Mechanistic pathways linking gut microbial metabolites, microbial structural products, and host-microbe co-metabolites to mitochondrial function.

As artificial intelligence (AI) and data-centric computing tasks become increasingly complex, the inherent energy inefficiency and latency of conventional von Neumann architectures have emerged as critical bottlenecks [–]. In response, brain-inspired neuromorphic computing has been extensively investigated as a promising paradigm to process vast amounts of unstructured data with unprecedented energy efficiency [–]. To realize such hardware-based artificial neural networks (ANNs), memristors have garnered significant attention as ideal artificial synapses due to their non-volatile analog resistance modulation, high integration density, and low power consumption [–]. For neuromorphic hardware to achieve high pattern recognition accuracy comparable to software-based algorithms, the artificial synapses must exhibit highly linear and symmetric weight updates, a large dynamic range (high ON/OFF ratio), and excellent operational reliability. Recently, two-dimensional (2D) transition metal dichalcogenides (TMDs) materials and their van der Waals (vdW) heterostructures have provided a breakthrough platform for next-generation flexible neuromorphic electronics [–]. A variety of TMD-based memristors have been investigated for in-memory computing, including floating-gate memristors (FGMEMs) [–], resistive memristors (RMEMs) [, ], phase-change memristors (PCMEMs) [, ], magnetic memristors (MMEMs) [, ], and ferroelectric field-effect transistors (FeFETs) [, ]. Among them, RMEMs using metallic niobium diselenide (NbSe) offer unique advantages. By precisely controlling the thermal oxidation of NbSe, a native NbO insulating layer can be formed on its surface, spontaneously creating an atomically flat, vertical NbO–NbSe heterostructure [, ]. In this architecture, the metallic NbSe acts as a seamless bottom electrode, while the NbO layer serves as a highly reliable active switching medium. Despite these structural advantages, realizing high-performance memristors based on ultra-thin 2D native oxides faces a severe technological hurdle regarding top electrode integration [–]. In conventional NbO–NbSe memristors, top metal electrodes are deposited using high-vacuum thermal evaporation. However, during this process, high-kinetic-energy metal atoms physically bombard the delicate surface of the ultra-thin oxide barrier. This high-energy metallization inevitably causes structural damage, atomic penetration, and the formation of defect-induced leakage pathways [, ]. In ultra-thin NbO layers, such interface degradation shifts the charge transport mechanism toward uncontrollable direct tunneling (DT). Consequently, the device suffers from a degraded ON/OFF ratio, high OFF-state leakage current, and excessively abrupt switching behaviors. In the context of neuromorphic computing, this severely degrades the dynamic range and the linearity of synaptic weight updates, which ultimately causes catastrophic failures in neural network training and classification accuracy.