Memristive devices are commonly benchmarked by the multi-level programmability of their resistance states. Neural networks utilizing memristor crossbar arrays as synaptic layers largely rely on this feature. However, the dynamical properties of memristors, such as the adaptive response times arising from the exponential voltage dependence of the resistive switching speed remain largely unexploited. Here, we propose an information processing scheme which fundamentally relies on the latter. We realize simple dynamical memristor circuits capable of complex temporal information processing tasks. We demonstrate an artificial neural circuit with one nonvolatile and one volatile memristor which can detect a neural spike pattern in a very noisy environment, fire a single voltage pulse upon successful detection and reset itself in an entirely autonomous manner. Furthermore, we implement a circuit with only two nonvolatile memristors which can learn the operation of an external dynamical system and perform the corresponding time-series prediction with high accuracy.

Achieving low-voltage, nanosecond multi-level programming and non-destructive read-out of ferroelectric non-volatile memories (NVM) is critical for analog in-memory computing architectures relying on ferroelectric capacitive devices (FeCap). We integrate HfO2/ZrO2 ferroelectric nanolayers concurrently in the BEOL of CMOS and on SiO2/Si, achieving nanosecond multilevel switching with programming voltages below 5 V. Partial ferroelectric switching enhances FeCap endurance above 1011 cycles, leading to MemCapacitance (MC) states with non-destructive read-out and 10-year retention. However, experiments reveal the collapse of the MC window for read frequencies above 1 MHz. To overcome this speed limit, we introduce a novel, non-destructive readout methodology. Using electrical pulses with widths down to 20 ps, below the RC time constant of the FeCaps, we enable measurement of the polarization-dependent leakage current, providing ultrafast and non-destructive read operations at only 14 fJ.

ABSTRACT Filamentary Ag memristors are at the forefront of next‐generation artificial neuromorphic networks. However, their fabrication routinely involves multiple lithography steps and vacuum techniques, driving processing complexity and length. In this study, we explore a lateral memristor structure consisting of an Ag‐based filamentary memristor on a SiO2 surface. A new, aqueous Ag nanoparticle (NP) nucleation technique is used for simplified memristor fabrication. The synthesis is an adaptation of an electroless metal deposition process used in microelectronic interconnect fabrication. Lithographically structured Pt gaps act as nucleation seeding sites, where single nanoparticles selectively grow. The resulting particles on the chip were then conditioned via an electromigration process to perform filamentary memristor operations. Operation voltages below 0.5 V and reproducible cycling are demonstrated. The location on a surface makes them ideal for sensing the environment. The responsivity to changes in the ambient was showcased by the broad influence of atmospheric humidity on the memristive hysteresis, enabling global gating strategies in future neuromorphic circuits.

Ta_2O_5 nonvolatile memristors are used as compact, traceable, and well-controllable dynamic reservoir computing layers to perform time-series prediction tasks. The strongly voltage-dependent switching speed is utilized for information processing. It enables the configuration of tailorable programming and forgetting times in response to the positive and negative driving voltage pulses. Benchmarking this framework on time-series prediction problems reveals that the configurable forgetting dynamics enables a high prediction accuracy using a rather small number of memristive input channels. The training is based either on optimizing the output layer using linear regression with fixed forgetting times, or on optimizing the forgetting times as well. In the first case, six memristive channels, while in the second, only two memristive channels are used to demonstrate excellent prediction accuracy for the benchmark tasks. This scheme allows for the tunability of the operating frequency over many orders of magnitude: by adjusting the input voltage levels, the information processing speed of the same memristive dynamic layer can be increased from the kHz to the MHz range while maintaining excellent prediction accuracy. These findings demonstrate the merits of memristor based dynamic networks in the analysis, prediction and recovery of fast temporal signals, approaching telecommunication data rates.

Oscillating neural networks are promising candidates for a new computational paradigm, where complex optimization problems are solved by physics itself through the synchronization of coupled oscillating circuits. VO 2 Mott memristors are particularly promising building blocks for such oscillating neural networks. Until now, however, not only the maximum frequency of VO 2 oscillating neural networks, but also the maximum frequency of individual VO 2 oscillators is severely limited, which has restricted their efficient and energy‐saving use. In this study, it is showed how to increase the oscillating frequency by more than an order of magnitude into the 100 MHz range utilizing ultrasmall, ≈30 nm wide active volume VO 2 devices and optimizing the circuit layout for high frequency operation. In addition, the physical limiting factors of the oscillation frequencies are studied by investigating the complex switching dynamics of our nanoscale VO 2 devices. These dynamical studies, together with simulations, provide a clear conclusion on the maximum achievable operating frequencies and the optimal operating parameters under which these can be reached.

