Coupled Spin Torque Nano Oscillators for Low Power Neural Computation

Abstract

We present coupled Spin Torque Nano Oscillators (STNOs) as "electronic neurons" for efficient brain-inspired computation. The coupled STNOs show two distinct outputs, depending on whether the frequencies are locked or not. The locking mechanisms are based on magnetic coupling or injection locking. The neuron firing threshold can be set by tuning the locking range of the coupled STNOs. We employ a crossbar array of programmable memory devices like memristors to implement "electronic synapses" that work seamlessly with the coupled STNO's for hardware implementation of neural networks. Results show that injection locking based neuron model can be attractive from scaling point of view and computation like character recognition can be performed with energy consumption per neuron of ~1.8X and ~3X lower than the digital and the analog CMOS counterpart, respectively.Index Terms-coupled spin torque nano-oscillators, frequency locking, injection locking, low power, neural networks, programmable resistive crossbar.

Prospects of efficient neural computing with arrays of magneto-metallic neurons and synapses

“…Simulation of the proposed spintronic devices, calibrated to experimental results, demonstrates that spintronic neurons can provide ∼ 10 − 100× lower energy consumption in compari- son to a corresponding analog/digital CMOS implementation [2], [4]- [6], [9], [22]. Such magneto-metallic low resistance spintronic neurons can enable ultra-low power operation of synaptic crossbar arrays.…”

“…Such magneto-metallic low resistance spintronic neurons can enable ultra-low power operation of synaptic crossbar arrays. Further, STNOs have recently proved to be efficient at emerging non-Boolean applications like edge detection and associative computing, in addition to providing low-power neuron functionality [9]. Proposed spintronic synapses, not only provide low programming energies, but also offer decoupled "write" and "read" current paths [12].…”

Computing with dynamical systems based on insulator-metal-transition oscillators

“…More recently, this effort has been augmented by advances in the development of compact oscillators in non-silicon technologies. One prominent effort is the use of spin torque oscillators (STOs) coupled with using spin diffusion currents and providing a computational platform for machine learning, spiking neural networks, and others [23,24]. However, the high current densities of STOs and the limited range of spin diffusion currents continue to pose serious challenges to technologists.…”

Low-Power Microwave Relaxation Oscillators Based on Phase-Change Oxides for Neuromorphic Computing

“…Early work on the hardware development based on neuromorphic computing relied heavily on the matured CMOS based platform [7]. However, recently approaches based on spin-torque [8,9], metal-insulator-transition [10,11], resistive switching [12], and ferroelectric phase change materials [13,14] are being explored as alternatives to the traditional CMOS based ring oscillators.…”