Self-assembling crystalline peptide microrod for neuromorphic function implementation

Abstract

•Memristor exhibits repeatable switching behavior and flexible relaxation process•Ion transport confinement was observed via in situ electrostatic force microscopy•Memristor emulates common nociceptive behaviors and neuropathic pain Resistive memory technologies have gained significant attention due to their potential for creating bio-inspired intelligent systems that memorize and learn locally. However, a major bottleneck of memristors lies in the reliability issue due to the uncontrollable filament dynamic induced by the inherent stochastic nature of ion migration, hindering their applications for neuromorphic computing. Here we introduce a convenient approach in which, in a planar structured cell, solution-processable diphenylalanine microrod (good ionic permeability) is used as the switching medium to facilitate a quasi-one-dimensional channel for Ag ion diffusion. The volatile feature, highly repeatable switching behavior, and flexible relaxation process of our peptide-based memristor render it a promising candidate to perform neuromorphic function implementation. For example, by tuning the external electrical stimuli, our device can emulate common nociceptive behaviors and neuropathic pain in damaged tissue. Memristive devices offer desirable voltage-regulated conductance switching and promise to address zettabyte storage challenges in the big-data era. Generally, most of these reported devices use amorphous solids, where the structural and compositional inhomogeneity is regarded as the origin of stochastic variability. Self-assembling peptide crystals with solution-processed fabrication, controllable morphologies, and structural stability are therefore a promising candidate to address the reliability issues. Here we report a planar diffusive memristor that possesses reliable switching characteristics based on a quasi-one-dimensional crystallized material: diphenylalanine (FF) microrod (MR). This element offers a preferential ion migration path, confining conductive filaments in a defined crystalline surface and therefore reducing programming stochasticity. Ion transport confinement along FF MR was observed via in situ electrostatic force microscopy technique. Additionally, FF MR memristor with high switching uniformity and reproducible relaxation dynamic provides an ideal hardware platform for reliable nociceptor emulation and hardware identification of four-bit decimal numbers. Memristive devices offer desirable voltage-regulated conductance switching and promise to address zettabyte storage challenges in the big-data era. Generally, most of these reported devices use amorphous solids, where the structural and compositional inhomogeneity is regarded as the origin of stochastic variability. Self-assembling peptide crystals with solution-processed fabrication, controllable morphologies, and structural stability are therefore a promising candidate to address the reliability issues. Here we report a planar diffusive memristor that possesses reliable switching characteristics based on a quasi-one-dimensional crystallized material: diphenylalanine (FF) microrod (MR). This element offers a preferential ion migration path, confining conductive filaments in a defined crystalline surface and therefore reducing programming stochasticity. Ion transport confinement along FF MR was observed via in situ electrostatic force microscopy technique. Additionally, FF MR memristor with high switching uniformity and reproducible relaxation dynamic provides an ideal hardware platform for reliable nociceptor emulation and hardware identification of four-bit decimal numbers. The development of artificial intelligence (AI) and big-data analytics is driving a revolution in the methods of data processing and storage. Confronting the speed and energy consumption issues, these fields require a new computing system based on the neuromorphic computing concept to parallel retrieve, process, and store massive amounts of data.1Sebastian A. Le Gallo M. Khaddam-Aljameh R. Eleftheriou E. Memory devices and applications for in-memory computing.Nat. Nanotechnol. 