Memristive Devices in Neuromorphic Computing

Memristive devices are two-terminal elements whose resistance state depends on the history of applied voltage or current, closely mimicking the activity-dependent plasticity of biological synapses. In neuromorphic computing architectures, these devices serve both as storage for synaptic weights and as active components in in-memory processing, thereby overcoming the energy and latency penalties of conventional von Neumann systems. Various material platforms, including transition metal oxides, phase change compounds and emerging two-dimensional frameworks, have been explored to achieve high endurance, low operating energy and multilevel conductance states. Crossbar arrays of memristors enable massively parallel matrix-vector multiplications, fundamental to both spiking and deep neural network implementations. Meanwhile, device innovations that exploit ionic drift, filamentary switching or ferroelectric polarisation switching support local learning rules such as spike-timing-dependent plasticity. Key challenges remain in controlling device variability, achieving linear weight updates, scaling to billions of elements and integrating with complementary metal–oxide–semiconductor circuitry. Nevertheless, recent advances portend energy-efficient, highly scalable platforms for real-time pattern recognition, temporal signal processing and edge AI applications.

Research from Nature Portfolio

Recent studies have demonstrated the potential of memristive arrays for temporal processing by harnessing intrinsic ionic dynamics. In one landmark experiment, a small ensemble of dynamic memristors formed a reservoir computing system capable of handwritten-digit recognition and nonlinear temporal task resolution without retraining internal weights. Another investigation achieved on-chip, multilayer learning by monolithically integrating hafnium-oxide memristors with transistor arrays; this work showcased in-situ training that adapts to device imperfections and attains classification accuracies comparable with software-based networks. In parallel, a multi-memristive synapse architecture based on phase change memory devices has been proposed, using a counter-based arbitration scheme to extend conductance range and precision. Simulations and large-scale experiments with over a million devices validated unsupervised temporal correlation learning in spiking networks, marking a significant step towards practical, large-scale neuromorphic hardware.

Memristive Devices in Neuromorphic Computing publication trend

The graph below shows the total number of articles in memristive devices in neuromorphic computing across all publications each year (not limited to Nature Index journals).

Technical terms

Memristor: A two-terminal electronic component whose resistance varies according to the history of voltage or current, enabling non-volatile weight storage and computation in neuromorphic circuits.

Synaptic plasticity: The ability of a synaptic connection to strengthen or weaken over time in response to electrical activity, modelled in hardware by conductance modulation of memristive devices.

Crossbar array: A grid of intersecting conductive lines with memristive elements at each junction, used to perform parallel matrix-vector multiplications directly within memory.

Reservoir computing: A neural network paradigm that employs a fixed, dynamic reservoir—such as a network of memristors—to project temporal inputs into a high-dimensional space, training only the readout layer.

In-memory computing: A processing approach in which computation is performed within the memory elements themselves, reducing data movement and improving energy efficiency.

References

1. Reservoir computing using dynamic memristors for temporal information processing. Nature Communications (2017).
2. Efficient and self-adaptive in-situ learning in multilayer memristor neural networks. Nature Communications (2018).
3. Neuromorphic computing with multi-memristive synapses. Nature Communications (2018).
4. Recent Advances in In-Memory Computing: Exploring Memristor and Memtransistor Arrays with 2D Materials. Nano-Micro Letters (2024).
5. High‐Performance Memristors Based on Ordered Imine‐Linked Two‐Dimensional Covalent Organic Frameworks for Neuromorphic Computing. Interdisciplinary Materials (2025).
6. Neuromorphic computing using non-volatile memory. Advances in Physics X (2016).
7. Learning through ferroelectric domain dynamics in solid-state synapses. Nature Communications (2017).