READ ME File For 'Dual-Mode Nanoporous SiO2 Memristors with Coexisting Volatile and Nonvolatile Dynamics for Reservoir Computing'

Dataset DOI: 10.5258/SOTON/D3940

ReadMe Author: Bohao Ding, University of Southampton 

This dataset supports the publication: Dual-Mode Nanoporous SiO2 Memristors with Coexisting Volatile and Nonvolatile Dynamics for Reservoir Computing
AUTHORS: Bohao Ding, Tongjun Zhang, Li Shao, Nikolay Zhelev, Andrew L. Hector, Ruomeng Huang
TITLE: Dual-Mode Nanoporous SiO2 Memristors with Coexisting Volatile and Nonvolatile Dynamics for Reservoir Computing
JOURNAL: Advanced Science
PAPER DOI IF KNOWN: doi.org/10.1002/advs.76163


This dataset contains:
The raw data of figure 1 to 8.

The figures are as follows:

Figure 1. Overall architecture and workflow of the unified nanoporous SiO2 memristor-based neuromorphic system for MNIST digit and ECG signal recognition.


Figure 2. Structural characterization of the 3D-ordered nanoporous silica (nSiO2) thin films. (a) Schematic illustration of the sol–gel fabrication process of nanoporous SiO2 thin films. (b) Top-view SEM image showing the uniform nanoporous architecture. (c) GISAXS pattern of the SiO2 thin film with simulated Bragg reflections confirming the periodic nanostructure.

Figure 3. Resistive switching characteristics of the nanoporous SiO2 memristor. (a) Top-view SEM image of the nanoporous SiO2 cross-point device (20 μm × 20 μm). (b) Forming-free volatile I–V characteristics from consecutive sweeps under a compliance current (CC) of 1 μA. (c) Endurance characteristics of the volatile mode at a 0.1 V read voltage. (d) Device-to-device volatile I–V characteristics. (e) Consecutive non-volatile bipolar I–V sweeps under a CC of 1 mA. (f) Endurance characteristics of the non-volatile switching mode under repeated DC sweep conditions measured at a read voltage of 0.1 V. (g) Endurance characteristics of the non-volatile switching mode under pulse-driven operation conditions. The pulse endurance measurements were performed using alternating ±3 V pulses with pulse width and interval time of 1 ms, while the resistance states were read at 0.1 V. (h) Retention characteristics of the non-volatile mode measured under a CC of 1 mA. (i) Device-to-device non-volatile I–V characteristics.

Figure 4. Short-term synaptic plasticity (STP) emulated by the nanoporous SiO2 memristor. (a) Schematic comparison between a biological synapse and the memristor, illustrating their analogous signal transmission and modulation mechanisms. (b) Progressive I–V evolution during 50 consecutive low-bias sweeps (0–1 V) without CC. (c) Corresponding gradual conductance potentiation extracted from panel (b), confirming analog-like conductance modulation. (d) Incremental increase of postsynaptic current (PSC) during 30 consecutive pulse stimulations (1 V amplitude, 10 ms width, 0.5 ms interval), followed by spontaneous relaxation after pulse cessation. (e) Frequency-dependent paired-pulse facilitation (PPF) responses recorded under paired stimuli (1 V, 10 ms) at 1, 2, 5, 10, 20, and 50 Hz. (f) Extracted PPF index as a function of inter-pulse interval (Δt), showing a double-exponential decay that characterizes the temporal dynamics of synaptic facilitation.

Figure 5. Stimulus-dependent short-term synaptic plasticity (STP) characteristics of the nanoporous SiO₂ memristor. (a) PSC responses to 30 consecutive voltage pulses (1 V amplitude, 0.5 ms gap) with pulse durations varied from 10 ms to 30 ms, showing larger conductance potentiation for longer pulses. (b) Experimental relaxation curves fitted using a stretched-exponential decay model, exhibiting excellent agreement. (c) Extracted relaxation time constants (τ) as a function of pulse duration. (d–f) Effect of inter-pulse gap (0.05, 0.5, and 5 ms) on PSC amplitude and relaxation dynamics, where longer gaps allow more ion diffusion and faster recovery. (g–i) Dependence of PSC response on pulse amplitude (1–3 V), demonstrating enhanced Ag-ion injection and prolonged decay with stronger electric fields.

Figure 6. Long-term synaptic plasticity and memory retention in the nanoporous SiO2 memristor. (a) Long-term potentiation (LTP) and long-term depression (LTD) triggered by sequential trains of positive and negative pulses, showing smooth, analog-like conductance modulation. (b) Multiple LTP/LTD cycles demonstrating stable and reversible conductance updates. (c) Retention characteristics of programmed conductance states measured over 600 s, confirming robust memory stability. (d) Cumulative probability distributions of the retained conductance states, exhibiting tightly clustered distributions and robust state stability.

Figure 7. Benchmark demonstration of MNIST handwritten-digit classification using a memristor-based reservoir computing (RC) system. (a) Schematic of the hardware RC architecture, where short-term plasticity-driven reservoir dynamics are coupled with a long-term plasticity-trained readout layer, both realized using cross-point-structured nanoporous SiO2 memristors. (b) Evolution of quantized-state loss during network training under mean-fitting and variability-aware conditions. (c) Training accuracy of the memristive reservoir-computing system as a function of training epoch under mean-fitting and variability-aware conditions. (d) Confusion matrix obtained under the mean-fitting condition for 10,000 MNIST test images, confirming accurate recognition across all digit categories.

Figure 8. ECG signal classification using the nanoporous SiO2 memristor-based RC system. (a) Schematic illustration of the RC framework, where STP-based virtual nodes form the reservoir and a LTP-trained readout layer performs classification. (b) Current response profiles of 34 virtual nodes for two heart conditions: normal (red solid lines) and abnormal (blue dashed lines). (c) Evolution of training loss and state loss (evaluated every 2 epochs) as a function of training epochs. (d) Training accuracy of the RC system during ECG classification as a function of training epochs. (e) Confusion matrix of the classification results for 266 ECG signals.    

Date of data collection: October 2024 to April 2026

Information about geographic location of data collection: United Kingdom

Licence: No

Related projects: ADEPT project funded by a Programme Grant from the EPSRC (EP/N035437/1). 


Date that the file was created: June, 2026