2020; 15: 529-544Crossref PubMed Scopus (301) Google Scholar, 2Ielmini D. Wong H.S.P. In-memory computing with resistive switching devices.Nat. Electron. 2018; 1: 333-343Crossref Scopus (638) Google Scholar, 3Fuller E.J. Keene S.T. Melianas A. Wang Z. Agarwal S. Li Y. Tuchman Y. James C.D. Marinella M.J. Yang J.J. et al.Parallel programming of an ionic floating-gate memory array for scalable neuromorphic computing.Science. 2019; 364: 570Crossref PubMed Scopus (253) Google Scholar, 4Xia Q. Yang J.J. Memristive crossbar arrays for brain-inspired computing.Nat. Mater. 2019; 18: 309-323Crossref PubMed Scopus (497) Google Scholar, 5Lv Z. Wang Y. Chen J. Wang J. Zhou Y. Han S.-T. Semiconductor quantum dots for memories and neuromorphic computing systems.Chem. Rev. 2020; 120: 3941-4006Crossref PubMed Scopus (82) Google Scholar, 6van de Burgt Y. Melianas A. Keene S.T. Malliaras G. Salleo A. Organic electronics for neuromorphic computing.Nat. Electron. 2018; 1: 386-397Crossref Scopus (345) Google Scholar Beyond complementary metal oxide semiconductor (CMOS) device and technology, two-terminal memristors stand out as a promising candidate for neuromorphic computing, offering desirable attributes including subnanosecond switching times, picojoule multi-bit programmability, and highly integratable structure.7Guo J. Wang L. Liu Y. Zhao Z. Zhu E. Lin Z. Wang P. Jia C. Yang S. Lee S.-J. et al.Highly reliable low-voltage memristive switching and artificial synapse enabled by van der Waals integration.Matter. 2020; 2: 965-976Abstract Full Text Full Text PDF Scopus (24) Google Scholar, 8Xu W. Min S.-Y. Hwang H. Lee T.-W. Organic core-sheath nanowire artificial synapses with femtojoule energy consumption.Sci. Adv. 2016; 2: e1501326Crossref PubMed Scopus (290) Google Scholar, 9Shi Y. Liang X. Yuan B. Chen V. Li H. Hui F. Yu Z. Yuan F. Pop E. Wong H.S.P. et al.Electronic synapses made of layered two-dimensional materials.Nat. Electron. 2018; 1: 458-465Crossref Scopus (293) Google Scholar, 10Liu C. Chen H. Wang S. Liu Q. Jiang Y.-G. Zhang D.W. Liu M. Zhou P. Two-dimensional materials for next-generation computing technologies.Nat. Nanotechnol. 2020; 15: 545-557Crossref PubMed Scopus (180) Google Scholar, 11Wang M. Cai S. Pan C. Wang C. Lian X. Zhuo Y. Xu K. Cao T. Pan X. Wang B. et al.Robust memristors based on layered two-dimensional materials.Nat. Electron. 2018; 1: 130-136Crossref Scopus (324) Google Scholar, 12Xue W. Gao S. Shang J. Yi X. Liu G. Li R.-W. Recent advances of quantum conductance in memristors.Adv. Electron. Mater. 2019; 5: 1800854Crossref Scopus (27) Google Scholar, 13Wu Z. Lu J. Shi T. Zhao X. Zhang X. Yang Y. Wu F. Li Y. Liu Q. Liu M. A habituation sensory nervous system with memristors.Adv. Mater. 2020; 32: 2004398Crossref Scopus (24) Google Scholar, 14Lee H.-S. Sangwan V.K. Rojas W.A.G. Bergeron H. Jeong H.Y. Yuan J. Su K. Hersam M.C. Dual-gated MoS2 memtransistor crossbar array.Adv. Funct. Mater. 2020; 30: 2003683Crossref Scopus (33) Google Scholar, 15Hwang B. Lee J.-S. Recent advances in memory devices with hybrid materials.Adv. Electron. Mater. 2019; 5: 1800519Crossref Scopus (59) Google Scholar, 16Goswami S. Rath S.P. Thompson D. Hedström S. Annamalai M. Pramanick R. Ilic B.R. Sarkar S. Hooda S. Nijhuis C.A. et al.Charge disproportionate molecular redox for discrete memristive and memcapacitive switching.Nat. Nanotechnol. 2020; 15: 380-389Crossref PubMed Scopus (29) Google Scholar Memristive devices generally consist of a simple metal-insulator-metal configuration with two or more switchable resistance states via the application of “set” and “reset” operation. The typical switching process relies on the formation and rupture of one or more conductive filaments with relatively high conductance as a result of a localized defect concentration. Defects could stem from switching materials (for example, anion vacancies in metal oxides in resistive random-access memory [RRAM]) or from active electrodes via ionic mass transport across the insulating media in conductive bridge random-access memory (CBRAM).17Nili H. Adam G.C. Hoskins B. Prezioso M. Kim J. Mahmoodi M.R. Bayat F.M. Kavehei O. Strukov D.B. Hardware-intrinsic security primitives enabled by analogue state and nonlinear conductance variations in integrated memristors.Nat. Electron. 