This study investigates resistive switching in amorphous undoped and Cr-doped vanadium oxide thin films synthesized by sputtering deposition at low oxygen partial pressure. Two different volatile threshold switching characteristics can occur as well as a non-volatile bipolar switching mechanism, depending on device stack symmetry and Cr-doping. The two threshold switching types are associated with different crystalline phases in the conduction filament created during an initial forming step. The first kind of threshold switching, observed for undoped vanadium oxide films, was, by its temperature dependence, proven to be associated with a thermally triggered insulator-to-metal transition in a crystalline VO₂ phase, whereas the threshold switch observed in chromium doped films is stable up to 90 °C and shows characteristics of an electronically induced Mott transition. This different behaviour for undoped versus doped films has been attributed to an increased stability of V³⁺ due to the Cr³⁺ doping (as evidenced by X-ray photoelectron spectroscopy analysis), probably favouring the creation of a crystalline Cr-doped V₂O₃ phase (rather than a Cr-doped VO₂ phase) during the energetic forming step. The symmetric Pt/a-(VCr)O_x/Pt device showing high temperature stable threshold switching may find interesting applications as a possible new selector device for resistive switching memory (ReRAM) crossbar arrays.

One of the key issues of resistive switching memory devices is the so called “forming” process, a one time process at a high voltage, which initializes the resistive switching at significantly lower voltages. With this study we identify the influence of the different layers – namely the insulating oxide layer (ZrO₂ and Ta₂O₅) and the reactive ohmic electrode layer (Hf, Ta and Pt) – on the forming voltage and the pristine capacitance of the devices. For this, the forming voltage and pristine capacitance is measured in dependence of the oxide layer thickness with different electrodes. The different slopes of the forming voltage – thickness relation for different top electrodes give an indication that the reactive ohmic electrode is oxidized from the oxide layer underneath and that the degree of the oxidation depends on the thickness of the oxide layer as well as the materials used for the oxide and electrode layer. This finding could be confirmed by X-ray photoelectron spectroscopy (XPS) and transmission electron microscopy (TEM) measurements. From the electrical measurements and the TEM images the thickness of the oxidized electrode layer could be estimated. The degree of the oxidation depends on the oxygen affinity of the oxide and electrode material. The interface dependent (thickness independent) part of the forming voltage is determined by the material of the electrode. The magnitude of this interface voltage could be correlated to the oxide free energy of the electrode material. These results can support the ongoing research towards resistive switching memory devices with a very low forming voltage or forming free behaviour.

The I–V switching curves of bipolar switching non-volatile ReRAM devices show peculiar characteristics, such as an abrupt ON switching and the existence of a universal switching voltage. This switching behavior has been explained by the presence of a filamentary process, in which the width of a conductive filament changes during switching resulting in different resistance states. Vice versa, similar (ON) switching behavior, e.g. that of volatile switching Cr-doped V₂O₃ devices, has been interpreted as an indication of the presence of similar filamentary switching. In this paper, we want to review the correlation between filamentary (width) switching and the (SET) I–V characteristics by discussing the existing models. For the Cr-doped V₂O₃ devices, on the other hand, it is argued that a different, constant filament width switching mode may be present.

The switching mechanism of valence change resistive memory devices is widely accepted to be an ionic movement of oxygen vacancies resulting in a valence change of the metal cations. However, direct experimental proofs of valence changes in memristive devices are scarce. In this work, we have employed hard X-ray photoelectron emission microscopy (PEEM) to probe local valence changes in Pt/ZrO_x/Ta memristive devices. The use of hard X-ray radiation increases the information depth, thus providing chemical information from buried layers. By extracting X-ray photoelectron spectra from different locations in the PEEM images, we show that zirconia in the active device area is reduced compared to a neighbouring region, confirming the valence change in the ZrO_x film during electroforming. Furthermore, we succeeded in measuring the Ta 4f spectrum for two different resistance states on the same device. In both states, as well as outside the device region, the Ta electrode is composed of different suboxides without any metallic contribution, hinting to the formation of TaO_x during the deposition of the Ta thin film. We observed a reduction of the Ta oxidation state in the low resistance state with respect to the high resistive state. This observation is contradictory to the established model, as the internal redistribution of oxygen between ZrO_x and the Ta electrode during switching would lead to an oxidation of the Ta layer in the low resistance state. Instead, we have to conclude that the Ta electrode takes an active part in the switching process in our devices and that oxygen is released and reincorporated in the ZrO_x/TaO_x bilayer during switching. This is confirmed by the degradation of the high resistance state during endurance measurements under vacuum.

Sequence learning, prediction and replay have been proposed to constitute the universal computations performed by the neocortex. The Hierarchical Temporal Memory (HTM) algorithm realizes these forms of computation. It learns sequences in an unsupervised and continuous manner using local learning rules, permits a context specific prediction of future sequence elements, and generates mismatch signals in case the predictions are not met. While the HTM algorithm accounts for a number of biological features such as topographic receptive fields, nonlinear dendritic processing, and sparse connectivity, it is based on abstract discrete-time neuron and synapse dynamics, as well as on plasticity mechanisms that can only partly be related to known biological mechanisms. Here, we devise a continuous-time implementation of the temporal-memory (TM) component of the HTM algorithm, which is based on a recurrent network of spiking neurons with biophysically interpretable variables and parameters. The model learns high-order sequences by means of a structural Hebbian synaptic plasticity mechanism supplemented with a rate-based homeostatic control. In combination with nonlinear dendritic input integration and local inhibitory feedback, this type of plasticity leads to the dynamic self-organization of narrow sequence-specific subnetworks. These subnetworks provide the substrate for a faithful propagation of sparse, synchronous activity, and, thereby, for a robust, context specific prediction of future sequence elements as well as for the autonomous replay of previously learned sequences. By strengthening the link to biology, our implementation facilitates the evaluation of the TM hypothesis based on experimentally accessible quantities. The continuous-time implementation of the TM algorithm permits, in particular, an investigation of the role of sequence timing for sequence learning, prediction and replay. We demonstrate this aspect by studying the effect of the sequence speed on the sequence learning performance and on the speed of autonomous sequence replay.