Reproducibility, endurance, driftless data retention, and fine resolution of the programmable conductance weights are key technological requirements against memristive artificial synapses in neural network applications. However, the inherent fluctuations in the active volume impose severe constraints on the weight resolution. In order to understand and push these limits, a comprehensive noise benchmarking and noise reduction protocol is introduced. Our approach goes beyond the measurement of steady-state readout noise levels and tracks the voltage-dependent noise characteristics all along the resistive switching I(V) curves. Furthermore, we investigate the tunability of the noise level by dedicated voltage cycling schemes in our filamentary Ta_2O_5 memristors. This analysis highlights a broad, order-of-magnitude variability of the possible noise levels behind seemingly reproducible switching cycles. Our nonlinear noise spectroscopy measurements identify a subthreshold voltage region with voltage-boosted fluctuations. This voltage range enables the reconfiguration of the fluctuators without resistive switching, yielding a highly denoised state within a few subthreshold cycles.

Memristive devices have drawn significant interest due to their use in novel paradigms such as neuromorphic computing. Neuromorphic systems are developed by implementing artificial neurons and synapses on a hardware level. Hence, memristors with multipurpose and reconfigurable neuromorphic functionalities could be highly beneficial in the design process. In this study, we experimentally verify that both neuronal and synaptic functions can be implemented on a single memristor. By controlling the device current at two different levels, the memristor operates in either a volatile or a nonvolatile retention regime. These two operation regimes are essential to mimic neuronal or synaptic behavior. Towards this end, we use an alloyed filamentary memristor (AgSn/SiO2/Pt) composed of ions with differing mobilities enabling both integrate and fire (IF) operation in the volatile regime and synaptic weights in the nonvolatile regime. By only changing the current compliance, these devices switch reliably between the aforementioned retention regimes. Additionally, our proposed training method significantly improves switching variability in the volatile regime. We show how the mean set voltage statistically reduce from 1.2 to 0.2 V; and the standard deviation of the set voltages reduced from 0.52 to 0.03 V.

Brain-inspired computing solutions require a suitable hardware platform, where complex operations can be realized at low power consumption. Ideally, the hardware can be reconfigured between multiple functionalities by tuning the corresponding device parameters. In this work, a three-terminal silver-tin alloyed memristor is demonstrated, where the resistive switching characteristics can be modulated by the gate voltage. The polarity of the gate voltage determines the volatility of the device. Positive gate voltages result in primarily nonvolatile switching, while negative gate voltages facilitate primarily volatile switching. In addition, the set voltage and low resistance state can be adjusted by the magnitude of gate voltage both in the volatile and nonvolatile regimes. The dimensions of the active switching volume are 40 nm x 6 nm x 10 nm, making the design one of the most compact three-terminal memristor. Such an ultrasmall, versatile memristive device represents a viable candidate for reconfigurable, neuromorphic hardware, where the basic building blocks can be conveniently customized to perform either synaptic or neural operations.

Oscillating neural networks are promising candidates for a new computational paradigm, where complex optimization problems are solved by physics itself through the synchronization of coupled oscillating circuits. VO2 Mott memristors are particularly promising building blocks for such oscillating neural networks. Until now, however, not only the maximum frequency of VO2 oscillating neural networks, but also the maximum frequency of individual VO2 oscillators is severely limited, which has restricted their efficient and energy-saving use. In this study, it is showed how to increase the oscillating frequency by more than an order of magnitude into the 100 MHz range utilizing ultrasmall, approximate to 30 nm wide active volume VO2 devices and optimizing the circuit layout for high frequency operation. In addition, the physical limiting factors of the oscillation frequencies are studied by investigating the complex switching dynamics of our nanoscale VO2 devices. These dynamical studies, together with simulations, provide a clear conclusion on the maximum achievable operating frequencies and the optimal operating parameters under which these can be reached.

AbstractThe human brain facilitates information processing via generating and receiving temporal patterns of short voltage pulses, a.k.a. neural spikes. This approach simultaneously grants low‐power operation as well as a high degree of noise immunity and fault tolerance at a small footprint and simplistic structure of the neurons. To date, the latter two key features are critically missing from the toolbox of artificial spiking neural network hardware, hindering the development of scalable and sustainable artificial intelligence (AI) platforms. Here, a compact, gate‐tunable neuron circuit is demonstrated, and its potential as a functional leaky integrate‐and‐fire (LIF) neuron is explored. It relies on a single nanoscale three‐terminal (3T) memristor device, which has been downscaled by 30% compared to previous work, where the set voltage and, thereby, the spiking probability of the neuron circuit can be widely tuned by the low‐voltage operation of the gate electrode. The influence of the gate voltage on the two‐terminal (2T) current–voltage characteristics is measured, statistically analyzed, and further utilized in a custom‐built LTspice model. The circuit simulations account for the experimentally observed, adjustable set voltage. The presented results demonstrate the merits of 3T memristors as compact, tunable, and versatile artificial neurons for neuromorphic computing applications.