2018; 1: 197-202Crossref Scopus (84) Google Scholar, 18Jiang H. Li C. Zhang R. Yan P. Lin P. Li Y. Yang J.J. Holcomb D. Xia Q. A provable key destruction scheme based on memristive crossbar arrays.Nat. Electron. 2018; 1: 548-554Crossref Google Scholar, 19Herder C. Yu M. Koushanfar F. Devadas S. Physical unclonable functions and applications: a tutorial.Proc. IEEE. 2014; 102: 1126-1141Crossref Scopus (647) Google Scholar Unfortunately, due to the three-dimensional uncontrollable filament dynamic induced by the inherent stochastic nature of ion/atom migration, there are great variations from cycle to cycle and device to device in set and reset voltages, high-resistance state (HRS), low-resistance state (LRS), etc., which lead to great uncertainty, nonuniformity, and unreliability of power consumption and resistance value, hindering its application for neuromorphic computing.20Alaghi A. Qian W. Hayes J.P. The promise and challenge of stochastic computing.IEEE Trans. Comput. Aid. Des. Integr. Circ. Syst. 2018; 37: 1515-1531Crossref Scopus (103) Google Scholar, 21Adam G.C. Khiat A. Prodromakis T. Challenges hindering memristive neuromorphic hardware from going mainstream.Nat. Commun. 2018; 9: 5267Crossref PubMed Scopus (39) Google Scholar, 22Midya R. Wang Z. Asapu S. Joshi S. Li Y. Zhuo Y. Song W. Jiang H. Upadhay N. Rao M. et al.Artificial neural network (ANN) to spiking neural network (SNN) converters based on diffusive memristors.Adv. Electron. Mater. 2019; 5: 1900060Crossref Scopus (42) Google Scholar In general, most of these reported devices use amorphous solids, where the structural and compositional inhomogeneity is regarded as the origin of intrinsically stochastic variability.23Wang Z. Wu H. Burr G.W. Hwang C.S. Wang K.L. Xia Q. Yang J.J. Resistive switching materials for information processing.Nat. Rev. Mater. 2020; 5: 173-195Crossref Scopus (266) Google Scholar, 24Wang Z. Joshi S. Savel’ev S.E. Jiang H. Midya R. Lin P. Hu M. Ge N. Strachan J.P. Li Z. et al.Memristors with diffusive dynamics as synaptic emulators for neuromorphic computing.Nat. Mater. 2017; 16: 101-108Crossref PubMed Scopus (1100) Google Scholar, 25Lu Y. Alvarez A. Kao C.-H. Bow J.-S. Chen S.-Y. Chen I.W. An electronic silicon-based memristor with a high switching uniformity.Nat. Electron. 2019; 2: 66-74Crossref Scopus (26) Google Scholar, 26Zhao X. Ma J. Xiao X. Liu Q. Shao L. Chen D. Liu S. Niu J. Zhang X. Wang Y. et al.Breaking the current-retention dilemma in cation-based resistive switching devices utilizing graphene with controlled defects.Adv. Mater. 2018; 30: 1705193Crossref Scopus (156) Google Scholar Crystalline resistive switching materials are therefore promising candidates to address the reliability issues of memristors for practical neuromorphic computing.27Milano G. Luebben M. Ma Z. Dunin-Borkowski R. Boarino L. Pirri C.F. Waser R. Ricciardi C. Valov I. Self-limited single nanowire systems combining all-in-one memristive and neuromorphic functionalities.Nat. Commun. 2018; 9: 5151Crossref PubMed Scopus (70) Google Scholar, 28Jiang H. Belkin D. Savel’ev S.E. Lin S. Wang Z. Li Y. Joshi S. Midya R. Li C. Rao M. et al.A novel true random number generator based on a stochastic diffusive memristor.Nat. Commun. 2017; 8: 882Crossref PubMed Scopus (177) Google Scholar, 29Lee D. Kwak M. Moon K. Choi W. Park J. Yoo J. Song J. Lim S. Sung C. Banerjee W. et al.Various threshold switching devices for integrate and fire neuron applications.Adv. Electron. Mater. 2019; 5: 1800866Crossref Scopus (46) Google Scholar, 30Fu T. Liu X. Gao H. Ward J.E. Liu X. Yin B. Wang Z. Zhuo Y. Walker D.J.F. Yang J.J. et al.Bioinspired bio-voltage memristors.Nat. Commun. 2020; 11: 1861Crossref PubMed Scopus (63) Google Scholar, 31Zhu X. Wang Q. Lu W.D. Memristor networks for real-time neural activity analysis.Nat. Commun. 2020; 11: 2439Crossref PubMed Scopus (42) Google Scholar, 32Yoon J.H. Wang Z. Kim K.M. Wu H. Ravichandran V. Xia Q. Hwang C.S. Yang J.J. An artificial nociceptor based on a diffusive memristor.Nat. Commun. 2018; 9: 417Crossref PubMed Scopus (171) Google Scholar, 33Choi S. Tan S.H. Li Z. Kim Y. Choi C. Chen P.-Y. Yeon H. Yu S. Kim J. SiGe epitaxial memory for neuromorphic computing with reproducible high performance based on engineered dislocations.Nat. Mater. 