Protonic Nickelates: H-Modified Perovskites

Updated 3 January 2026

Protonic nickelates are perovskite materials, such as NdNiO₃, that integrate interstitial protons to enable tunable electronic properties and robust metal–insulator transitions.
They are synthesized using reactive co-sputtering and controlled hydrogenation that modulate lattice expansion and resistive switching through precise proton incorporation.
These materials underpin neuromorphic and memristive devices by supporting rapid, low-energy, multilevel switching in CMOS-compatible architectures.

Protonic nickelates are a class of perovskite nickelate materials, exemplified by NdNiO₃, that are chemically and electronically modified by the controlled incorporation of interstitial protons (H⁺) into their crystalline lattice. This process, typically referred to as hydrogenation, enables tunable modulation of local conductivity, strong resistive switching, and nonlinear transport properties. Such features have positioned protonic nickelates as promising candidates for correlated oxide-based neuromorphic hardware and memristive device applications, integrating ultrafast and multilevel electronic functionality in a CMOS-compatible material system (Zhou et al., 27 Dec 2025, Gamage et al., 2023).

1. Material Synthesis, Proton Incorporation, and Structural Effects

Hydrogenated NdNiO₃ (H-NNO) thin films are commonly grown by reactive co-sputtering of neodymium and nickel targets onto lattice-matched perovskite or silicon substrates. A representative synthesis consists of depositing a 50–60 nm film on LaAlO₃ (100) at 5 mTorr with an Ar/O₂ ratio of 40/10 sccm, followed by post-annealing at 550 °C in O₂ for 24 h to optimize crystallinity (Zhou et al., 27 Dec 2025, Gamage et al., 2023). Proton incorporation is achieved by annealing the device in forming gas (5% H₂ in Ar or N₂) at 115–120 °C for 20 min, using patterned Pd electrodes as catalytic sources for H₂ dissociation and H⁺ injection.

X-ray diffraction reveals a systematic shift and broadening of characteristic film peaks (e.g., NNO (002)), evidencing lattice expansion (Δc) and increased structural disorder linked to interstitial H. The typical proton content reaches $x \sim 0.3–0.5$ in $H_x$ NdNiO₃ under these conditions, sufficient to significantly reduce the Ni oxidation state (from 3+ towards 2+) and drive a robust metal–insulator transition. The volumetric expansion of NiO₆ octahedra with increasing hydrogenation is quantitatively described by DFT calculations. The octahedral volume increases by 6.7% at $x_H=0.25$ , 12.6% at $x_H=0.50$ , and up to 21.2% at full protonation ( $x_H=1.0$ ) (Gamage et al., 2023). Strain fields generated by volumetric expansion create an internal energetic barrier (ΔE_strain ∼ 0.2–0.4 eV) that spatially regulates further proton ingress.

2. Device Architectures and Fabrication Methodologies

Protonic nickelate devices are fabricated in a variety of lateral junction geometries, primarily Pd–Pd (symmetric) or Pd–Au (asymmetric) pad configurations. Symmetric Pd–Pd devices form “hydrogen clouds” under each electrode, with NNO channel lengths in the 2–6 μm range (Gamage et al., 2023, Zhou et al., 27 Dec 2025). Asymmetric Pd–Au devices establish a spatially graded hydrogen profile, facilitating distinct local switching characteristics.

The fabrication flow involves optical or electron-beam lithographic definition of electrode pads (typically 100–120 μm per side), with Ti/Au capping for improved contact. Following hydrogenation, array architectures with 2×3 to 8×16 nodes are realized, with inter-pad pitches near 130 μm. The equivalent circuit at each node incorporates a voltage-controlled resistive element $R_h(x)$ (modulated by local H-concentration), and a parasitic substrate capacitance $C_s$ . In Pd–Pd devices, two series-coupled $R_h(x)$ elements encode the volatile, anti-polar switching response.

3. Proton Transport, Nonlinear Switching, and Nanoscale Inhomogeneity

The key mechanism underlying protonic nickelate devices is the migration of interstitial H⁺ under applied electric fields. This migration is described by the Nernst–Planck drift–diffusion expression,

$J = -D \nabla c_H + \mu c_H E$

where $c_H$ is the spatial H⁺ concentration, $D$ the proton diffusion coefficient (typically $10^{-12}$ – $10^{-13}$  m²/s at 120 °C), and $\mu$ the mobility. Interstitial hydrogen both donates an electron (reducing Ni³⁺ to Ni²⁺) and gives rise to local OH groups, as confirmed by density functional theory (DFT) and vibrational spectroscopy (Gamage et al., 2023).

Under voltage pulsing (e.g., $V_p≈5$  V, 500 ns duration), resistive switching proceeds via fast ( $\tau_C ∼ 500$  ns) capacitive and slower ( $\tau_H ∼ 5$  μs) proton redistribution. Asymmetric junctions (Pd–Au) exhibit nonvolatile multilevel switching, with 16 linearly spaced resistance states (40 kΩ to 350 kΩ) and an ON/OFF ratio of $>10^3$ ; symmetric Pd–Pd devices yield volatile, accumulative responses suitable for temporal memory (Zhou et al., 27 Dec 2025).

