The rapid advancement of artificial intelligence has driven the need for energy-efficient and high-performance computing architectures. Traditional von Neumann computing, which separates memory and processing units, faces significant limitations due to the “von Neumann bottleneck,” where data transfer between components consumes substantial time and power. Inspired by the human brain’s ultra-efficient, parallel, and event-driven computation, neuromorphic computing offers a promising alternative. At its core are artificial synapses—key components that mimic biological synaptic plasticity through adjustable conductance states. Among various candidates, memristors have emerged as particularly suitable due to their non-volatility, low power consumption, and scalability. Oxide-based memristors, in particular, benefit from compatibility with complementary metal-oxide-semiconductor (CMOS) fabrication processes, making them ideal for integration into existing electronic systems.
However, conventional oxide memristors suffer from performance instability caused by stochastic formation and rupture of nanoscale conducting filaments—typically composed of oxygen or metal vacancies. This randomness leads to significant device-to-device and cycle-to-cycle variations, hindering large-scale implementation in neural networks. To overcome these challenges, hybrid oxide structures have been developed, combining multiple materials or introducing dopants to enhance control over ion migration and stabilize resistive switching behavior. These strategies enable more reliable, uniform, and reproducible synaptic operations essential for practical neuromorphic applications.
This review focuses on recent advances in hybrid oxide memristive synapses, categorized by their structural design and underlying physical mechanisms. The discussion begins with multilayered oxide configurations, where distinct layers govern different aspects of switching. For example, bilayers such as HfOx/TaOy or TiOx/AlOy improve stability by confining filament formation at interfacial regions, reducing stochasticity. The introduction of conductive layers like Sn-doped In2O3 (ITO) acts as an oxygen reservoir, enhancing linearity and long-term reliability. Similarly, Ar plasma treatment modifies surface chemistry to promote homogeneous oxygen vacancy distribution, leading to consistent resistance modulation.
Another class involves hybrid structures incorporating non-oxide materials. MoOx/MoS2 and WOx/WS2 devices fabricated via solution processing exhibit repeatable bipolar switching with high ON/OFF ratios and multilevel capability, attributed to field-driven oxygen ion migration at heterojunctions. Graphene quantum dots (GQDs) embedded in FeOx matrices serve as nano-oxygen reservoirs, enabling precise manipulation of oxygen vacancy dynamics and achieving narrowly distributed switching voltages. Moisture-powered memristors based on WOx/oxygen-plasma-treated amorphous carbon (OAC) further expand functionality by allowing human breath-driven reading through reversible oxygen ion migration across interfaces.
Element-doped oxides represent another effective approach. Nitrogen doping in TiOx or ZnO strengthens N–O bonds, effectively trapping oxygen ions during SET processes and enabling controlled filament rupture during RESET. Mn-, Al-, or Y-doped HfOx suppress crystallization, preserving amorphous phases crucial for uniform switching. Si-doped HfOx devices have been successfully integrated into spiking neural networks (SNNs), enabling time-dependent synaptic weight modulation through transistor-coupled circuits. Such systems demonstrate biorealistic learning via spike-timing-dependent plasticity (STDP), capable of detecting sound location using interaural time differences—a key function in auditory perception.
Composite oxides, such as HfAlOx and HfZrOx, offer enhanced performance through tailored composition and microstructure. Flexible 3D crossbar arrays made from HfAlOx membranes on polyethylene terephthalate substrates achieve ultralow energy consumption (~4 aJ/spike), surpassing even biological systems. These devices support multilevel information transmission and retain functionality under bending, demonstrating excellent mechanical robustness. Furthermore, integrating TiN buffer layers and electro-thermal modulation improves cell-to-cell uniformity and enables implementation of associative memory functions via Hopfield neural networks, capable of recovering emotional images from partial inputs.p300 Antibody Autophagy
Finally, purely electronic memristors based on carrier trapping/detrapping mechanisms avoid microstructural changes during operation, offering superior stability.MMACHC Antibody MedChemExpress Structures like HfO2/Ta2O5 show self-rectifying behavior, eliminating sneak paths in crossbar arrays without additional selector devices.PMID:34856454 This intrinsic rectification simplifies circuit design and enhances scalability.
In conclusion, hybrid oxide memristive synapses represent a transformative step toward practical neuromorphic computing. By leveraging multilayer engineering, material hybridization, elemental doping, and composite design, researchers have significantly improved device stability, uniformity, and functionality. These advances bring us closer to realizing brain-inspired machines that match the human brain’s energy efficiency and computational power. Future efforts should focus on optimizing endurance, retention, and multi-level precision while enabling large-scale integration. With continued innovation, oxide-based hybrid memristors hold strong promise for the hardware realization of next-generation artificial intelligence.MedChemExpress (MCE) offers a wide range of high-quality research chemicals and biochemicals (novel life-science reagents, reference compounds and natural compounds) for scientific use. We have professionally experienced and friendly staff to meet your needs. We are a competent and trustworthy partner for your research and scientific projects.Related websites: https://www.medchemexpress.com