2018; 17: 335-340Crossref PubMed Scopus (316) Google Scholar, 34Yeon H. Lin P. Choi C. Tan S.H. Park Y. Lee D. Lee J. Xu F. Gao B. Wu H. et al.Alloying conducting channels for reliable neuromorphic computing.Nat. Nanotechnol. 2020; 15: 574-579Crossref PubMed Scopus (59) Google Scholar Although widely used in electronic hardware, conventional inorganic crystals/semiconductors display several disadvantages, such as harsh conditions for relatively complex crystal growth and the utilization of toxic elements. Organic crystalline materials overcome some of these shortcomings, but several issues, such as poor oxidation stability and, in some cases, complicated synthetic processes, may hinder their commercial translation. Peptide crystallized self-assemblies, typically based on aromatic short peptides, could address these issues owing to their low-cost solution-processed fabrication, biocompatibility and biodegradability, strong oxidation stability, and well-crystallized structure. Besides, peptide-based electronic devices may offer a bio-machine interface. Diphenylalanine (FF), the most-studied short peptide, self-assembles into various crystalline architectures from nanofiber and nanosphere on the nanoscale to microtube and microrod (MR) on the microscale, by controlling FF solution conditions.35Fleming S. Ulijn R.V. Design of nanostructures based on aromatic peptide amphiphiles.Chem. Soc. Rev. 2014; 43: 8150-8177Crossref PubMed Google Scholar These supramolecular structures possess the same crystallographic lattice structures, indicating a facile and low-temperature method to develop abundant nanostructures by regulating the dewetting process of FF solution. In this regard, FF supramolecular structures could serve as an ideal solution to scale down the device size beyond the limits of top-down lithographic patterning, and may offer a suitable platform to regulate ionic movement in a customized channel and probe the mechanism of the switching effect. In addition, recent studies indicated that FF self-assemblies (e.g., FF MR) process desirable ferroelectric and photoluminescent properties, potentially granting memristors unique functions such as in-sensor computing.36Gan Z. Wu X. Zhu X. Shen J. Light-induced ferroelectricity in bioinspired self-assembled diphenylalanine nanotubes/microtubes.Angew. Chem. Int. Ed. 2013; 52: 2055-2059Crossref PubMed Scopus (66) Google Scholar Hence, in this article, by confining the formation/rupture of Ag conductive filaments along a short-peptide self-assembling structure (FF MR), the planar structured Ag/FF MR/Ag memristor shows reproducible threshold voltage with temporal variation (σ/μ) of lower than 0.09, high uniform cell-to-cell performance (five cells constructed by a single MR), and short-term memory features with repeatable relaxation dynamics compared with Ag/evaporated FF film/Ag and Ag/SiO2/Ag. The MR crystalline surface with low activation energy for Ag ion migration offers a preferential ion transport path, and thus promotes confinement of conducting filament formation. This confinement effect was directly observed via in situ electrostatic force microscopy, which is sensitive to ionic dynamic with sufficiently high resolution. In this MR memristor, the volatile and nonvolatile switching dynamic could be achieved by simply adjusting the electrode spacing, and we discovered that our basic device (2-μm electrode spacing) shows repeatable volatile switching and uniform relaxation dynamics. Furthermore, emulation of nociceptor's neuropathic pain and hardware identification of four-bit decimal numbers (0–15) have been demonstrated on FF MR memristor by virtue of its reliable threshold switching characteristic. Additionally, the bioresorbable FF MR shows ultrafast dissolution behavior (within 100 s) upon high relative humidity, implying its suitability for further application in degradable electronic systems. FF MRs were realized with a bottom-up approach by finely optimizing the ratio and concentration of 1,1,1,3,3,3-hexafluoro-2-propanol (HFP)/deuteroxide (D2O) (Figure 1A, see experimental procedures), according to previously developed methods.37Li Q. Jia Y. Dai L. Yang Y. Li J. Controlled rod nanostructured assembly of diphenylalanine and their optical waveguide