represent. This approach can be implemented using low-energy bitstream generators based on superparamagnetic tunnel junctions.
The article also discusses the potential of spin-torque oscillators as non-Boolean or neuromorphic computational devices. These oscillators can be used for tasks such as calculating the degree of match between test and reference values, image reconstruction, and recurrent network processing.
In summary, the article highlights the potential of spintronics, particularly magnetic tunnel junctions, for neuromorphic computing. The use of MTJs in various computing schemes, such as neural networks, stochastic computing, and oscillators, can lead to more energy-efficient and efficient computers for tasks like voice and video recognition.
Source: https://www.nist.gov/programs-projects/spintronics-neuromorphic-computing
Keywords: Superparamagnetic, Tunnel junction, Stochastic computing, Spin torque, Oscillators
