Two-dimensional materials have emerged as a promising platform for next-generation optoelectronic memory and neuromorphic computing devices. Their unique electronic, optical, and mechanical properties enable novel device architectures that bridge the gap between data storage and processing. Among these materials, transition metal dichalcogenides like molybdenum disulfide (MoS2) have demonstrated exceptional potential for resistive switching, photonic synapses, and in-memory computing applications.
Resistive switching in 2D materials forms the foundation of optoelectronic memory devices. The phenomenon relies on the reversible modulation of a material’s resistance under electrical or optical stimuli. In MoS2-based resistive random-access memory (RRAM), defects such as sulfur vacancies act as trapping sites for charge carriers, enabling the formation and rupture of conductive filaments. The high on/off ratios, often exceeding 10^5, and low switching voltages, typically below 1 V, make these devices attractive for non-volatile memory applications. The switching dynamics can be further tuned by controlling layer thickness, defect engineering, and interfacial properties.
Photonic synapses represent another breakthrough, where light pulses emulate synaptic plasticity in neural networks. Unlike purely electrical devices, optoelectronic synapses leverage the photoresponse of 2D materials to achieve synaptic weight modulation. MoS2 exhibits strong excitonic effects and tunable photoconductivity, allowing for precise control over synaptic behaviors such as short-term plasticity (STP) and long-term potentiation (LTP). For instance, pulsed laser illumination can induce persistent photoconductivity, mimicking the strengthening of synaptic connections. The ability to integrate optical and electrical stimuli enables more complex neuromorphic functionalities, including spike-timing-dependent plasticity (STDP), a critical mechanism for unsupervised learning in artificial neural networks.
In-memory computing architectures leverage 2D materials to overcome the von Neumann bottleneck by performing computations directly within memory arrays. Crossbar arrays of MoS2 memristors can execute vector-matrix multiplication in parallel, significantly accelerating machine learning tasks. The inherent scalability of 2D materials allows for ultra-dense integration, with device footprints potentially shrinking to the atomic scale. Additionally, their compatibility with flexible substrates opens new avenues for wearable and edge computing applications.
The role of MoS2 in mimicking neural plasticity stems from its dynamic response to external stimuli. Defect-mediated ion migration, charge trapping, and phase transitions contribute to analog switching behaviors that closely resemble biological synapses. For example, gradual resistance changes under repeated electrical pulses emulate the strengthening or weakening of synaptic weights, essential for adaptive learning. Heterostructures combining MoS2 with other 2D materials, such as hexagonal boron nitride (hBN), further enhance device performance by improving interface quality and reducing leakage currents.
Despite these advantages, challenges remain in endurance and variability. Cycle-to-cycle and device-to-device variations can arise from stochastic filament formation or environmental factors like humidity. Endurance is often limited by material degradation or irreversible defect generation, with typical lifetimes ranging from 10^4 to 10^6 cycles in MoS2-based devices. Strategies to mitigate these issues include encapsulation layers, doping, and optimized electrode materials. For instance, graphene electrodes have been shown to reduce interfacial reactions and improve switching uniformity.
Applications in AI hardware are particularly compelling. Neuromorphic devices based on 2D materials can enable energy-efficient deep learning accelerators, with power consumption potentially orders of magnitude lower than conventional CMOS-based systems. Prototypes have demonstrated pattern recognition, image classification, and even reinforcement learning tasks with high accuracy. The integration of photonic synapses also paves the way for optoelectronic neural networks, where light-based communication could further reduce latency and power consumption.
Looking ahead, the development of large-scale fabrication techniques will be critical for commercial adoption. Chemical vapor deposition (CVD) and transfer methods have advanced significantly, but achieving uniform, defect-free films over wafer-scale areas remains a challenge. Hybrid integration with silicon photonics and complementary metal-oxide-semiconductor (CMOS) technology could provide a practical pathway for near-term deployment.
In summary, 2D materials like MoS2 are redefining the landscape of optoelectronic memory and neuromorphic computing. Their unique properties enable resistive switching, photonic synapses, and in-memory computing architectures that closely emulate biological neural networks. While endurance and variability issues persist, ongoing material and device engineering efforts continue to push the boundaries of what is possible in AI hardware and beyond. The convergence of electronics and photonics in these systems promises to unlock unprecedented efficiencies and functionalities in the era of post-von Neumann computing.