properties.ACS Nano. 2015; 9: 2689-2695Crossref PubMed Scopus (158) Google Scholar The obtained FF MRs process a smooth surface with diameters of ∼200–1,000 nm (Figure S1). We selected FF MR with the diameter of 500–600 nm as the switching element of the single MR device owing to its mechanical stability for next-stage device fabrication. Mechanical damage events during the transferring process are more likely to happen in FF MR with non-optimal diameter (e.g., <400 nm, Figure S1). The as-synthesized FF MR morphology studied by scanning electron microscopy (SEM) and atomic force microscopy (AFM) are depicted in Figures 1B and 1C. The morphology of the self-assembled FF MR is highly ordered with a nearly linear shape with the diameter and height of 542 nm and 408 nm, respectively. X-ray diffraction (XRD) measurement reveals that self-assembling FF MRs present much better crystalline characteristic than that of evaporated FF thin film (Figure 1D).37Li Q. Jia Y. Dai L. Yang Y. Li J. Controlled rod nanostructured assembly of diphenylalanine and their optical waveguide properties.ACS Nano. 2015; 9: 2689-2695Crossref PubMed Scopus (158) Google Scholar The highly ordered structures of FF MRs provide a good platform for high localization of switching event, which is expected to reduce stochastic variability in corresponding device performance.27Milano G. Luebben M. Ma Z. Dunin-Borkowski R. Boarino L. Pirri C.F. Waser R. Ricciardi C. Valov I. Self-limited single nanowire systems combining all-in-one memristive and neuromorphic functionalities.Nat. Commun. 2018; 9: 5151Crossref PubMed Scopus (70) Google Scholar Then, single FF MR-based memristor was obtained by bridging two electrodes (i.e., a 5-nm Cr adhesion film followed by a 60-nm Ag electrode; electrode spacing, 1, 2, and 5 μm) with isolated FF MR, resting on an insulating SiO2 substrate, the schematic diagram of the circuit of the FF MR-based memristor (Ag/FF MR/Ag) is displayed in Figure 1E (see experimental procedures, Figures S2 and S3 for fabrication details). Resistive switching with Ag or Cu active metal is usually an electrochemical process accompanied by atomic and ionic mass transport.38Waser R. Dittmann R. Staikov G. Szot K. Redox-based resistive switching memories – nanoionic mechanisms, prospects, and challenges.Adv. Mater. 2009; 21: 2632-2663Crossref Scopus (3989) Google Scholar The metal ion migration is expected to be a fundamental factor affecting the switching performance. Thus, we first evaluated the activation barrier of Ag ion migration in our active materials. Figure 1F depicts the Arrhenius plot of the electrical conductivity for Ag/FF MR/Ag and Ag/FF film (thermal evaporation)/Ag devices in a partially electroformed state achieved by a subthreshold voltage pulse (amplitude, 5 V; width, 8 s) (Figure S4). The conductance for both devices increases when temperature increases from 298 K to 373 K (Figure S4). From temperature-dependent Arrhenius plot, low activation energy for Ag+ migration of 0.23 eV in FF MRs that possess well-crystallized structure and 0.30 eV in evaporated FF thin film can be achieved. The activation energy of ion migration relies on the material's structure, ion-jumping distance, ionic radius, and the charge of ions. Difference in activation energy between FF MR and evaporated FF film may stem from distinct materials’ structures and ion-jumping distances. The value in FF MR device is significantly lower than the value reported for Ag transport on SiO2 (i.e., 0.77–1.24 eV),39Pi S. Ghadiri-Sadrabadi M. Bardin J.C. Xia Q. Nanoscale memristive radiofrequency switches.Nat. Commun. 2015; 6: 7519Crossref PubMed Scopus (79) Google Scholar,40McBrayer J.D. Swanson R.M. Sigmon T.W. Diffusion of metals in silicon dioxide.J. Electrochem. Soc. 1986; 133: 1242-1246Crossref Scopus (488) Google Scholar implying FF MR element could offer a preferential ion migration path. Therefore, we built Ag memristor in a planar configuration with FF MR switching element to determine if it would address variability issues (Figure 1G). To probe the initial electrical characteristics of the planar MR-based memristor, the current-voltage (I-V) features of Ag/SiO2/Ag and Au/FF MR/Au devices were compared with Ag/FF MR/Ag device, as presented