Exploring the path of the variable resistance

In handling computer hardware, the last thing anyone would like to do is expose electronic components to electrostatic discharges. Nevertheless, this is exactly an approach that researchers are taking toward faster and more energy-efficient computing. Inspired by the functions of neurons and synapses in the brain, resistive switching devices or “memristors” are being explored as building blocks for neuromorphic circuitry. In such devices, the resistance properties are durably altered by applying voltage pulses. On page 907 of this issue, del Valle et al. ([ 1 ][1]) have imaged the early stages of electric field–induced electronic breakdown and formation of a conducting filament in vanadium oxide. By doing this in a space- and time-resolved manner, the authors provide useful insight into the characteristic length and time scales involved. ![Figure][2] Moving toward neural networksGRAPHIC: KELLIE HOLOSKI/ SCIENCE Computing systems are commonly based on the Von Neumann architecture, in which the memory is physically separated from the logic circuitry. Data are continuously shuttled between these units. This process is time consuming and presents an important cause of energy dissipation. Both aspects become very noticeable in data-intensive applications, like training deep neural networks. Neural networks are composed of layers of neuron-like devices connected through synapses. The latter comprise weight factors that are adjusted in the training process. In conventional complementary metal-oxide semiconductor (CMOS)–based technology, the weights need to be fetched, adjusted, and put back into the memory in every learning step. In an alternative and ultimately more efficient approach, the weights are embodied in the hardware itself, and training implies an alteration of the physical properties of the synapse, similar to what happens in the brain. In a fully electronic implementation, this requires the ability to controllably adjust the electrical resistance of a material. This is achieved using the electric field–driven motion of defect states, such as oxygen vacancies and impurity atoms ([ 2 ][3]), which are resistive switching concepts used also in binary resistive random access memory (ReRAM). Alternatives involve thermally induced alterations of the crystallinity of the material ([ 3 ][4]) and organic memristors ([ 4 ][5]). A complication in many techniques is that they involve atomic displacements and reconfigurations, which can lead to a spread in device properties and fatigue. This problem is circumvented by exploiting tunable electronic and/or magnetic ordering phenomena. The Mott insulator VO2 is an attractive example, exhibiting a hysteretic resistive transition just above room temperature ([ 5 ][6]). Applying electric field pulses to the material in the high-resistive state creates a metallic filament with a conductance that depends on the pulse intensity and duration. Notably, the resistance can be programmed over several orders of magnitude. By studying thin film microdevices with various vanadium oxide stoichiometries, del Valle et al. found that the transition starts with resistance fluctuations and nucleation of the conducting filament in hotspots on a hundreds-of-nanoseconds time scale (see the figure). In an avalanche-like process, the filament subsequently grows, as a result of Joule heating, over a time scale of microseconds. The authors investigated the growth dynamics and the final width of the conducting filament, which depends on both the characteristics of the voltage pulse and the resistivities of the material in the insulating and conducting states. Inhomogeneities play an important role in triggering the transition and in the filament formation by focusing the current. These findings can help to optimize the switching processes—e.g., by deliberately incorporating nanoscopic elements that act as optimized hotspots. The storing of synaptic weights in the neural network hardware is an example of the upcoming in-memory computing paradigm, which aims to circumvent the Von Neumann bottleneck. The practical implementation of this is typically in the form of cross-bar arrays ([ 6 ][7]), with the current lines acting as the pre- and postsynaptic connections to the neurons. The variable conductance properties of the barrier materials encode for the synaptic weight. Using this setup, Ohm's law and Kirchhoff's circuit law are used for matrix-vector multiplications, which are a key processing step in neural network operation. Also, other data-intensive applications can benefit from outsourcing data processing from the logic units to the memory—large-scale database queries being one example ([ 7 ][8]). In addition to storing information, the switching of VO2 when exceeding a certain threshold voltage can also be used for the realization of the artificial neurons. Using a negative differential resistance that can be invoked in the resistive transition, Yi et al. have even demonstrated 23 different neuronal functionalities with VO2-based memristors ([ 8 ][9]). Spiking modes of neural network operation are facilitated by this, with further expected enhancements in energy efficiency. The optical reflectivity modulation, as studied by del Valle et al. , presents a coupling between the electronic and photonic domains. This allows, for example, for the storing of synaptic weights in a photonic processor—a principle recently used in a photonic tensor core accelerator using phase change materials ([ 9 ][10]). Future computer systems will likely comprise a heterogeneous mix of electronic, optical, and spintronic components, and efficient coupling between these domains will then be indispensable. The next stage in vanadium oxide memristor research will be to make the step from single resistive switching devices to functional network structures, like multilayer artificial neural networks, and to explore their operation. In this endeavor, other more exotic post–Von Neumann information processing concepts are also of interest ([ 10 ][11], [ 11 ][12]). The space- and time-resolved optical reflectometry technique as demonstrated by del Valle et al. will enable current pulses and associated resistance modulations passing through such networks to be monitored without interference—tracing, so to say, the path of the variable resistance. 