Three-dimensional (3D) imaging captures spatial depth and multidimensional attributes, enabling precise scene reconstruction for diverse applications in robotics, augmented reality, precision medicine, and industrial processing. Fundamentally, 3D imaging relies on integrating front-end sensing devices with back-end computational algorithms. Recent algorithmic advances, particularly those leveraging artificial intelligence, impose stringent hardware demands, rendering traditional sensors such as charge-coupled device and complementary metal oxide semiconductor cameras insufficient for emerging applications. Driven by these novel paradigms, a new generation of front-end devices has emerged, including metasurfaces, neuromorphic cameras, and reconfigurable optoelectronic components, collectively overcoming limitations in speed, sensitivity, and resolution. These hardware innovations enhance multidimensional data acquisition, facilitating real-time processing, and robust depth extraction. This Mini-Review systematically introduces algorithm-driven front-end sensors, encompassing conventional, emerging optoelectronic, and photonic devices, highlighting their distinct performance advantages. Finally, we address the current manufacturing and integration limitations of these novel devices, providing perspectives on future opportunities.

Two-dimensional ferroelectric van der Waals heterojunctions offer high-density integration and superior performance in devices, facilitating efficient information storage and manipulation. However, their digital/analog computing architectures remain insufficiently explored. Here, we develop multi-gate ferroelectric field-effect transistors using the CuInP2S6/MoS2 heterostructure. The devices operate as programmable rectifiers, content-addressable memory units, and electrically programmable Boolean logic gates. By leveraging the multi-gate ferroelectric layer for efficient channel conductivity modulation, we achieve an on/off ratio of 106, a subthreshold swing of 45.1 mV/dec, and a drain current as low as 10−11 A, demonstrating device stability and low power. As a programmable rectifier, the devices exhibit a rectification ratio of 104. Their asymmetric polarization control at each channel end enables distinct switching behaviors and parallel data search. Furthermore, incorporating controlled-NOT (CNOT)-like logic into ferroelectric circuits enables adaptable, time-constrained weight-update kernels, achieving 96% accuracy in efficient handwritten digit recognition. This hybrid approach paves the way for multifunctional, electrically programmable devices with smaller footprints, low-power consumption, and high-density chip integration.

As the size of state-of-the-art copper interconnects shrinks to the nanoscale, carrier scattering greatly increases their resistance, causing signal delay and reliability issues. Alternative conductors are therefore needed to sustain interconnect scaling beyond conventional charge-transport mechanisms. Here we demonstrate CoSi semimetal for highly scalable, conductive and reliable interconnects. As the CoSi thickness decreases from 1 µm to ∼20 nm, its resistivity decreases from 7.0 to 0.72 μΩ·cm due to the highly conductive surface path. The room-temperature resistivity of 20-nm-thick CoSi is one-tenth that of copper at the same thickness. The high cohesive energy (5.4 eV) and migration barrier (3.7 eV) of CoSi confer excellent reliability at current densities of up to 108 A cm-2 and temperatures up to 450 °C. Radiofrequency measurements demonstrate CoSi interconnect operation at frequencies up to 40 GHz. We further integrate a CoSi interconnect with a 16-nm-node silicon ring oscillator, which operates at the same frequency as its metal-interconnected counterpart.

Neuromorphic computing aims to mimic the neural architecture of the biology to process information efficiently. Valley transistors, with their unique low-power information carrier properties, offer a promising approach for implementing bio-inspired computing systems. This study investigates the use of nonlocal valley transistors as graded neuron. After integrating the graded response properties in processing tempoeral signals, we have achieved high classification accuracy (95 %) of dynamic icons.