in Figures 2A–2C (electrode spacing, 2 μm). The planar MR-based device can be regarded as an electrochemical metallization memristor, where the low-conductive FF MR functioned as a dielectric (Figure S5). The original state of the memristor is insulating and needs to be electroformed prior to cyclical switching events. A typical forming process can be achieved as shown in Figure 2A where a sufficient applied voltage (e.g., ∼7 V) abruptly switched the memristor to the high-conductance state (compliance current, 5 μA), which spontaneously relaxed back to the low-conductance state at close-to-zero voltage, suggesting transient switching behavior. After electroforming, nearly symmetric hysteresis loops were detected in both positive and negative sweeps, indicating a bipolar threshold switching process that is dramatically different from nonvolatile electrochemical metallization cell (Figure 2B), especially in the reset step. Due to the highly resistive property of FF MR (i.e., 0.043 S/m, Figure S5), the device exhibits a large switching on/off ratio of ∼105 and a steep switch-on slope of ∼40 mV/dec. The threshold voltage was highly independent of compliance settings, which remained at ∼1.6 V with various compliance currents, implying a field-driven mechanism of the memristor (Figure S6). In contrast to the Ag/FF MR/Ag device, both Ag/SiO2/Ag and Au/FF MR/Au devices remain in insulating configurations even under an applied high-voltage sweep (e.g., 20 V), as depicted in Figure 2C, suggesting that the observed transient switching effect of Ag/FF MR/Ag is originated from synergistic effort of the isolated FF MR and the active Ag electrodes. It is worth noting that the switching performance is closely related to the electrode spacing and the morphology of switching materials. Control devices of two configurations were prepared to compare the switching capabilities. The first type is Ag/FF MR/Ag devices with varying electrode spacing ranging from 1 μm to 5 μm (electrode spacing of our basic device is 2 μm) (Figures 2D–2F). The second type is Ag/thermal evaporated FF film/Ag with fixed electrode spacing of 2 μm (Figure 2G). The transient switching effect greatly depends on the electrode spacing. Similar threshold switching was also observed in the device with a shorter electrode spacing of 1 μm after electroforming process, as depicted in Figure 2D. The threshold voltage was 5-fold lower than our basic device. This change of operation voltage can be explained because shortening the electrode spacing can enhance the electric field intensity (electric field E is defined by E = V/day), which facilitate the transport of the metal ions through the dielectrics, thus accelerate the filament growth between Ag electrodes. However, a “dead” device stuck at an LRS was induced after just three positive sweep cycles due to the formation of a permanent conductive pathway in the device of a shorter electrode spacing. Although further reduction of electrode spacing (below 1 μm) means a higher integration density and may facilitate low-voltage nonvolatile memory applications, we selected devices of 2-μm electrode spacing as our basic devices due to their highly repeatable threshold switching characteristic and potentially volatile applications. In contrast, for longer electrode spacing (5 μm), the device maintained in an HRS even under voltage sweep from 0 V to 20 V, suggesting the difficulty of creating a filament that brings two Ag electrodes with such separation distance (Figure 2F). The Ag/thermal evaporated FF film/Ag struggled with high cycle-to-cycle variability with particularly precarious threshold voltage ranging from 3V to 6V, as indicated by the randomly distributed and unsmooth I-V loops in Figure 2G. In contrast, reproducible switching behavior was observed in our basic device (Figure 2E). Compared with Ag/thermal evaporated FF film/Ag, the improved switching uniformity in our basic device mainly originates from the one-dimensional (1D) structure confinement of ion transport in FF MR. To evaluate the switching repeatability of the FF MR device, we analyzed the cell-to-cell (cell 1, cell 2, cell 3, cell 4 and cell 5; electrode spacing, 2 μm) and cycle-to-cycle variations (100 cycles) of five FF MR devices based