Infrared nanoimaging and s-SNOM/FTIR spectroscopy reveal stripe-like nanoscale phase patterns, where metallic (M) and insulating (I) domains with periods 0.8–1.2 μm coexist perpendicularly to the field. These stripes, driven by proton accumulation and strain, yield resistivity contrasts between $10^2–10^6$  Ω·μm, segmenting device conduction into parallel resistor networks (Gamage et al., 2023).

4. Electronic, Optical, and Memory Characteristics

Hydrogenated NdNiO₃ junctions function as memristive devices, displaying nonlinear I–V curves and programmable resistance under voltage biasing. In Pd–Au configurations, large resistance hysteresis and fine multilevel state control are achieved by sequential pulsing ( $\pm$ 9 V, 10 μs pulses), with state retention stable over tens of minutes and robust switching endurance (Gamage et al., 2023, Zhou et al., 27 Dec 2025). Resistance states relax exponentially ( $I(t) ≃ (I_{peak} – I_{ss})\,e^{-t/\tau_H} + I_{ss}$ , $\tau_H\sim5$  μs) after excitation, with frequency-dependent accumulative effects for short ( $\sim$ 500 ns) interpulse intervals.

The underlying carrier dynamics are further corroborated by local conductivity maps derived from s-SNOM amplitude and vibrational signatures corresponding to Ni–O octahedral and OH stretching modes (ω₀ ≈ 600 cm⁻¹, ω₀ ≈ 1064 cm⁻¹, respectively). DFT and molecular dynamics simulate the dielectric response across varying H-concentrations, reproducing the observed growth and spatial evolution of the OH peak in IR spectra (Gamage et al., 2023).

Switching energy per event is extremely low ($\sim$0.2 nJ), with sub-microsecond operation windows. Resistivity changes span $\rho_{NNO} ≈ 2.5×10^6$  Ω·m for pristine films to $\rho_{H-NNO} ≈ 8.85$  Ω·m after maximal protonation (Zhou et al., 27 Dec 2025).

5. Neuromorphic Implementations and Device Networks

Protonic nickelate devices form the basis for spatiotemporal neuromorphic networks that emulate dynamic biological neural computation. A canonical implementation employs a two-layer architecture: (1) a reservoir of 128 volatile (short-term) Pd–Pd junctions enabling emergent dynamic connectivity, and (2) a linear classifier layer using static (long-term) Pd–Au output weights.

The spatiotemporal reservoir leverages both local temporal memory (given by proton drift/relaxation) and global spatial modulation. Activation (spiking) of a node invokes proton-driven conduction changes, while also shifting the substrate potential, thereby modulating all other nodes via long-range, distance-independent coupling. Experiments demonstrate that active neighboring nodes can increase a node’s current output ( $I_{\text{out}}$ ) by up to 30% (Zhou et al., 27 Dec 2025).

Demonstrated machine learning tasks include: spoken-digit recognition using AudioMNIST (64-channel cochlea input into Pd–Pd reservoir; 95.3% accuracy with only 4 temporal spikes) and early seizure detection on EEG (23–46 input channels; 85% accuracy in a 1 s window). Performance exceeds temporal-only and no-processing baselines (improvements of up to 37% absolute detection rate), with an overall computation time below 10 μs per data frame and $>10^3\times$ energy advantage compared to slow ionic memory systems.

6. Nanoscale Imaging, Theoretical Characterization, and Fundamental Limitations

Operando IR nanoimaging (s-SNOM/nano-FTIR) and in situ x-ray nanodiffraction provide direct spatial and spectral diagnosis of proton, strain, and phase inhomogeneity. Local maxima and minima of the AFM topography coincide with domain boundaries, corresponding to up to 6% lattice expansion over 100 nm thickness, and local strain as high as +1% beneath Pd (fully hydrogenated) electrodes (Gamage et al., 2023).

Theoretical modeling using DFT (U(Ni 3d)=4.6 eV, J=0.6 eV) establishes the H interstitial formation energy $E_{form} ≈ -0.4$  eV, and a H⁺ hopping activation barrier $E_a ≈ 0.3–0.5$  eV. In regions with large fractional $x_H$ , strain further increases this barrier by ∼0.1 eV, suggesting a strain-moderated limit for device speed and write endurance at high proton densities.

7. CMOS Compatibility, Scalability, and Prospects

Hydrogenated NdNiO₃ films grown on SiO₂/Si, in addition to LaAlO₃, exhibit analogous protonic resistive switching, indicating CMOS process compatibility. Standard sputtering, optical lithography, lift-off patterning, and low-temperature hydrogenation (≤120 °C) integrate natively with back-end-of-line process flows (Zhou et al., 27 Dec 2025). Pad and device scaling to submicron dimensions enhances integration density, increases the speed of spatial coupling, and reduces device area.

Uniformity of resistance states across arrays achieves R-state dispersions under 5%. Scalability constraints include the management of parasitic capacitance and line resistance, which are addressable through three-dimensional stacking and monolithic CMOS driver integration. A plausible implication is that future architectures could combine protonic nickelate arrays with other functional oxide or 2D material layers, further expanding the domain of in-material machine intelligence.

References:

(Zhou et al., 27 Dec 2025) "Protonic Nickelate Device Networks for Spatiotemporal Neuromorphic Computing" (Gamage et al., 2023) "Infrared Nanoimaging of Hydrogenated Perovskite Nickelate Synaptic Devices"