Scalable platform enabling reservoir computing with nanoporous oxide memristors for image recognition and time series prediction

Typical mammal brains have some form of random connectivity between neurons. Reservoir computing, a neural network approach, uses random weights within its processing layer along with built-in recurrent connections and short-term, fading memory, and is shown to be time and training efficient in processing spatiotemporal signals. Here we prepared a niobium oxide-based thin film memristor device with intrinsic structural in-homogeneity in the form of random nanopores and performed computational tasks of XOR operations, image recognition, and time series prediction and reconstruction. For the latter task we chose a complex three-dimensional chaotic Lorenz-63 time series. By applying three temporal voltage waveforms individually across the device and training the readout layer with electrical current signals from a three-output physical reservoir, we achieved satisfactory prediction and reconstruction accuracy in comparison to the case of no reservoir. This work highlights the potential for scalable, on-chip devices using all-oxide reservoir systems, paving the way for energy-efficient neuromorphic electronics dealing with time signals.

💡 Research Summary

This paper presents a scalable hardware platform for physical reservoir computing (RC) based on nanoporously structured niobium‑oxide (NbOₓ) memristors. The authors deliberately introduce random nanopores into a Pt bottom electrode, which, after annealing, yields a porous network with an average pore radius of ~29 nm. A nitrogen‑doped NbOₓ switching layer (≈80 nm) is deposited on top of this porous substrate, followed by a thin Ti adhesion layer and a conventional Pt top electrode. The resulting two‑terminal device exhibits volatile resistive switching: under a voltage sweep (0 → 3 V → 0 V and the negative counterpart) the resistance continuously moves from a high‑resistance state (HRS) to a low‑resistance state (LRS) and returns to HRS when the bias is removed, providing short‑term memory (hysteresis) and strong non‑linearity. Because the nanopores create a multitude of random conductive pathways, each of the four top electrodes shows a distinct I‑V characteristic; three of them are sufficiently different to serve as independent reservoir output channels, while the fourth is overly conductive and excluded from further experiments.

In the RC scheme, an input voltage waveform u(t) is applied across the device, and the three simultaneous current responses r(t) are recorded. The concatenated vector X =