Bioinspired in-sensor computing devices can process information at sensory terminals by leveraging physical principles, thereby reducing latency and energy consumption during computation while simultaneously enhancing the efficiency of data processing and real-time analysis. Optoelectronic devices exhibit in-sensor computing functions, such as feature enhancement and data compression, by tuning the defect states of the semiconductor channels and thereby modulating the photoresponsivity and time constants of the sensors. These functionalities are critically dependent on precise fabrication and testing protocols. Here we present a detailed procedure for fabricating and characterizing in-sensor computing devices based on nanoscale semiconductor thin films. We explain how to test such optoelectronic devices, including the testing of visual adaptation and motion perception responses. When using semiconductor materials obtained from commercial suppliers, this procedure is time efficient and results in highly reproducible device performance. Nevertheless, all device fabrication and testing steps are generalizable and can be extended to other semiconductor thin films grown using different methods. The procedure is intended for researchers experienced in cleanroom operations and microfabrication techniques and can be completed in ~14 d. The use of bioinspired optoelectronic devices enables the development of a framework for advancing in-sensor computing technologies. Optoelectronic devices based on two-dimensional layered thin film semiconductors can be tuned to directly sense and process spatial and temporal information, making them suitable to use as in-sensor computing devices.

Schottky diodes are widely utilized in optoelectronics. In this study, we fabricated a vertical photodiode using a WSe2/ITO Schottky junction. The device exhibited an impressively low dark current of 0.7 pA and an ultrafast light response time of 13 ns, with a broadband photo-response across the entire visible spectrum. These exceptional results offer a promising approach for developing high-performance photodiodes in the future.

Sodium metal batteries (SMBs) are promising candidates for next-generation high-energy batteries due to the abundance and low cost of sodium. However, their practical deployment is hindered by uneven sodium deposition and dendritic growth, which leads to poor cycling stability and safety concerns. Herein, sodiophilic porous carbon hosts are fabricated via a defect engineering and surface functionalization strategy. The incorporation of moderate structural defects effectively reduces the sodium nucleation overpotential, while the oxygen-containing functional groups such as carboxyl (-COOH) and carbonyl (C=O) guides the formation of dense and uniform sodium layers. Theoretical calculations and experimental characterizations collectively reveal the synergistic role of defect sites and surface functionalities in promoting Na+ diffusion and regulating nucleation behavior. Consequently, the optimized ABC-11 achieves exceptional cycling stability with a Coulombic efficiency of 99.91 % over 1200 cycles at 1 mA cm-2/1 mAh cm-2 in asymmetric cells. The Na@ABC-11 symmetric cell also exhibits ultra-stable plating/stripping durability exceeding 2200 h. Furthermore, the assembled Na3V2(PO4)3 || Na@ABC11 full cell delivers a high reversible capacity of 102 mAh g-1 and maintains 94 % of its initial capacity after 1200 cycles. This work highlights a cost-effective and scalable strategy for the structural design of carbon-based hosts, offering a viable pathway toward high-performance SMBs.

Bioinspired vision systems are revolutionizing next-generation visual sensors through their unique neural mechanisms, offering great potential for power-efficient and real-time perception. This chapter explores the remarkable visual systems of insects and their transformative potential for advanced visual sensors. Specifically, we focus on the neural mechanisms of insect vision, particularly the adaptive signal processing by graded neurons and direction-selective circuits, we examine how these biological systems achieve unparalleled efficiency in motion detection, object tracking, and optic flow analysis. The chapter systematically bridges biological principles with engineering applications, detailing the design and implementation of bioinspired vision sensors that replicate these capabilities. Furthermore, we present cutting-edge developments in neuromorphic hardware and adaptive processing architectures while evaluating their performance against conventional systems. The discussion extends to industrial applications such as high-speed defect detection and unmanned aerial vehicle obstacle avoidance and concludes with the future directions and challenges in this interdisciplinary field. This chapter aims to provide researchers and engineers with neurobiology-inspired design principles and practical guidance for developing next-generation visual technologies.

