Robust Deep Reservoir Computing Through Reliable Memristor With Improved Heat Dissipation Capability

An, Hongyu; Al‐Mamun, Mohammad; Orłowski, M.; Liu, Lingjia; Yi, Yang

doi:10.1109/tcad.2020.3002539

Cited by 15 publications

(10 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To improve the performance of DetectX under the impact of hardware non-idealities like device conductance variations, process-voltage-temparature (PVT) variations among others, variation aware training methods [45] can be employed during the dual-phase training. Additionally, new 3D crossbar architectures [46] and memristive devices [47] can be utilized to minimize the bit-line parasitic capacitance/resistances and device conductance variations, respectively which can ultimately improve the performance of DetectX.…”

Section: Hardware Evaluation Of the Crossbar+detectx Systemmentioning

confidence: 99%

DetectX -- Adversarial Input Detection using Current Signatures in Memristive XBar Arrays

Moitra,

Panda

2021

Preprint

View full text Add to dashboard Cite

Adversarial input detection has emerged as a prominent technique to harden Deep Neural Networks(DNNs) against adversarial attacks. Most prior works use neural network-based detectors or complex statistical analysis for adversarial detection. These approaches are computationally intensive and vulnerable to adversarial attacks. To this end, we propose DetectX -a hardware friendly adversarial detection mechanism using hardware signatures like Sum of column Currents (SoI) in memristive crossbars (XBar). We show that adversarial inputs have higher SoI compared to clean inputs. However, the difference is too small for reliable adversarial detection. Hence, we propose a dual-phase training methodology: Phase1 training is geared towards increasing the separation between clean and adversarial SoIs; Phase2 training improves the overall robustness against different strengths of adversarial attacks. For hardware-based adversarial detection, we implement the DetectX module using 32nm CMOS circuits and integrate it with a Neurosim-like analog crossbar architecture. We perform hardware evaluation of the Neurosim+DetectX system on the Neurosim platform using datasets-CIFAR10(VGG8), CIFAR100(VGG16) and Tiny-Imagenet(ResNet18). Our experiments show that DetectX is 10x-25x more energy efficient and immune to dynamic adversarial attacks compared to previous state-of-the-art works. Moreover, we achieve high detection performance (ROC-AUC>0.95) for strong white-box and black-box attacks. The code has been released at https://github.com/Intelligent-Computing-Lab-Yale/DetectX

show abstract

Section: Hardware Evaluation Of the Crossbar+detectx Systemmentioning

confidence: 99%

DetectX -- Adversarial Input Detection using Current Signatures in Memristive XBar Arrays

Moitra,

Panda

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In recent years, the Ferroelectric Field-Effect Transistor (FeFET) [12] and the Resistive Random-Access Memory (ReRAM) [13] have been introduced in neuromorphic applications as two emerging building blocks for emulating artificial neural networks. This makes these emerging devices possible candidates for implementation of IRS, due to their intrinsic reconfigurability.…”

Section: B Hardware Nonlinearity In Irsmentioning

confidence: 99%

Making Intelligent Reflecting Surfaces More Intelligent: A Roadmap Through Reservoir Computing

Zhou¹,

Bai²,

Mohammadi³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

This article introduces a neural networkbased signal processing framework for intelligent reflecting surface (IRS) aided wireless communications systems. By modeling radio-frequency (RF) impairments inside the "meta-atoms" of IRS (including nonlinearity and memory effects), we present an approach that generalizes the entire IRS-aided system as a reservoir computing (RC) system, an efficient recurrent neural network (RNN) operating in a state near the "edge of chaos". This framework enables us to take advantage of the nonlinearity of this "fabricated" wireless environment to overcome link degradation due to model mismatch. Accordingly, the randomness of the wireless channel and RF imperfections are naturally embedded into the RC framework, enabling the internal RC dynamics lying on the edge of chaos. Furthermore, several practical issues, such as channel state information acquisition, passive beamforming design, and physical layer reference signal design, are discussed.

show abstract

“…Recently metallic nanofilaments have attracted a great deal of attention due to its interesting physics and their great potentials for applications like neuromorphic computing, resistive non-volatlie memory, flexible touch screen, transparent electrodes and solar cells and transparent electrodes 1 – 7 . For the design and optimization of these devices, the electrical and thermal properties of an individual nanofilament are both fundamental and critical.…”

Section: Introductionmentioning

confidence: 99%

“…In recent studies, conductance quantization phenomena have been demonstrated to be present in the nanofilaments in ReRAM structures 51 – 54 . Hence, it is important to understand the interplay of electrical and thermal effects in a nanofilaments in order to fine-tune the performance, assess the reliability, and control the variability 55 , 56 of ReRAM memory arrays or of neuromorphic devices used in deep machine learning 1 .…”

Section: Introductionmentioning

confidence: 99%

“…As the vibrations of the Cu atoms subside during the cooling off period, the electron tunneling between the Cu atoms becomes more likely. At elevated temperatures, the average tunneling distance increases, leading to a sharp decrease of the tunneling probability and, consequently, to a sharp increase in transient resistance.Recently metallic nanofilaments have attracted a great deal of attention due to its interesting physics and their great potentials for applications like neuromorphic computing, resistive non-volatlie memory, flexible touch screen, transparent electrodes and solar cells and transparent electrodes [1][2][3][4][5][6][7] . For the design and optimization of these devices, the electrical and thermal properties of an individual nanofilament are both fundamental and critical.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Electron tunneling between vibrating atoms in a copper nano-filament

Al‐Mamun

Orłowski

2021

Sci Rep

View full text Add to dashboard Cite

Nanowires, atomic point contacts, and chains of atoms are one-dimensional nanostructures, which display size-dependent quantum effects in electrical and thermal conductivity. In this work a Cu nanofilament of a defined resistance and formed between a Cu and Pt electrode is heated remotely in a controlled way. Depending on the robustness of the conductive filament and the amount of heat transferred several resistance-changing effects are observed. In case of sufficiently fragile nanofilament exhibiting electrical quantum conductance effects and moderate heating applied to it, a dramatic increase of resistance is observed just after the completion of the heating cycle. However, when the filament is allowed to cool off, a spontaneous restoration of the originally set resistance of the filament is observed within less than couple tens of seconds. When the filament is sufficiently fragile or the heating too excessive, the filament is permanently ruptured, resulting in a high resistance of the cell. In contrast, for robust, low resistance filaments, the remote heating does not affect the resistance. The spontaneous restoration of the initial resistance value is explained by electron tunneling between neighboring vibrating Cu atoms. As the vibrations of the Cu atoms subside during the cooling off period, the electron tunneling between the Cu atoms becomes more likely. At elevated temperatures, the average tunneling distance increases, leading to a sharp decrease of the tunneling probability and, consequently, to a sharp increase in transient resistance.

show abstract

Robust Deep Reservoir Computing Through Reliable Memristor With Improved Heat Dissipation Capability

Cited by 15 publications

References 30 publications

DetectX -- Adversarial Input Detection using Current Signatures in Memristive XBar Arrays

DetectX -- Adversarial Input Detection using Current Signatures in Memristive XBar Arrays

Making Intelligent Reflecting Surfaces More Intelligent: A Roadmap Through Reservoir Computing

Electron tunneling between vibrating atoms in a copper nano-filament

Contact Info

Product

Resources

About