M e m r is t o r s f o r N e u r o m o r p h ic C ir c u it s a n d A r t if ic ia l I Memristors for n t e llig e n Neuromorphic c e A p p lic a Circuits and t io n s • Jo Artificial Intelligence r d i S u ñ é Applications Edited by Jordi Suñé Printed Edition of the Special Issue Published in Materials www.mdpi.com/journal/materials Memristors for Neuromorphic Circuits and Artificial Intelligence Applications Memristors for Neuromorphic Circuits and Artificial Intelligence Applications SpecialIssueEditor JordiSun˜e´ MDPI•Basel•Beijing•Wuhan•Barcelona•Belgrade SpecialIssueEditor JordiSun˜e´ UniversitatAuto`nomade Barcelona,Departament d‘EnginyeriaElectro`nica Spain EditorialOffice MDPI St.Alban-Anlage66 4052Basel,Switzerland ThisisareprintofarticlesfromtheSpecialIssuepublishedonlineintheopenaccessjournalMaterials (ISSN 1996-1944) from 2018 to 2020 (available at: https://www.mdpi.com/journal/materials/ specialissues/memristors). Forcitationpurposes,citeeacharticleindependentlyasindicatedonthearticlepageonlineandas indicatedbelow: LastName,A.A.; LastName,B.B.; LastName,C.C.ArticleTitle. JournalNameYear,ArticleNumber, PageRange. ISBN978-3-03928-576-1(Pbk) ISBN978-3-03928-577-8(PDF) Cover image courtesy of Jaime Moroldo, Italian-Venezuelan plastic artist based in Spain, who is a long-lasting friend of Jordi Sun˜´e, editor of this book. (cid:2)c 2020 by the authors. Articles in this book are Open Access and distributed under the Creative Commons Attribution (CC BY) license, which allows users to download, copy and build upon published articles, as long as the author and publisher areproperly credited, which ensures maximum dissemination and a wider impact of our publications. The book as a whole is distributed by MDPI under the terms and conditions of the Creative Commons license CC BY-NC-ND. Contents AbouttheSpecialIssueEditor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii EnriqueMirandaandJordiSun˜e´ MemristorsforNeuromorphicCircuitsandArtificialIntelligenceApplications Reprintedfrom:Materials2020,13,938,doi:10.3390/ma13040938. . . . . . . . . . . . . . . . . . . 1 LuisA.Camun˜as-Mesa,Bernabe´Linares-BarrancoandTeresaSerrano-Gotarredona Neuromorphic Spiking Neural Networks and Their Memristor-CMOS Hardware Implementations Reprintedfrom:Materials2019,12,2745,doi:10.3390/ma12172745 . . . . . . . . . . . . . . . . . . 10 ValerioMilo,GerardoMalavena,ChristianMonzioCompagnoniandDanieleIelmini MemristiveandCMOSDevicesforNeuromorphicComputing Reprintedfrom:Materials2020,13,166,doi:10.3390/ma13010166. . . . . . . . . . . . . . . . . . . 38 AlejandroFerna´ndez-Rodr´ıguez,JordiAlcala`,JordiSun˜e,NarcisMestresandAnnaPalau Multi-Terminal Transistor-Like Devices Based on Strongly Correlated Metallic Oxides for NeuromorphicApplications Reprintedfrom:Materials2020,13,281,doi:10.3390/ma13020281. . . . . . . . . . . . . . . . . . . 71 RuiWang,TuoShi,XumengZhang,WeiWang,JinsongWei,JianLu,XiaolongZhao,Zuheng Wu,RongrongCao,ShibingLong,QiLiuandMingLiu BipolarAnalogMemristorsasArtificialSynapsesforNeuromorphicComputing Reprintedfrom:Materials2018,11,2102,doi:10.3390/ma11112102 . . . . . . . . . . . . . . . . . . 82 WookyungSun,SujinChoi,BokyungKimandJunheePark Three-Dimensional (3D) Vertical Resistive Random-Access Memory (VRRAM) Synapses for NeuralNetworkSystems Reprintedfrom:Materials2019,12,3451,doi:10.3390/ma12203451 . . . . . . . . . . . . . . . . . . 96 PaoloLaTorraca,FrancescoMariaPuglisi,AndreaPadovaniandLucaLarcher Multiscale Modeling for Application-Oriented Optimization of Resistive Random-AccessMemory Reprintedfrom:Materials2019,12,3461,doi:10.3390/ma12213461 . . . . . . . . . . . . . . . . . . 