A Spiking Neural Network Based on the Model of VO2—Neuron

Collective dynamics and long-range order in thermal neuristor networks

In the pursuit of scalable and energy-efficient neuromorphic devices, recent research has unveiled a novel category of spiking oscillators, termed "thermal neuristors." These devices function via thermal interactions among neighboring vanadium dioxide resistive memories, emulating biological neuronal behavior. Here, we show that the collective dynamical behavior of networks of these neurons showcases a rich phase structure, tunable by adjusting the thermal coupling and input voltage. Notably, we identify phases exhibiting long-range order that, however, does not arise from criticality, but rather from the time non-local response of the system. In addition, we show that these thermal neuristor arrays achieve high accuracy in image recognition and time series prediction through reservoir computing, without leveraging long-range order. Our findings highlight a crucial aspect of neuromorphic computing with possible implications on the functioning of the brain: criticality may not be necessary for the efficient performance of neuromorphic systems in certain computational tasks.

Tuning the Intrinsic Stochasticity of Resistive Switching in VO2 Microresistors

Vanadium dioxide (VO2) microresistors exhibit resistive switching above a certain threshold voltage, allowing them to emulate neurons in neuromorphic systems. However, such devices present intrinsic cycle-to-cycle variations in their resistances and threshold voltages, which can be detrimental or beneficial, depending on their use. Here, we study this stochasticity in VO2 microresistors with various grain sizes and dimensions, through high-resolution electrical and optical measurements across numerous cycles. Our results highlight that the cycle-to-cycle variations in threshold voltage increase as the grain size becomes comparable to the device dimensions. We also present observations of multimodal threshold voltage distributions in the smaller-length resistors. To understand the underlying phenomenon, we investigate the relationship between the device insulating resistance and threshold voltage distributions, showing that these modes could correspond to distinct percolation paths and filaments. Our findings provide the first experimentally verified guidelines for designing VO2 devices with minimized/maximized stochasticity, depending on the targeted application.

Physical Origin of Negative Differential Resistance in V3 O5 and Its Application as a Solid-State Oscillator

Oxides that exhibit an insulator-metal transition can be used to fabricate energy-efficient relaxation oscillators for use in hardware-based neural networks but there are very few oxides with transition temperatures above room temperature. Here the structural, electrical, and thermal properties of V₃ O₅ thin films and their application as the functional oxide in metal/oxide/metal relaxation oscillators are reported. The V₃ O₅ devices show electroforming-free volatile threshold switching and negative differential resistance (NDR) with stable (<3% variation) cycle-to-cycle operation. The physical mechanisms underpinning these characteristics are investigated using a combination of electrical measurements, in situ thermal imaging, and device modeling. This shows that conduction is confined to a narrow filamentary path due to self-confinement of the current distribution and that the NDR response is initiated at temperatures well below the insulator-metal transition temperature where it is dominated by the temperature-dependent conductivity of the insulating phase. Finally, the dynamics of individual and coupled V₃ O₅ -based relaxation oscillators is reported, showing that capacitively coupled devices exhibit rich non-linear dynamics, including frequency and phase synchronization. These results establish V₃ O₅ as a new functional material for volatile threshold switching and advance the development of robust solid-state neurons for neuromorphic computing.

Parameter Estimation for Hindmarsh–Rose Neurons

In the paper, a new adaptive model of a neuron based on the Hindmarsh–Rose third-order model of a single neuron is proposed. The learning algorithm for adaptive identification of the neuron parameters is proposed and analyzed both theoretically and by computer simulation. The proposed algorithm is based on the Lyapunov functions approach and reduced adaptive observer. It allows one to estimate parameters of the population of the neurons if they are synchronized. The rigorous stability conditions for synchronization and identification are presented.

Voltage Pulse Driven VO2 Volatile Resistive Transition Devices as Leaky Integrate-and-Fire Artificial Neurons

In a hardware-based neuromorphic computation system, using emerging nonvolatile memory devices as artificial synapses, which have an inelastic memory characteristic, has attracted considerable interest. In contrast, the elastic artificial neurons have received much less attention. An ideal material system that is suitable for mimicking biological neurons is the one with volatile (or mono-stable) resistive change property. Vanadium dioxide (VO2) is a well-known material that exhibits an abrupt and volatile insulator-to-metal transition property. In this work, we experimentally demonstrate that pulse-driven two-terminal VO2 devices behave in a leaky integrate-and-fire (LIF) manner, and they elastically relax back to their initial value after firing, thus, mimicking the behavior of biological neurons. The VO2 device with a channel length of 20 µm can be driven to fire by a single long-duration pulse (>83 µs) or multiple short-duration pulses. We further model the VO2 devices as resistive networks based on their granular domain structure, with resistivities corresponding to the insulator or metallic states. Simulation results confirm that the volatile resistive transition under voltage pulse driving is caused by the formation of a metallic filament in an avalanche-like process, while this volatile metallic filament will relax back to the insulating state at the end of driving pulses. The simulation offers a microscopic view of the dynamic and abrupt filament formation process to explain the experimentally observed LIF behavior. These results suggest that VO2 insulator–metal transition could be exploited for artificial neurons.

Resistive switching on individual V2O5 nanoparticles encapsulated in fluorinated graphene films

Memristors currently attract much attention as basic building blocks for future neuromorphic electronics. Due to their unusual electronic, optical, magnetic, electrochemical, and structural properties, transition metal oxides offer much potential in the development of memristors. Recent trends in the design and fabrication of electronic devices have led to miniaturization of their working elements, with nanometer-sized structures enjoying increasing demand. In the present study, we investigated resistive switching on individual vanadium oxide (V₂O₅) crystal-hydrate nanoparticles, 2 to 10 nm in size, encapsulated in fluorinated graphene (FG). Measurements using a conductive atomic force microscope (c-AFM) probe showed that the core-shell V₂O₅/FG nanoparticles make it possible to achieve bipolar resistive switching, reproducible during 10⁴ switching cycles, with the ON/OFF current ratio reaching 10³-10⁵. The switching voltage of the structures depends on the thickness of the FG shells of the composite particles and equals ∼2-4 V. It is shown that the encapsulation of V₂O₅ particles in fluorinated graphene ensures a high stability of the resistive switching effect and, simultaneously, prevents the escape of water from the crystalline vanadium oxide hydrates. A qualitative model is proposed to describe the bipolar resistive switching effect in examined structures. Results reported in the present article will prove useful in creating bipolar nanoswitches.

A Low-Power Spike-Like Neural Network Design

Modern massively-parallel Graphics Processing Units (GPUs) and Machine Learning (ML) frameworks enable neural network implementations of unprecedented performance and sophistication. However, state-of-the-art GPU hardware platforms are extremely power-hungry, while microprocessors cannot achieve the performance requirements. Biologically-inspired Spiking Neural Networks (SNN) have inherent characteristics that lead to lower power consumption. We thus present a bit-serial SNN-like hardware architecture. By using counters, comparators, and an indexing scheme, the design effectively implements the sum-of-products inherent in neurons. In addition, we experimented with various strength-reduction methods to lower neural network resource usage. The proposed Spiking Hybrid Network (SHiNe), validated on an FPGA, has been found to achieve reasonable performance with a low resource utilization, with some trade-off with respect to hardware throughput and signal representation.