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.

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.

Összefoglalás. Napjainkra az információs technológiák fejlődése elérte azt a szintet, ahol a gyorsuló ütemben létrejövő adattömeg feldolgozásához már sok esetben elégtelenek a klasszikus, Neumann-elvek alapján működő számítógépek. A jelenség újszerű szoftveres megoldások, biológiai ihletésű algoritmusok, neurális hálózatok elterjedéséhez vezetett, ám ezek hatékony alkalmazásához teljesen új hardveres megoldások szükségesek. Jelen kézirat ilyen újszerű architektúrákhoz fejlesztett, Si-mikrochip-alapú memóriatulajdonsággal rendelkező nanoméretű áramköri elemek kísérleti eredményeit mutatja be, illetve azok egy-egy specifikus információfeldolgozási feladatra történő alkalmazhatóságával foglalkozik. Summary. Resistive switching memory devices, also known as memristors, are generally metal-insulator-metal nanostructures whose conductivity can be varied via electrical signals, enabling information storage in the value of the conductivity. Based on this property, memristive devices provide a promising platform for hardware-level encoding of large matrices. With a network of memristors, computationally intensive vector-matrix operations can be performed in a single step, considerably speeding up the operation of an artificial neural network (ANN). Memristors can also serve as real physical building blocks for biologically inspired algorithms through their neuromorphic properties. Via building simple circuits, such devices can be used to create oscillators or artificial neurons, which can be utilized for the implementation of oscillatory neural networks (ONN) or spiking neural networks (SNN). Another interesting feature of these memristive neuromorphic circuits is that they can be directly used for information processing tasks at the edge of a network. Memristors facilitate such edge computing applications which usually require energy-efficient operation and small size of the processor unit. Edge computing approaches have several advantages over centralized data processing from the aspect of security, e.g., significantly reducing time latency of sending large amounts of data to a central hub, and by processing sensitive data locally and independently of the central servers. Present work focuses on the experimental investigation of purpose-built nanoscale memristive devices, revealing their physical processes. A superconducting spectroscopy measurement technique is developed for the non-destructive detection of atomic scale memristive filaments during device operation (Török et al. 2020; Török et al. 2023). In addition, the tunable stochasticity of the nucleation process is revealed by statistical studies of the set process in silicon oxide memristors (Török et al. 2022). This finding provides a basis for the physical realization of neural activation functions, stochastically firing neurons or energy-efficient true random number generation. Finally, the applicability of the investigated nanoscale memristive devices is illustrated through two examples. The concept of a neuromorphic, memristor-based auditory sensing unit is presented, leading towards medical application in a fully implantable cochlear implant. Last, the feasibility of a hardware-level stochastic optimization procedure is introduced (Fehérvári et al. 2023), based entirely on memristive elements, utilizing tunable noise characteristics of the devices (Sánta et al. 2021).

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.

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.

Nanometer-scale resistive switching devices operated in the metallic conductance regime offer ultimately scalable and widely reconfigurable hardware elements for novel in-memory and neuromorphic computing architectures. Moreover, they exhibit high operation speed at low power arising from the ease of the electric-field-driven redistribution of only a small amount of highly mobile ionic species upon resistive switching. We investigate the memristive behavior of a so-far less explored representative of this class, the Ag/AgI material system in a point contact arrangement established by the conducting PtIr tip of a scanning probe microscope. We demonstrate stable resistive switching duty cycles and investigate the dynamical aspects of non-volatile operation in detail. The high-speed switching capabilities are explored by a custom-designed microwave setup that enables time-resolved studies of subsequent set and reset transitions upon biasing the Ag/AgI/PtIr nanojunctions with sub-nanosecond voltage pulses. Our results demonstrate the potential of Ag-based filamentary memristive nanodevices to serve as the hardware elements in high-speed neuromorphic circuits.

Nanometer-scale resistive switching devices operated in the metallic conductance regime offer ultimately scalable and widely reconfigurable hardware elements for novel in-memory and neuromorphic computing architectures. Moreover, they exhibit high operation speed at low power arising from the ease of the electric-field-driven redistribution of only a small amount of highly mobile ionic species upon resistive switching. We investigate the memristive behavior of a so-far less explored representative of this class, the Ag/AgI material system in a point contact arrangement established by the conducting PtIr tip of a scanning probe microscope. We demonstrate stable resistive switching duty cycles and investigate the dynamical aspects of non-volatile operation in detail. The high-speed switching capabilities are explored by a custom-designed microwave setup that enables time-resolved studies of subsequent set and reset transitions upon biasing the Ag/AgI/PtIr nanojunctions with sub-nanosecond voltage pulses. Our results demonstrate the potential of Ag-based filamentary memristive nanodevices to serve as the hardware elements in high-speed neuromorphic circuits.

The dynamical aspects of bipolar resistive switchings have been investigated in Nb/Nb2O5/PtIr nanojunctions. We found that the widely tuneable ON and OFF state resistances are well separated at low bias. On the other hand, the high-bias regime of the resistive switchings coincides with the onset of a high nonlinearity in the current-voltage characteristics, where the impedance of both states rapidly decreases and becomes equivalent around 50 Ω. This phenomenon enables the overriding of the RC limitations of fast switchings between higher resistance ON and OFF states. Consequently, nanosecond switching times between multiple resistance states due to subnanosecond voltage pulses are demonstrated. Moreover, this finding provides the possibility of impedance engineering by the appropriate choice of voltage signals, which facilitates that both the set and reset transitions take place in an impedance matched manner to the surrounding circuit, demonstrating the merits of ultra-fast operation of Nb2O5 based neuromorphic networks.