Energy-Efficient All-Spin BNN Using Voltage-Controlled Spin-Orbit Torque Device for Digit Recognition

Shreya, Sonal; Verma, Gaurav; Piramanayagam, S. N.; Kaushik, Brajesh Kumar

doi:10.1109/ted.2020.3038140

Cited by 26 publications

(5 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In conclusion, using MTJs as neurons and synapses in BNN is feasible. Shreya et al implemented an all-spin BNN, reducing power consumption by 82% compared to a SRAM-based BNN [118].…”

Section: Mtjs For Bnnsmentioning

confidence: 99%

Tunneling magnetoresistance materials and devices for neuromorphic computing

et al. 2023

View full text Add to dashboard Cite

Artificial intelligence has become indispensable in modern life, but its energy consumption has become a significant concern due to its huge storage and computational demands. Artificial intelligence algorithms are mainly based on deep learning algorithms, relying on the backpropagation of convolutional neural networks or binary neural networks. While these algorithms aim to simulate the learning process of the human brain, their low bio-fidelity and the separation of storage and computing units lead to significant energy consumption. The human brain is a remarkable computing machine with extraordinary capabilities for recognizing and processing complex information while consuming very low power. Tunneling magnetoresistance (TMR)-based devices, namely magnetic tunnel junctions (MTJs), have great advantages in simulating the behavior of biological synapses and neurons. This is not only because MTJs can simulate biological behavior such as Spike-Timing Dependent Plasticity(STDP) and Leaky Integrate-Fire(LIF), but also because MTJs have intrinsic stochastic and oscillatory properties. These characteristics improve MTJs' bio-fidelity and reduce their power consumption. MTJs also possess advantages such as ultrafast dynamics and non-volatile properties, making them widely utilized in the field of neuromorphic computing in recent years. We conducted a comprehensive review of the development history and underlying principles of TMR, including a detailed introduction to the material and magnetic properties of MTJs and their temperature dependence. We also explored various writing methods of MTJs and their potential applications. Furthermore, we provided a thorough analysis of the characteristics and potential applications of different types of MTJs for neuromorphic computing. TMR-based devices have demonstrated promising potential for broad application in neuromorphic computing, particularly in the development of spiking neural networks. Their ability to perform on-chip learning with ultra-low power consumption makes them an exciting prospect for future advances in the era of the Internet of Things.

show abstract

“…In conclusion, using MTJs as neurons and synapses in BNN is feasible. Shreya et al implemented an all-spin BNN, reducing power consumption by 82% compared to a SRAM-based BNN [118].…”

Section: Mtjs For Bnnsmentioning

confidence: 99%

Tunneling magnetoresistance materials and devices for neuromorphic computing

et al. 2023

View full text Add to dashboard Cite

show abstract

“…These non-volatile and energy-efficient devices have the potential for unsupervised learning through dynamic conductance adjustments, guided by an innovative learning rule. Although there are a few works that proposed the use of MTJs for plasticity dynamics in SNN (Shreya et al, 2020 ; Jang et al, 2021 ; Leonard et al, 2022 ), there is a need to demonstrate this in a full network trained with a device-specific learning rule. The key contribution of this paper focuses on the synaptic design and a compatible learning rule .…”

Section: Introductionmentioning

confidence: 99%

Bi-sigmoid spike-timing dependent plasticity learning rule for magnetic tunnel junction-based SNN

Daddinounou,

Vatajelu

2024

Front. Neurosci.

View full text Add to dashboard Cite

In this study, we explore spintronic synapses composed of several Magnetic Tunnel Junctions (MTJs), leveraging their attractive characteristics such as endurance, nonvolatility, stochasticity, and energy efficiency for hardware implementation of unsupervised neuromorphic systems. Spiking Neural Networks (SNNs) running on dedicated hardware are suitable for edge computing and IoT devices where continuous online learning and energy efficiency are important characteristics. We focus in this work on synaptic plasticity by conducting comprehensive electrical simulations to optimize the MTJ-based synapse design and find the accurate neuronal pulses that are responsible for the Spike Timing Dependent Plasticity (STDP) behavior. Most proposals in the literature are based on hardware-independent algorithms that require the network to store the spiking history to be able to update the weights accordingly. In this work, we developed a new learning rule, the Bi-Sigmoid STDP (B2STDP), which originates from the physical properties of MTJs. This rule enables immediate synaptic plasticity based on neuronal activity, leveraging in-memory computing. Finally, the integration of this learning approach within an SNN framework leads to a 91.71% accuracy in unsupervised image classification, demonstrating the potential of MTJ-based synapses for effective online learning in hardware-implemented SNNs.

show abstract

“…Moreover, it is expected to solve reliability, area cost and energy efficacy bottlenecks and provide more possibilities for learning and computing processes of large-scale NC. Typically, the magnetoresistive random-access memory (MRAM) bears competitive advantages of high read and write speed, high reliability, ultra-low power consumption, and nearly infinite endurace 25 , 26 . Therefore, developing neuromorphic hardware based on the advanced MRAM components is paramount important from the practical application perspectives, in which the synthetic antiferromagnet (SAF) 27 – 29 poises as the kernel of the commercialized MRAM cell, i.e., magnetic tunnel junction (MTJ) as described in early reports and our previous works 3 , 30 – 38 .…”

Section: Introductionmentioning

confidence: 99%

Spintronic leaky-integrate-fire spiking neurons with self-reset and winner-takes-all for neuromorphic computing

et al. 2023

View full text Add to dashboard Cite

Neuromorphic computing using nonvolatile memories is expected to tackle the memory wall and energy efficiency bottleneck in the von Neumann system and to mitigate the stagnation of Moore’s law. However, an ideal artificial neuron possessing bio-inspired behaviors as exemplified by the requisite leaky-integrate-fire and self-reset (LIFT) functionalities within a single device is still lacking. Here, we report a new type of spiking neuron with LIFT characteristics by manipulating the magnetic domain wall motion in a synthetic antiferromagnetic (SAF) heterostructure. We validate the mechanism of Joule heating modulated competition between the Ruderman–Kittel–Kasuya–Yosida interaction and the built-in field in the SAF device, enabling it with a firing rate up to 17 MHz and energy consumption of 486 fJ/spike. A spiking neuron circuit is implemented with a latency of 170 ps and power consumption of 90.99 μW. Moreover, the winner-takes-all is executed with a current ratio >104 between activated and inhibited neurons. We further establish a two-layer spiking neural network based on the developed spintronic LIFT neurons. The architecture achieves 88.5% accuracy on the handwritten digit database benchmark. Our studies corroborate the circuit compatibility of the spintronic neurons and their great potential in the field of intelligent devices and neuromorphic computing.

show abstract

Energy-Efficient All-Spin BNN Using Voltage-Controlled Spin-Orbit Torque Device for Digit Recognition

Cited by 26 publications

References 30 publications

Tunneling magnetoresistance materials and devices for neuromorphic computing

Tunneling magnetoresistance materials and devices for neuromorphic computing

Bi-sigmoid spike-timing dependent plasticity learning rule for magnetic tunnel junction-based SNN

Spintronic leaky-integrate-fire spiking neurons with self-reset and winner-takes-all for neuromorphic computing

Contact Info

Product

Resources

About