108 N. Rodriguez, D. Maldonado, F. J. Romero, F. J. Alonso, A. M. Aguilera, A. Godoy, F. Jimenez-Molinos,F.G.RuizandJ.B.Roldan Resistive Switching and Charge Transport in Laser-Fabricated Graphene Oxide Memristors: ATimeSeriesandQuantumPointContactModelingApproach Reprintedfrom:Materials2019,12,3734,doi:10.3390/ma12223734 . . . . . . . . . . . . . . . . . . 133 Da´nielHajto´,A´da´mRa´kandGyo¨rgyCserey RobustMemristorNetworksforNeuromorphicComputationApplications Reprintedfrom:Materials2019,12,3573,doi:10.3390/ma12213573 . . . . . . . . . . . . . . . . . . 142 SonNgocTruong A Parasitic Resistance-Adapted Programming Scheme for Memristor Crossbar-Based NeuromorphicComputingSystems Reprintedfrom:Materials2019,12,4097,doi:10.3390/ma12244097 . . . . . . . . . . . . . . . . . . 153 v Agustı´nCisternasFerri,AlanRapoport,GermanPatterson,PabloFierens,EnriqueMiranda andJordiSun˜e´ OntheApplicationofaDiffusiveMemristorCompactModeltoNeuromorphicCircuits Reprintedfrom:Materials2019,12,2260,doi:10.3390/ma12142260 . . . . . . . . . . . . . . . . . . 165 Marta Pedro´, Javier Martı´n-Mart´ınez, Marcos Maestro-Izquierdo, Rosana Rodr´ıguez and MontserratNafrı´a Self-Organizing Neural Networks Based on OxRAM Devices under a Fully Unsupervised TrainingScheme Reprintedfrom:Materials2019,12,3482,doi:10.3390/ma12213482 . . . . . . . . . . . . . . . . . . 183 TienVanNguyen,KhoaVanPhamandKyeong-SikMin Memristor-CMOS Hybrid Circuit for Temporal-Pooling of Sensory and Hippocampal ResponsesofCorticalNeurons Reprintedfrom:Materials2019,12,875,doi:10.3390/ma12060875. . . . . . . . . . . . . . . . . . . 201 TienVanNguyen,KhoaVanPhamandKyeong-SikMin Hybrid Circuit of Memristor and Complementary Metal-Oxide-Semiconductor for Defect-TolerantSpatialPoolingwithBoost-FactorAdjustment Reprintedfrom:Materials2019,12,2122,doi:10.3390/ma12132122 . . . . . . . . . . . . . . . . . . 215 vi About the Special Issue Editor Jordi Su˜n´eis a full professor of Electronics at the Universitat Aut`onomad e Barcelona (UAB). He is the coordinator of the NANOCOMP research group, dedicated to the modeling and simulation of electron devices with a multi-scale approach. His main contributions are in the area of gate oxide reliability for CMOS technology. In terms of research achievements in this field, he was upgraded to IEEE Fellow for contributions to the understanding of gate oxide failure and reliability methodology. In 2008, he received the IBM Faculty award for a long-lasting collaboration with IBM Microelectronics in this field. Since 2008, he has worked in the area of memristive devices and their application to neuromorphic circuits. In 2010, he received the ICREA ACADEMIA award and, in 2012 and 2013, he was awarded the Chinese Academy of Sciences Professorship for Senior International Scientists, for a collaboration with IMECAS (Beijing, China). Recently, he launched a new research group/network (neuromimeTICs.org) dedicated to the application of neuromorphic electronics to artificial intelligence and to dissemination activities. He has (co)authored more than 400 papers (h-index = 44) in international journals and relevant conferences, including 14 IEDM papers, several invited papers, and five tutorials on oxide reliability at the IEEE-IRPS. At present, he’s the local UAB coordinator of a European project (EU2020-ECSEL-WAKeMeUP) on emerging non-volatile memories embedded in microprocessors for automotive, secure, and general electronics applications. His present research interests are: transition metal oxide-based filamentary RRAM memristors; complex perovskite oxide based memristors; bio-realistic compact modeling of memristors for neuromorphic applications; RRAM fabrication, characterization and modelling; biomimetic electrical circuit simulation with SPICE; analog circuits based on the combination of CMOS and memristors; and, in general, artificial neural networks for artificial intelligence applications. Jordi Suñé was funded by the WAKeMeUP 783176 project, co-funded by grants from the