1. [↵][13]1. J. del Valle et al ., Science 373, 907 (2021). [OpenUrl][14][Abstract/FREE Full Text][15] 2. [↵][16]1. R. Waser, 2. R. Dittmann, 3. G. Staikov, 4. K. Szot , Adv. Mater. 21, 2632 (2009). [OpenUrl][17] 3. [↵][18]1. I. Boybat et al ., Nat. Commun. 9, 2514 (2018). [OpenUrl][19][CrossRef][20][PubMed][21] 4. [↵][22]1. S. Goswami, 2. S. Goswami, 3. T. Venkatesan , Appl. Phys. Rev. 7, 021303 (2020). [OpenUrl][23] 5. [↵][24]1. T. Driscoll, 2. H.-T. Kim, 3. B.-G. Chae, 4. M. Di Ventra, 5. D. N. Basov , Appl. Phys. Lett. 95, 043503 (2009). [OpenUrl][25][CrossRef][26] 6. [↵][27]1. Q. Xia, 2. J. J. Yang , Nat. Mater. 18, 309 (2019). [OpenUrl][28][CrossRef][29][PubMed][30] 7. [↵][31]1. I. Giannopoulos et al ., Adv. Intell. Syst. 2, 2000141 (2020). [OpenUrl][32] 8. [↵][33]1. W. Yi et al ., Nat. Commun. 9, 4661 (2018). [OpenUrl][34][CrossRef][35][PubMed][36] 9. [↵][37]1. J. Feldmann et al ., Nature 589, 52 (2021). [OpenUrl][38] 10. [↵][39]1. M. Di Ventra, 2. F. L. Traversa , J. Appl. Phys. 123, 180901 (2018). [OpenUrl][40] 11. [↵][41]1. M. A. Nugent, 2. T. W. Molter , PLOS ONE 9, e85175 (2014). [OpenUrl][42] [1]: #ref-1 [2]: pending:yes [3]: #ref-2 [4]: #ref-3 [5]: #ref-4 [6]: #ref-5 [7]: #ref-6 [8]: #ref-7 [9]: #ref-8 [10]: #ref-9 [11]: #ref-10 [12]: #ref-11 [13]: #xref-ref-1-1 "View reference 1 in text" [14]: {openurl}?query=rft.jtitle%253DScience%26rft.stitle%253DScience%26rft.aulast%253Ddel%2BValle%26rft.auinit1%253DJ.%26rft.volume%253D373%26rft.issue%253D6557%26rft.spage%253D907%26rft.epage%253D911%26rft.atitle%253DSpatiotemporal%2Bcharacterization%2Bof%2Bthe%2Bfield-induced%2Binsulator-to-metal%2Btransition%26rft_id%253Dinfo%253Adoi%252F10.1126%252Fscience.abd9088%26rft_id%253Dinfo%253Apmid%252F34301856%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [15]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEyOiIzNzMvNjU1Ny85MDciO3M6NDoiYXRvbSI7czoyMjoiL3NjaS8zNzMvNjU1Ny84NTQuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9 [16]: #xref-ref-2-1 "View reference 2 in text" [17]: {openurl}?query=rft.jtitle%253DAdv.%2BMater.%26rft.volume%253D21%26rft.spage%253D2632%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [18]: #xref-ref-3-1 "View reference 3 in text" [19]: {openurl}?query=rft.jtitle%253DNat.%2BCommun.%26rft.volume%253D9%26rft.spage%253D2514%26rft_id%253Dinfo%253Adoi%252F10.1038%252Fs41467-018-04933-y%26rft_id%253Dinfo%253Apmid%252F29955057%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [20]: /lookup/external-ref?access_num=10.1038/s41467-018-04933-y&link_type=DOI [21]: /lookup/external-ref?access_num=29955057&link_type=MED&atom=%2Fsci%2F373%2F6557%2F854.atom [22]: #xref-ref-4-1 "View reference 4 in text" [23]: {openurl}?query=rft.jtitle%253DAppl.%2BPhys.%2BRev.%26rft.volume%253D7%26rft.spage%253D021303%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [24]: #xref-ref-5-1 "View reference 5 in text" [25]: {openurl}?query=rft.jtitle%253DAppl.%2BPhys.%2BLett.%26rft.volume%253D95%26rft.spage%253D043503%26rft_id%253Dinfo%253Adoi%252F10.1063%252F1.3187531%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [26]: /lookup/external-ref?access_num=10.1063/1.3187531&link_type=DOI [27]: #xref-ref-6-1 "View reference 6 in text" [28]: {openurl}?query=rft.jtitle%253DNat.%2BMater.%26rft.volume%253D18%26rft.spage%253D309%26rft_id%253Dinfo%253Adoi%252F10.1038%252Fs41563-019-0291-x%26rft_id%253Dinfo%253Apmid%252F30894760%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [29]: /lookup/external-ref?access_num=10.1038/s41563-019-0291-x&link_type=DOI [30]: /lookup/external-ref?access_num=30894760&link_type=MED&atom=%2Fsci%2F373%2F6557%2F854.atom [31]: #xref-ref-7-1 "View reference 7 in text" [32]: {openurl}?query=rft.jtitle%253DAdv.%2BIntell.%2BSyst.%26rft.volume%253D2%26rft.spage%253D2000141%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [33]: #xref-ref-8-1 "View reference 8 in text" [34]: {openurl}?query=rft.jtitle%253DNat.%2BCommun.%26rft.volume%253D9%26rft.spage%253D4661%26rft_id%253Dinfo%253Adoi%252F10.1038%252Fs41467-018-07052-w%26rft_id%253Dinfo%253Apmid%252F30405124%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [35]: /lookup/external-ref?access_num=10.1038/s41467-018-07052-w&link_type=DOI [36]: /lookup/external-ref?access_num=30405124&link_type=MED&atom=%2Fsci%2F373%2F6557%2F854.atom [37]: #xref-ref-9-1 "View reference 9 in text" [38]: {openurl}?query=rft.jtitle%253DNature%26rft.volume%253D589%26rft.spage%253D52%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [39]: #xref-ref-10-1 "View reference 10 in text" [40]: {openurl}?query=rft.jtitle%253DJ.%2BAppl.%2BPhys.%26rft.volume%253D123%26rft.spage%253D180901%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [41]: #xref-ref-11-1 "View reference 11 in text" [42]: {openurl}?query=rft.jtitle%253DPLOS%2BONE%26rft.volume%253D9%26rft.spage%253De85175%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx