Hardware-Aware 
<i>In Situ</i>
 Learning Based on Stochastic Magnetic Tunnel Junctions

Kaiser, Jan; Borders, William A.; Fukami, Shunsuke; Ohno, Hideo; Datta, Supriyo

doi:10.1103/physrevapplied.17.014016

Cited by 61 publications

(33 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Importantly, the distribution of conductance noise is dictated by the magnetic energy landscape, which can be manipulated using a variety of methods including magnetic field (Hayakawa et al, 2021), spin transfer torque (Borders et al, 2019), spin orbit torque (Ostwal and Appenzeller, 2019), and voltage-controlled magnetic anisotropy (VCMA) (Cai et al, 2019;Safranski et al, 2021). As a result, the tunable random bitstream readout of stochastic MTJs can be used to implement Boltzmann machines for probabilistic computing (Kaiser et al, 2022). While proposals for spin-based BNNs have been made (Yang et al, 2020;Lu et al, 2022), they relied upon either streaming generated RNGs from the periphery into each array or using digital circuitry to fully compose the weight used in the sampling step.…”

Section: Introductionmentioning

confidence: 99%

Bayesian neural networks using magnetic tunnel junction-based probabilistic in-memory computing

Liu

Xiao

Kwon

et al. 2022

Front. Nanotechnol.

View full text Add to dashboard Cite

Bayesian neural networks (BNNs) combine the generalizability of deep neural networks (DNNs) with a rigorous quantification of predictive uncertainty, which mitigates overfitting and makes them valuable for high-reliability or safety-critical applications. However, the probabilistic nature of BNNs makes them more computationally intensive on digital hardware and so far, less directly amenable to acceleration by analog in-memory computing as compared to DNNs. This work exploits a novel spintronic bit cell that efficiently and compactly implements Gaussian-distributed BNN values. Specifically, the bit cell combines a tunable stochastic magnetic tunnel junction (MTJ) encoding the trained standard deviation and a multi-bit domain-wall MTJ device independently encoding the trained mean. The two devices can be integrated within the same array, enabling highly efficient, fully analog, probabilistic matrix-vector multiplications. We use micromagnetics simulations as the basis of a system-level model of the spintronic BNN accelerator, demonstrating that our design yields accurate, well-calibrated uncertainty estimates for both classification and regression problems and matches software BNN performance. This result paves the way to spintronic in-memory computing systems implementing trusted neural networks at a modest energy budget.

show abstract

Section: Introductionmentioning

confidence: 99%

Bayesian neural networks using magnetic tunnel junction-based probabilistic in-memory computing

Liu

Xiao

Kwon

et al. 2022

Front. Nanotechnol.

View full text Add to dashboard Cite

show abstract

“…RBMs have non-deterministic activation functions, and usually have a relatively small number of parameters, which fits well with the small and noisy RSM crossbars that are currently available. In the same way as BNNs, the probabilistic aspect of RBMs is not very desirable for CMOS chips, whereas RSMs offer new design perspectives (Kaiser et al, 2022). For example, suggested using the HRS and LRS variability to build a stochastic activation function an RBM and Mahmoodi et al (2019) experimentally demonstrated the benefits of thermal noise to realize a stochastic dot product computation.…”

Section: Rsm-based Methodsmentioning

confidence: 99%

Exploiting Non-idealities of Resistive Switching Memories for Efficient Machine Learning

et al. 2022

View full text Add to dashboard Cite

Novel computing architectures based on resistive switching memories (also known as memristors or RRAMs) have been shown to be promising approaches for tackling the energy inefficiency of deep learning and spiking neural networks. However, resistive switch technology is immature and suffers from numerous imperfections, which are often considered limitations on implementations of artificial neural networks. Nevertheless, a reasonable amount of variability can be harnessed to implement efficient probabilistic or approximate computing. This approach turns out to improve robustness, decrease overfitting and reduce energy consumption for specific applications, such as Bayesian and spiking neural networks. Thus, certain non-idealities could become opportunities if we adapt machine learning methods to the intrinsic characteristics of resistive switching memories. In this short review, we introduce some key considerations for circuit design and the most common non-idealities. We illustrate the possible benefits of stochasticity and compression with examples of well-established software methods. We then present an overview of recent neural network implementations that exploit the imperfections of resistive switching memory, and discuss the potential and limitations of these approaches.

show abstract

“…This can be used in neuromorphic circuits to mimic the neuron's leaky integration, or to detect the pulse order and delay that determine the synaptic weights [140,143]. The smooth analog switching and relaxation functions are distinct from ferromagnetic neuromorphic devices whose characteristic behavior are stochastic fluctuations between two states with opposite magnetization [144,145].…”

Section: B Ultra-fast Optics and Neuromorphicsmentioning

confidence: 99%

Emerging research landscape of altermagnetism

Šmejkal¹,

Sinova²,

Jungwirth³

2022

Preprint

View full text Add to dashboard Cite

Magnetism is one of the largest, most fundamental, and technologically most relevant fields of condensed-matter physics. Traditionally, two basic magnetic phases have been considered -ferromagnetism and antiferromagnetism. The breaking of the time-reversal symmetry and spin splitting of the electronic states by the magnetization in ferromagnets underpins a range of macroscopic responses in this extensively explored and exploited type of magnets. By comparison, antiferromagnets have vanishing net magnetization. This Perspective reflects on recent observations of materials hosting an intriguing ferromagnetic-antiferromagnetic dichotomy, in which spin-split spectra and macroscopic observables, akin to ferromagnets, are accompanied by antiparallel magnetic order with vanishing magnetization, typical of antiferromagnets. An unconventional non-relativistic symmetry-group formalism offers a resolution of this apparent contradiction by delimiting a third basic magnetic phase, dubbed altermagnetism. Our Perspective starts with an overview of the still emerging unique phenomenology of the phase, and of the wide array of altermagnetic material candidates. In the main part of the article, we illustrate how altermagnetism can enrich our understanding of overarching condensed-matter physics concepts, and have impact on prominent condensed-matter research areas. CONTENTS I. Introduction 1 II. Altermagnetic phase 3 A. Ab initio band-structures 3 B. Symmetry classification and description 4 C. Identification rules 6 D. Material candidates 7 III. Physical concepts 9 A. Kramers theorem 9 B. Fermi-liquid instabilities 10 C. Electron and magnon quasiparticles 12 D. Berry phase and non-dissipative transport 14 IV. Research areas 15 A. Spintronics 15 B. Ultra-fast optics and neuromorphics 17 C. Thermoelectrics, field-effect electronics and multiferroics 19 D. Superconductivity 20 V. Conclusion 20 Acknowledgement 20 References 21

show abstract

Hardware-Aware In Situ Learning Based on Stochastic Magnetic Tunnel Junctions

Cited by 61 publications

References 47 publications

Bayesian neural networks using magnetic tunnel junction-based probabilistic in-memory computing

Bayesian neural networks using magnetic tunnel junction-based probabilistic in-memory computing

Exploiting Non-idealities of Resistive Switching Memories for Efficient Machine Learning

Emerging research landscape of altermagnetism

Contact Info

Product

Resources

About