Spanish Ministerio de Ciencia, Innovación y Universidades (PCI2018-093107 grant) and the ECSEL EU Joint Undertaking. vii Preface to ”Memristors for Neuromorphic Circuits and Artificial Intelligence Applications” Theapplicationsofartificialintelligence(AI)andtheirimpactsonglobalsocietyarecurrently growingatanexponentialpace.Imageandspeechrecognitionandprocessing,businessoptimization, medical diagnosis, autonomous cars, and science discovery are only some of these applications. Although the term AI was coined in the late 1950s, it is only in the past decade that due to the impressive improvements in computing power, AI has found applications in many areas, evenexceedinghumansinsometasks.Itisconvenienttodistinguishbetweenconventional(narrow) AIapplicationsdesignedforonespecifictask,andartificialgeneralintelligence(AGI),whichaims atemulatinghumansinthemostgeneralsituations. AllbigAIindustrialplayersarecommittedto achievingAGIwiththeideathatonceyousolveintelligence,youcanuseittosolveeverythingelse. AIisheraldedasarevolutionarytechnologyforthe21stcentury; ithasmanyapplicationsforthe goodbut,initsAGIversion,ithasalsobeensignaledasoneofthesignificantrisksforthefutureof humanity.Artificialneuralnetworks(ANNs)areinspiredbythestructureofthebrainandareformed by artificial neurons interconnected by artificial synapses exhibiting plasticity effects. During the trainingofanANN,largeamountsofdataareprovidedthroughtheinputneuronsandthestrength ofthesynapticinterconnectionsareprogressivelymodifieduntilthenetworklearnshowtoclassify notonlythetrainingdatabutalsounforeseendataofasimilarkind. MostAIalgorithmshavebeen implementedbysoftwareprogramsrunonconventionalcomputingsystemswithaVonNeumann architecture, suchascentralprocessingunits, graphicalprocessingunits, andfieldprogrammable gatearrays.Recently,speciallydesignedintegratedcircuits,suchasthetensorprocessingunit(TPU), havebeenintroducedtooptimizethetypeofoperations(vector-matrixmultiplication)requiredfor trainingandinference.Inthesecomputingsystems,thememoryandprocessingunitsarephysically separatedsothatsignificantamountsofdataneedtobeshuttledbackandforthduringcomputation. Thiscreatesaperformancebottleneckbothinprocessingspeedduetothememorylatencyandin powerconsumptionduetotheenergyrequirementsfordataretrieval,transportation,andstorage. Theproblemisaggravatedbydeeplearningsystemsgrowingsignificantlyincomplexityforbetter recognitionaccuracy; asaconsequence, trainingtime, cost, andenergyconsumptionsignificantly increase.Thistrendhasthedrawbackthattheover-the-airdistributionrequiredbyedgeapplications becomes moredifficult, if notimpossible. Giventhe requiredcomplexity ofcomputing resources andthehugeamountsofenergyconsumed,alternativeapproachestosoftware-basedAItoolsare urgently needed. In this regard, the hardware realization of AI applications and, in particular, theuseofmemristorstoimplementANNsmightbethenextstepinthewaytowardfast,compact, andenergyefficientAIsystemswithaperformancemuchclosertothatofthehumanbrain. In this book, various reputed authors cover different aspects of the implementation of these memristiveneuromorphicsystems. Thebookstartswithaneditorialandtwoinvitedcontributions that review the state-of-the-art ANNs implemented in hardware and present the basic concepts of neuromorphics. After this, different papers cover the whole field by focusing on advanced memristor devices for synaptic applications, including modelling for device improvement and circuit simulation; on some issues related to the organization of memristors in dense crossbar arrays, and finally on the application of these circuits to AI problems. The device-related papers cover promising three-terminal structures based on complex perovskite oxides, devices based on binaryoxidesandnewtrainingprotocolstoachieveimprovedsynapticproperties,andnewvertical ix