on a single FF MR (Figure 2H). As presented in Figures 2I and 2J, the as-fabricated five FF MR memristors all start from HRS, and show comparable forming and threshold switching behaviors with narrow distributions of the LRS/HRS current, as well as the forming and threshold voltages under positive voltage sweeps (forming voltage, 6.5–8.0 V; threshold voltage, 1.4–1.8 V), implying good cell-to-cell consistency of FF MR devices. Repeatable forming and threshold switching characteristics have been also achieved in 30 memory cells, each of which contains an independent FF MR (Figure S7). Repeatable I-V hysteresis loops in consecutive positive/negative sweeping over 100 cycles were observed, as depicted in Figure 2K. Figure 2L (bottom panel) shows the histogram of the threshold voltage distributions in both positive and negative sweeping of cell 1. The memristor exhibited a narrow distribution of threshold voltage at ∼1.64 V (−1.65 V) and quite a low degree of variability. According to the Gaussian fits to the threshold voltage histograms, the variabilities (defined as σ/μ, σ being the standard deviation and μ being the mean of the threshold voltage distribution) of threshold voltages in positive/negative sweeping are 0.095 and 0.060 (top panel), respectively, which are much lower than other amorphous material-based memristive devices.27Milano G. Luebben M. Ma Z. Dunin-Borkowski R. Boarino L. Pirri C.F. Waser R. Ricciardi C. Valov I. Self-limited single nanowire systems combining all-in-one memristive and neuromorphic functionalities.Nat. Commun. 2018; 9: 5151Crossref PubMed Scopus (70) Google Scholar,41Fang Y. Yu Z. Wang Z. Zhang T. Yang Y. Cai Y. Huang R. Improvement of HfOx-based RRAM device variation by inserting ALD TiN buffer layer.IEEE Electron Device Lett. 2018; 39: 819-822Crossref Scopus (32) Google Scholar, 42Parreira P. Paterson G.W. McVitie S. MacLaren D.A. Stability, bistability and instability of amorphous ZrO2 resistive memory devices.J. Phys. D Appl. Phys. 2016; 49: 095111Crossref Scopus (42) Google Scholar, 43Fantini A. Goux L. Degraeve R. Wouters D.J. Raghavan N. Kar G. Belmonte A. Chen Y. Govoreanu B. Jurczak M. Intrinsic switching variability in HfO2 RRAM.in: 5th IEEE International Memory Workshop. IEEE, 2013: 30-33Google Scholar Temporal switching repeatability in term of temporal dynamics was elucidated with voltage pulse tests (100 cycles). The memristor needs a limited delay time to be switched from HRS to LRS, and requires a relaxation time to return back to the HRS. The applied voltage pulses could yield repeatable delay and relaxation features (Figure 2M), indicating good reproducibility of the switching dynamics. In addition, by using a voltage pulse with larger amplitude (8 V, 1 ms), the turn-on delay time can be greatly reduced to 400 μs (see Figure S8). The FF MR device maintained a similar switching feature over a range of relative humidity (RH) (15%–65%), indicating good stability at low humidity (Figure S9). As a self-assembled peptide-based nanostructure, FF MR is sufficiently stable for weeks under an RH of ∼80% (Figure S10), which can fully degrade in a fast manner by a humidifying treatment (see Figures S9 and S11 and Video S1). https://www.cell.com/cms/asset/c66993f8-f34f-45c5-8db4-91abf03c5cfe/mmc2.mp4Loading ... Download .mp4 (12.31 MB) Help with .mp4 files Video S1. Degradation process of FF MR induced by a humidifying treatment Dynamic properties of the fading memory (an electroformed state) were then investigated by applying stimulation pulses and monitoring the response currents. Under applied pulses with different amplitudes (2.5V, 3.5 V, 4.5 V, 5 V; duration, 25 ms), the memristor showed threshold switching to a high-conductance state (with current larger than 10−4 A). As the input pulse ended, the memristor then relaxed back to its initial low-conductance state over a finite relaxation time, suggesting a typical short-term memory feature (Figure 3A). The relaxation time (or decay time) increased from 8 ms to more than 100 ms with the increasing input amplitude. To evaluate the device relaxation uniformity, stimulation voltage pulses with intervals of 200 ms (larger than decay time) were applied successively, and the device exhibited similar on-state cur