Improving the Recognition Accuracy of Memristive Neural Networks via Homogenized Analog Type Conductance Quantization.

Qilai Chen,Tingting Han,Gang Liu,Minghua Tang,Xuejun Zheng,Zhang Zhang

doi:10.3390/mi11040427

Abstract

Conductance quantization (QC) phenomena occurring in metal oxide based memristors demonstrate great potential for high-density data storage through multilevel switching, and analog synaptic weight update for effective training of the artificial neural networks. Continuous, linear and symmetrical modulation of the device conductance is a critical issue in QC behavior of memristors. In this contribution, we employ the scanning probe microscope (SPM) assisted electrode engineering strategy to control the ion migration process to construct single conductive filaments in Pt/HfOx/Pt devices. Upon deliberate tuning and evolution of the filament, 32 half integer quantized conductance states in the 16 G0 to 0.5 G0 range with enhanced distribution uniformity was achieved. Simulation results revealed that the numbers of the available QC states and fluctuation of the conductance at each state play an important role in determining the overall performance of the neural networks. The 32-state QC behavior of the hafnium oxide device shows improved recognition accuracy approaching 90% for handwritten digits, based on analog type operation of the multilayer perception (MLP) neural network.

Highlights

Resembling the operating principle of human brains that transmit and process information through huge amounts of interconnected neurons and synapses [1–4], neuromorphic computing paradigm based on new solid-state electronic devices have demonstrated advantages of high efficiency, low power consumption, and parallel processing ability when handling big-data analysis tasks [5–7]
Diguraritniognoopfeorxaytigoenn, athneiobnost/tvoamcanPct ielsectotrofodremwiansdaelnwtaatyiosngsro(Fuingduered,1a).ndDaulrlinthgeooppeerraattiioonn,sthweebreotctonmduPcttedlecintroadmebwieanst aelnwvairyosngmroeunnt.dAefdt,earwndaradll, Pthtetooppeerlaetcitornosdewsewreecreonddeupcotseidteidnianmtobtihenetceirncvuilraornpmaettnetr.nAtfhterorwugahrdE, -Pbtetaomp elveacptroordaetisown.ere deposited into the circular pattern through E-beam evaporation
We demonstrate a reliable hafnium oxide based memristor device that displays homogenized conductance quantization characteristics with 32 half-integer QC states

Summary

Introduction

Resembling the operating principle of human brains that transmit and process information through huge amounts of interconnected neurons and synapses [1–4], neuromorphic computing paradigm based on new solid-state electronic devices have demonstrated advantages of high efficiency, low power consumption, and parallel processing ability when handling big-data analysis tasks [5–7]. The ion migration and filamentary conduction mechanism make the memristor devices extremely scalable, enabling them to approach the lithographic limitations [18–21]. It was demonstrated that the computing and energy efficiency of memristor-based. 22 ooff 1100 memristor-based in-memory computation was comparable to that of the complementary metalionx-imdee-mseomryiccoonmdupcuttoarti(oCnMwOaSs)cpolmatpfoarrmabsle[2t2o,2t3h]a.t of the complementary metal-oxide-semiconductor (CMOShS)ripnlkaitnfogrmthse[d22im,2e3n].sion of the conductive filaments (CFs) into the atomic scale of quantum pointShcroinnktainctg tahlelodwims emnseimonriosftothr ebcaolnlidstuicctiveleefictlraomnentrtsan(CspFos)rtinwtoitthheouattomsciacttsecrailnegofaqnudanqtuumanptioziendt ccoonntdauccttaalnlocwe s(QmCe)mcrhiasrtoacrtbearilslitsictiscienleacntraolongtrdaonmspaoinrtsw[2it4h–o2u6]t.sIctanttoetrionnglyansdigqnuifaicnatniztelyd icnocnrdeauscetsanthcee (dQaCta) scthoarraagcetecraisptiaccsitiyn oafnatlhoeg ddeovmicaeisn,sb[u2t4–a2l6so]. Mbyoremiumltpi-olratyaenrtlyp,eerncehpatniocend, eremcopglonyitiniogn HacfcOuxramcyemaprpisrtooarchwiintgh 9h0o%mcoagnenbiezaecdh, iaenvaeldogbytympuelctio-nladyuecrtpaenrcceepqtuioann,tiezmatpiolonyfionrgbHoftOhxhmanedmwrirsittoternwditighiht opmatotegrennsi.zed, analog type conductance quantization for both handwritten digit patterns

Methods

Results

Conclusion