Understanding electric double layers (EDLs) and electrochemicalprocesses represents significant challenges in electrocatalysis. Inthis study, we employed classic molecular dynamics (MD) and ab initio molecular dynamics (AIMD) simulationswith an explicit water solvent to investigate interfacial structureson the Fe-N-4-C catalyst and acquire dynamicobservation of the oxygen reduction reaction (ORR). The orientationand population of interfacial water are potential-dependent. Whenpotential shifts positively, water molecules evolve from structurally"two O-H down" to 'one O-H parallel,one O-H down,' "two O-H parallel,"and eventually to "two O-H up." Our finding alsosuggests that hydrogen bonds (denoted as H-bonds) vary depending onthe potential and follow an asymmetric M-shape pattern.It confirms that interfacial water with "two O-H parallel"structures maximizes the number of hydrogen bonds (H-bonds), whilemore water with 'one O-H down, one O-H up'suppresses H-bond formation. We provided detailed information on howthe electrode potential influences H-bonds by impacting the orientationsof interfacial water molecules. The above analysis of interfacialwater is completely general and could be applicable to any water-basedenergy conversion and storage systems. Then, we focused on the reactionprocess and local environments around the ORR reaction center. Oursimulations show that the proton transfer to oxygenous intermediatesdirectly occurs at the applied potential. We also demonstrate thatthe applied potential induces charge redistribution and water reorientationaround the reaction center. We further identified a quadratic functionrelationship between the reaction free energy/activation barrier andthe potential for the key elementary ORR steps, in which the hydrogenationof the oxygenous intermediate is less favorable at higher potentialsas the local water is stabilized and pulled away from the reactioncenter through the H-bond interaction. Our analysis provides a profoundunderstanding of electric double layers and electrochemical processes,which are critical to experimental exploration and electrocatalystapplication.

The sluggish kinetics of anodic oxygen evolution reaction (OER) largely impedes the energy conversion efficiency in electrolytic industries. Fundamentally, the OER process is governed by the chemical reactivity of oxygen intermediates. Here, we demonstrate an effective approach to regulate the oxygen reactivity through the double-exchange interaction (DEI) towards OER activity improvement, using Ti-substituted pyrochlore ruthenate Y2Ru2-xTixO7 as model catalyst. We unveil that DEI-induced electron hopping in asymmetric Ru-O-Ti backbone enables the charge depletion in Ru active center to tune on the oxygen activation by ligand hole introduction. In addition, the ferromagnetic nature of DEI constructs a spin-selected channel for charge transport to govern the spin state in oxygen intermediates. Such comprehensive manipulation of oxygen reactivity substantially facilitates the nucleophilic attack process during O-O coupling as rate-determining step, lowering the energy barrier to improve the intrinsic OER activity.

The growing computational demand in artificial intelligence calls for hardware solutions that are capable of in situ machine learning, where both training and inference are performed by edge computation. This not only requires extremely energy-efficient architecture (such as in-memory computing) but also memory hardware with tunable properties to simultaneously meet the demand for training and inference. Here we report a duplex device structure based on a ferroelectric field-effect transistor and an atomically thin MoS 2 channel, and realize a universal in-memory computing architecture for in situ learning. By exploiting the tunability of the ferroelectric energy landscape, the duplex building block demonstrates an overall excellent performance in endurance (>10 13 ), retention (>10 years), speed (4.8 ns) and energy consumption (22.7 fJ bit –1 μm –2 ). We implemented a hardware neural network using arrays of two-transistors-one-duplex ferroelectric field-effect transistor cells and achieved 99.86% accuracy in a nonlinear localization task with in situ trained weights. Simulations show that the proposed device architecture could achieve the same level of performance as a graphics processing unit under notably improved energy efficiency. Our device core can be combined with silicon circuitry through three-dimensional heterogeneous integration to give a hardware solution towards general edge intelligence.