Beyond-Moore computing technologies are expected to provide a sustainable alternative to the von Neumann approach not only due to their down-scaling potential but also via exploiting device-level functional complexity at the lowest possible energy consumption. The dynamics of the Mott transition in correlated electron oxides, such as vanadium dioxide, has been identified as a rich and reliable source of such functional complexity. However, its full potential in high-speed and low-power operation has been largely unexplored. We fabricated nanoscale VO2 devices embedded in a broadband test circuit to study the speed and energy limitations of their resistive switching operation. Our picosecond time-resolution, real-time resistive switching experiments and numerical simulations demonstrate that tunable low-resistance states can be set by the application of 20 ps long, <1.7 V amplitude voltage pulses at 15 ps incubation times and switching energies starting from a few femtojoule. Moreover, we demonstrate that at nanometer-scale device sizes not only the electric field induced insulator-to-metal transition but also the thermal conduction limited metal-to-insulator transition can take place at time scales of 100s of picoseconds. These orders of magnitude breakthroughs can be utilized to design high-speed and low-power dynamical circuits for a plethora of neuromorphic computing applications from pattern recognition to numerical optimization.

Filamentary resistive switching (RS) devices are not only considered as promising building blocks for brain-inspired computing architectures but also realize an unprecedented operation regime where the active device volume reaches truly atomic dimensions. Such atomically sized RS filaments represent the quantum transport regime, where the transmission eigenvalues of the conductance channels are considered a specific device fingerprint. Here, we gain insight into the quantum transmission properties of close-to-atomic-sized RS filaments formed across an insulating Ta2O5 layer through superconducting subgap spectroscopy. This method reveals the transmission density function of the open conduction channels contributing to the device's conductance. Our analysis confirms the formation of truly atomic-sized filaments composed of 3-8 Ta atoms at their narrowest cross-section. We find that this diameter remains unchanged upon RS. Instead, the switching is governed by the redistribution of oxygen vacancies or tantalum cations within the filamentary volume. The set/reset process results in the reduction/formation of an extended barrier at the bottleneck of the filament, which enhances/reduces the transmission of the highly open conduction channels. This transmission variability facilitates neuromorphic electronic applications in nanosized artificial synapses reaching the ultimate atomic scale.

Analog tunable memristors are widely utilized as artificial synapses in various neural network applications. However, exploiting the dynamical aspects of their conductance change to implement active neurons is still in its infancy, awaiting the realization of efficient neural signal recognition functionalities. Here we experimentally demonstrate an artificial neural information processing unit that can detect a temporal pattern in a very noisy environment, fire a single output spike upon successful detection and reset itself in a fully unsupervised, autonomous manner. This circuit relies on the dynamical operation of only two memristive blocks: a non-volatile Ta$_2$O$_5$ device and a volatile VO$_2$ unit. A fading functionality with exponentially tunable memory time constant enables adaptive operation dynamics, which can be tailored for the targeted temporal pattern recognition task. In the trained circuit false input patterns only induce short-term variations. In contrast, the desired signal activates long-term memory operation of the non-volatile component, which triggers a firing output of the volatile block.

The resistance state of filamentary memristors can be tuned by relocating only a few atoms at interatomic distances in the active region of a conducting filament. Thereby the technology holds promise not only in its ultimate downscaling potential and energy efficiency but also in unprecedented speed. Yet, the breakthrough in high-frequency applications still requires the clarification of the dominant mechanisms and inherent limitations of ultra-fast resistive switching. Here we investigate bipolar, multilevel resistive switchings in tantalum pentoxide based memristors with picosecond time resolution. We experimentally demonstrate cyclic resistive switching operation due to 20 ps long voltage pulses of alternating polarity. Through the analysis of the real-time response of the memristor we find that the set switching can take place at the picosecond time-scale where it is only compromised by the bandwidth limitations of the experimental setup. In contrast, the completion of the reset transitions significantly exceeds the duration of the ultra-short voltage bias, demonstrating the dominant role of thermal diffusion and underlining the importance of dedicated thermal engineering for future high-frequency memristor circuit applications.

