Nanoscale neural network using non-linear spin-wave interference

Abstract

We demonstrate the design of a neural network hardware, where all neuromorphic computing functions, including signal routing and nonlinear activation are performed by spin-wave propagation and interference. Weights and interconnections of the network are realized by a magnetic-field pattern that is applied on the spin-wave propagating substrate and scatters the spin waves. The interference of the scattered waves creates a mapping between the wave sources and detectors. Training the neural network is equivalent to finding the field pattern that realizes the desired input-output mapping. A custom-built micromagnetic solver, based on the Pytorch machine learning framework, is used to inverse-design the scatterer. We show that the behavior of spin waves transitions from linear to nonlinear interference at high intensities and that its computational power greatly increases in the nonlinear regime. We envision small-scale, compact and low-power neural networks that perform their entire function in the spin-wave domain.

Giant nonlinear self-phase modulation of large-amplitude spin waves in microscopic YIG waveguides

“…On one hand, such phase uncertainty can severely influence the operation of spin-wave devices utilizing the phase of spin waves for information encoding 27 – 30 . On the other hand, the amplitude-dependent phase modulation can be used for implementation of devices with advanced functionality, e.g., systems for neuromorphic-like computing utilizing nonlinear interference of spin waves 32 .

Figure 6 Effects of self-phase modulation in the time domain.

…”

“…This effect also leads to temporal variation of the spin-wave phase by an amount significantly exceeding 2π over the pulse duration. These phenomena are important not only for understanding the propagation of strongly nonlinear microscopic-scale wave packets and solitons, but are also crucial for the operation of spin-wave devices relying on the phase of spin waves 28 – 32 and can be used to extend their functionality.…”

Inverse-design magnonic devices

“…It is noted, that our manuscript is submitted simultaneously with Ref. [32] in which the authors use the inverse-design magnonics approach to demonstrate the concept of a nanoscale neural network.…”

Exchange spin waves in thin films with gradient composition

“…Practical advantages of the magnonics include tunability of the magnon frequencies via the external field, material choice, or the geometry of the magnon media, charge-current-free nature and low power consumption of the spin-wave transfer, and micro-and sub-microscale dimensions of spin waves at microwave frequencies, which allows to target creation of micro-devices for processing of microwave information. Currently, the applied magnonics is progressed towards development of magnon logic devices [7], i.g., waveguides [9], magnon transistors [10], directional couplers [11], majority gates [12,13], non-Boolean devices [5], and neuromorphic circuits [14,15]. Also, magnonics finds its application in spintronic systems [4,[16][17][18][19].…”