SOT-MRAM Based Analog in-Memory Computing for DNN Inference

Doevenspeck, Jonas; Garello, Kévin; Verhoef, Bram-Ernst; Degraeve, R.; Beek, Simon Van; Crotti, D.; Yasin, F.; Couet, S.; Jayakumar, Ganesh; Papistas, Ioannis A.; Debacker, Peter; Lauwereins, Rudy; Dehaene, Wim; Kar, Gouri Sankar; Cosemans, Stefan; Mallik, Arindam; Verkest, D.

doi:10.1109/vlsitechnology18217.2020.9265099

Cited by 51 publications

(42 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…PCM and ReRAM typically have ON-resistance of 10–100 kΩ, while FeFET can operate at ON-resistance of higher than 500 kΩ (Jerry et al, 2017 ). Emerging spintronic devices such as spin orbit torque (SOT)-MTJ can potentially work under MΩ ON-resistance, due to separated read/write paths (Doevenspeck et al, 2020 ). The ability of providing large ON-state resistance can potentially enable deployment of large crossbars with minimal impact on computing accuracy.…”

Section: Discussionmentioning

confidence: 99%

“…Magnetic devices are further challenged by limited readout resolutions due to the relatively small magneto-resistance effect in practical MTJs. Fortunately, emerging SOT-MTJ with separate read and write paths could lead to device optimized in new design space with possibly higher ON/OFF ratios (up to 6x larger TMR ratio) that could be better suited for neuromorphic computing (Ikeda et al, 2008 ; Doevenspeck et al, 2020 ). And novel materials such as half metal have been predicted to potentially boost the magnetoresistive signals in future (Bhatti et al, 2017 ).…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Multi-Level Neuromorphic Devices Built on Emerging Ferroic Materials: A Review

Cheng

Agrawal

et al. 2021

Front. Neurosci.

View full text Add to dashboard Cite

Achieving multi-level devices is crucial to efficiently emulate key bio-plausible functionalities such as synaptic plasticity and neuronal activity, and has become an important aspect of neuromorphic hardware development. In this review article, we focus on various ferromagnetic (FM) and ferroelectric (FE) devices capable of representing multiple states, and discuss the usage of such multi-level devices for implementing neuromorphic functionalities. We will elaborate that the analog-like resistive states in ferromagnetic or ferroelectric thin films are due to the non-coherent multi-domain switching dynamics, which is fundamentally different from most memristive materials involving electroforming processes or significant ion motion. Both device fundamentals related to the mechanism of introducing multilevel states and exemplary implementations of neural functionalities built on various device structures are highlighted. In light of the non-destructive nature and the relatively simple physical process of multi-domain switching, we envision that ferroic-based multi-state devices provide an alternative pathway toward energy efficient implementation of neuro-inspired computing hardware with potential advantages of high endurance and controllability.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Multi-Level Neuromorphic Devices Built on Emerging Ferroic Materials: A Review

Cheng

Agrawal

et al. 2021

Front. Neurosci.

View full text Add to dashboard Cite

show abstract

“…It has been proposed that Boolean arithmetic computations can be done near or in the memory by modifying the peripheral circuits and hence help in the reduction of data movements [97]- [99]. Moreover, crossbar arrays of MRAM devices can execute analog MVM operations, where both in-memory processing of data and intrinsic hardware parallelism can be exploited [100]. Such a computing architecture provides an efficient hardware primitive for running DNNs, which heavily rely on MVM operations.…”

Section: Analog In-memory Computing With Mram Crossbar Arraysmentioning

confidence: 99%

“…SOT-MRAM, on the other hand, has no such restriction, since the write operation occurs through the underlying HM layer. This means, the tunneling barrier can be tuned for achieving higher R ON (> 1M) without affecting the writing operation [100]. One crucial point to consider while designing high R ON is the choice of V i used to encode the input vectors on the BL.…”

Section: B Design Considerations For Mram-based Crossbarsmentioning

confidence: 99%

“…Thus, the algorithm parameters can also play a significant role in determining the design parameters of crossbars for optimum accuracy and energy-efficiency. In [100], SOT-MTJs with R ON ≈ 5M were experimentally demonstrated. Their system-level simulations showed negligible accuracy drop on LeNet-5 and ResNet-26, for classifying MNIST and CIFAR-100 dataset, respectively, even under process variations and IR-drops.…”

Section: B Design Considerations For Mram-based Crossbarsmentioning

confidence: 99%

See 1 more Smart Citation

Magnetoresistive Circuits and Systems: Embedded Non-Volatile Memory to Crossbar Arrays

Agrawal

Wang

Sharma

et al. 2021

IEEE Trans. Circuits Syst. I

View full text Add to dashboard Cite

This overview article describes Magnetoresistive Random Access Memory (MRAM) from a circuits and systems perspective. We discuss various tradeoffs and design challenges of MRAM in three broad application areas: 1) embedded nonvolatile memory (eNVMs), 2) crossbar-based analog in-memory computing, and 3) stochastic computing. Certain MRAM characteristics, such as high retention, high endurance and fast read and write operations, make them ideal for replacing the standard CMOS memories for last-level cache applications with future scaling. However, various tradeoffs in power, performance and area pose conflicting requirements on MRAM design. We explore these challenges and various circuit techniques that have been developed to mitigate them. Further, we present various requirements of memristive crossbar arrays for accelerating matrixvector-multiplication (MVM) operations in light of MRAM devices, and highlight various challenges, design considerations, and applicability of MRAM as crossbar arrays. Finally, we will elaborate on how inherent stochasticity of MRAM devices can be leveraged for implementing energy-efficient true random number generators (TRNGs) and stochastic units for performing certain tasks, such as developing fast solvers for combinatorial optimization, and stochastic neural networks.

show abstract