Memristive devices have attracted significant attention due to their downscaling potential, low power operation, and fast switching performance. Their inherent properties make them suitable for emerging applications such as neuromorphic computing, in-memory computing, and reservoir computing. However, the different applications demand either volatile or nonvolatile operation. In this study, we demonstrate how compliance current and specific material choices can be used to control the volatility and nonvolatility of memristive devices. Especially, by mixing different materials in the active electrode, we gain additional design parameters that allow us to tune the devices for different applications. We found that alloying Ag with Sn stabilizes the nonvolatile retention regime in a reproducible manner. Additionally, our alloying approach improves the reliability, endurance, and uniformity of the devices. We attribute these advances to stabilization of the filament inside the switching medium by the inclusion of Sn in the filament structure. These advantageous properties of alloying were found by investigating a choice of six electrode materials (Ag, Cu, AgCu-1, AgCu-2, AgSn-1, AgSn-2) and three switching layers (SiO2, Al2O3, HfO2).

Filamentary resistive switching devices are not only considered as promising building blocks for brain-inspired computing architectures, but they also realize an unprecedented operation regime, where the active device volume reaches truly atomic dimensions. Such atomic-sized resistive switching filaments represent the quantum transport regime, where the transmission eigenvalues of the conductance channels are considered as a specific device fingerprint. Here, we gain insight into the quantum transmission properties of close-to-atomic-sized resistive switching filaments formed across an insulating Ta$_2$O$_5$ layer through superconducting subgap spectroscopy. This method reveals the transmission density function of the open conduction channels contributing to the device conductance. Our analysis confirms the formation of truly atomic-sized filaments composed of 3-8 Ta atoms at their narrowest cross-section. We find that this diameter remains unchanged upon resistive switching. Instead, the switching is governed by the redistribution of oxygen vacancies within the filamentary volume. The set/reset process results in the reduction/formation of an extended barrier at the bottleneck of the filament which enhances/reduces the transmission of the highly open conduction channels.

Two-dimensional materials have been widely investigated to implement memristive devices for data storage or neuromorphic computing applications because of their ultra-scaled thicknesses and clean interfaces. For example, resistance switching in hexagonal boron nitride (h-BN) has been demonstrated. This mechanism is most of the time attributed to the movement of metallic ions. It has however also been reported when h-BN is contacted with two inert electrodes such as graphene or Pt. We suggest here that the switching mechanism of the latter devices, which has not yet been clearly established, relies on locals change of the electronic structure of h-BN as caused by atomic defects, e.g., multi-vacancies. This class of intrinsic h-BN defects can create electrically controllable interlayer bridges. We use a combination of hybrid density functional theory and the Non-equilibrium Green’s function formalism to show that a single interlayer bridge resulting from the presence of a trivacancy in a graphene/h-BN/graphene stack leads to a switching voltage of ~5 V and a high-to-low resistance ratio >100. Both values lie within the reported experimental range and thus confirm the likelihood that intrinsic defects play a key role in the resistance switching of h-BN in contact with inert electrodes.

In this study, the possibilities of noise tailoring in filamentary resistive switching memory devices are investigated. To this end, the resistance and frequency scaling of the low-frequency 1/f-type noise properties are studied in representative mainstream material systems. It is shown that the overall noise floor is tailorable by the proper material choice, as demonstrated by the order-of-magnitude smaller noise levels in Ta2O5 and Nb2O5 transition-metal oxide memristors compared to Ag-based devices. Furthermore, the variation of the resistance states allows orders-of-magnitude tuning of the relative noise level in all of these material systems. This behavior is analyzed in the framework of a point-contact noise model highlighting the possibility for the disorder-induced suppression of the noise contribution arising from remote fluctuators. These findings promote the design of multipurpose resistive switching units, which can simultaneously serve as analog-tunable memory elements and tunable noise sources in probabilistic computing machines.

Graphene nanogaps are considered as essential building blocks of two-dimensional electronic circuits, as they offer the possibility to interconnect a broad range of atomic-scale objects. Here we provide an insight into the microscopic processes taking place during the formation of graphene nanogaps through the detailed analysis of their low-frequency noise properties. Following the evolution of the noise level, we identify the fundamentally different regimes throughout the nanogap formation. By modeling the resistance and bias dependence of the noise, we resolve the major noise-generating processes: atomic-scale junction-width fluctuations in the nanojunction regime and sub-atomic gap-size fluctuations in the nanogap regime. As a milestone toward graphene-based atomic electronics, our results facilitate the automation of an optimized electrical breakdown protocol for high-yield graphene nanogap fabrication.