Abstract

Dealing with big data, especially the videos and images, is the biggest challenge of existing Von-Neumann machines while the human brain, benefiting from its massive parallel structure, is capable of processing the images and videos in a fraction of second. The most promising solution, which has been recently researched widely, is brain-inspired computers, so-called neuromorphic computing systems (NCS). The NCS overcomes the limitation of the word-at-a-time thinking of conventional computers benefiting from massive parallelism for data processing, similar to the brain. Recently, spintronic-based NCSs have shown the potential of implementation of low-power high-density NCSs, where neurons are implemented using magnetic tunnel junctions (MTJs) or spin torque nano-oscillators (STNOs) and memristors are used to mimic synaptic functionality. Although using STNOs as neuron requires lower energy in comparison to the MTJs, still there is a huge gap between the power consumption of spintronic-based NCSs and the brain due to high bias current needed for starting the oscillation with a detectable output power. In this manuscript, we propose a spintronic-based NCS (196 × 10) proof-of-concept where the power consumption of the NCS is reduced by assisting the STNO oscillation through a microwatt nanosecond laser pulse. The experimental results show the power consumption of the STNOs in the designed NCS is reduced by 55.3% by heating up the STNOs to 100°C. Moreover, the average power consumption of spintronic layer (STNOs and memristor array) is decreased by 54.9% at 100°C compared with room temperature. The total power consumption of the proposed laser assisted STNO-based NCS (LAO-NCS) at 100°C is improved by 40% in comparison to a typical STNO-based NCS at room temperature. Finally, the energy consumption of the LAO-NCA at 100°C is expected to reduce by 86% compared with a typical STNO-based NCS at the room temperature.

Highlights

  • The grand challenge of exascale computing, 1018 operations/second, calls for a dramatic change in hardware of the current petascale supercomputers

  • For the first time to our knowledge, we propose to design a laser-assisted spin torque nano-oscillators (STNOs)-based neuromorphic computing systems (NCS) (LAO-NCS) to improve power consumption of the state-of-the-art NCSs by at least 40%; narrowing the gap of power efficiency between the Brain and the current NCSs

  • Considering the fact that the power consumption improvement in the spintronic layer is significantly higher than the power consumption of the laser and the CMOS interfacing circuit, the total power consumption of the laser assisted STNO-based NCS (LAO-NCS) decreases by 40% at 100◦C compared with the room temperature

Read more

Summary

INTRODUCTION

The grand challenge of exascale computing, 1018 operations/second, calls for a dramatic change in hardware of the current petascale supercomputers. The weighted input currents are sufficiently large, the FL magnetic moment of the STNO starts to oscillate that will be detected by a sensing circuit immediately, and translated to neuron firing. Increasing the temperature of the STNO will decrease its energy barrier, which leads to a lower bias current needed for starting the oscillation in the STNO. The sensing circuit is to sense the magnetization oscillation of the STNOs (neurons) in order to find the fired neuron(s) and activate the corresponding input(s) in the post-synaptic neuronal layer This can be done either by sensing the frequency or the output power of the oscillating signal across the STNOs, and comparing it with a threshold frequency or a threshold output power. The output voltage of the comparator switches from “0” to “1” and it will be considered as neuron firing

RESULTS
CONCLUSION
DATA AVAILABILITY STATEMENT
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call