Brain-inspired memristors open new pathways to energy-efficient artificial intelligence

As artificial intelligence continues to advance, conventional computing architectures face growing challenges in energy consumption, data-transfer latency, and hardware scalability. The separation of memory and processing units in traditional von Neumann systems creates a significant bottleneck, particularly for data-intensive AI applications. Neuromorphic computing, inspired by the parallel and energy-efficient information processing of the human brain, offers a promising alternative. In a comprehensive review, researchers from Shandong University, Fudan University, and collaborating institutions examine the development of neuromorphic memristors, highlighting how materials, device architectures, and system integration can be tailored to meet the demands of next-generation artificial intelligence.

Why Neuromorphic Memristors Matter

Biological neurons and synapses process information through electrical signals and dynamically adjustable connections. Neuromorphic memristors mimic these functions by regulating and retaining electrical conductance states, enabling artificial synaptic plasticity and neuron-like behavior. Their compact structures, nonvolatile memory characteristics, and potential for dense integration make them attractive for in-memory computing, where data storage and processing can occur within the same hardware. By reducing data movement between separate memory and processing units, these devices offer opportunities to improve computational efficiency and reduce power consumption.

Innovative Design and Mechanism

The review systematically examines representative memristor platforms, including oxide-based materials, ferroelectric systems, polymers, and two-dimensional materials. It also explores different device architectures and switching mechanisms, such as conductive filament formation, ionic migration, polarization modulation, phase transitions, and charge storage. Importantly, the authors emphasize that device optimization should be guided by the intended computational task rather than conventional switching performance alone. Different platforms offer distinct advantages: ferroelectric oxides provide opportunities for CMOS-compatible integration, polymer systems support flexible electronics, and layered materials enable ion-controlled neuromorphic functions. Reconfigurable devices capable of switching between volatile and nonvolatile operating modes further expand the range of possible computing applications.

Outstanding Performance

Recent demonstrations highlighted in the review illustrate the practical potential of these technologies. A multifunctional memristive synapse achieves approximately two orders of magnitude lower energy consumption than a GPU-based computing solution. Memristor arrays have also enabled epilepsy-related neural signal analysis with 93.46% accuracy and nearly 400-fold higher power efficiency than state-of-the-art CMOS systems. Other examples include artificial neurons operating at voltages as low as 80 mV, flexible three-dimensional memristor arrays, and wearable textile networks capable of combining sensing and neuromorphic processing. Together, these advances demonstrate the potential of memristors for efficient, adaptive, and mechanically flexible intelligent hardware.

Applications and Future Outlook

The review connects neuromorphic memristor technologies with a broad range of AI applications, including medical signal recognition, image reconstruction, artificial vision, intelligent sensing, and gesture-based human–machine interaction. Nevertheless, challenges remain in manufacturing reproducibility, device variability, standardized performance evaluation, scalable fabrication, and system-level integration. The authors highlight the need to coordinate materials engineering, CMOS-compatible architectures, peripheral circuits, and software–hardware co-design. Ultimately, integrating sensing, memory, computing, and interaction into compact hardware platforms could bring energy-efficient neuromorphic intelligence closer to practical applications in wearable electronics, edge AI, and intelligent human–machine interfaces.