Abstract The growing computational demand in artificial intelligence (AI) calls for hardware solutions that are capable of in-situ machine learning, where both training and inference are performed by edge computation. This not only requires extremely energy-efficient architecture (such as in-memory computing, IMC) but also memory hardware with tunable properties to simultaneously meet the demand for training and inference. Here, we report a duplex device structure based on ferroelectric field-effect transistor (FeFET) and atomically thin MoS2 channel and realize a universal IMC architecture for in-situ learning. By exploiting the tunability of ferroelectric energy landscape, the duplex building block demonstrates overall excellent performance in endurance (>1013), retention (>10 years), speed (4.8 ns) and energy consumption (22.7 fJ/(bit·μm2)). We implemented a hardware neural network using arrays of two-transistor-one-duplex-FeFET (2T1D) cells and achieved 99.86% accuracy in non-linear localization task with in-situ trained weights. Simulations show that the proposed device architecture could achieve the same level of performance as graphics processing unit under notably improved energy efficiency. Our device core can be combined with silicon circuitry through three-dimensional heterogeneous integration to give a hardware solution toward general edge intelligence (EI).

The prevailing transmission of image information over the Internet of Things demands trustworthy cryptography for high security and privacy. State-of-the-art security modules are usually physically separated from the sensory terminals that capture images, which unavoidably exposes image information to various attacks during the transmission process. Here we develop in-sensor cryptography that enables capturing images and producing security keys in the same hardware devices. The generated key inherently binds to the captured images, which gives rise to highly trustworthy cryptography. Using the intrinsic electronic and optoelectronic characteristics of the 256 molybdenum disulfide phototransistor array, we can harvest electronic and optoelectronic binary keys with a physically unclonable function and further upgrade them into multiple-state ternary and double-binary keys, exhibiting high uniformity, uniqueness, randomness, and coding capacity. This in-sensor cryptography enables highly trustworthy image encryption to avoid passive attacks and image authentication to prevent unauthorized editions.

Motion processing has proven to be a computational challenge and demands considerable computational resources. Contrast this with the fact that flying insects can agilely perceive real-world motion with their tiny vision system. Here we show that phototransistor arrays can directly perceive different types of motion at sensory terminals, emulating the non-spiking graded neurons of insect vision systems. The charge dynamics of the shallow trapping centres in MoS 2 phototransistors mimic the characteristics of graded neurons, showing an information transmission rate of 1,200 bit s −1 and effectively encoding temporal light information. We used a 20 × 20 photosensor array to detect trajectories in the visual field, allowing the efficient perception of the direction and vision saliency of moving objects and achieving 99.2% recognition accuracy with a four-layer neural network. By modulating the charge dynamics of the shallow trapping centres of MoS 2 , the sensor array can recognize motion with a temporal resolution ranging from 10 1 to 10 6 ms.

The electrocatalytic nitrogen reduction reaction (NRR) is one of the most promising ways to achieve NH3 production at room temperature and pressure. However, there exists significant disagreement between the theoretically predicted potentials required for the NRR by the conventional quantum-theoretical calculations and those observed experimentally. Here, an explicit computational model incorporating the solvation effect and electrode potential has been proposed for NRR on single iron atoms supported on nitrogen-doped graphene. We find that the aqueous environment plays an essential role in NRR by promoting N-2 adsorption, whereas the electrode potential impacts considerably on the electrode-electrolyte interface where NRR occurs. The constrained molecular dynamics (cMD) simulations and a thermodynamic integration method are used to explore the free energy profiles of N-2 adsorption and the proton transfer process. The results are consistent with experimental observations, i.e., the NRR can take place at a relatively low electrode potential, thus revealing the critical role of the explicit inclusion of the solvation effect and electrode potential in computationally studying electrochemical reactions. With this approach, we have provided atomic-level mechanistic insights into the electrode-electrolyte interface for NRR through electrochemical catalysis.

Achieving the fundamental understanding of electrochemical processes occurring at the complex electrode-liquid interface is a grand challenge in catalysis. Herein, to gain theoretical insights into the experimentally observed potential-dependent activity and selectivity for CO2 reduction reaction (CO2RR) on the popular single-iron-atom catalyst, we performed ab initio molecular dynamics (AIMD) simulation, constrained MD sampling and the thermodynamic integration to acquire the free energy profiles for the proton and electron transfer processes of CO2 at different potentials. We have demonstrated that the adsorption of CO2 is significantly coupled with the electron transfer from the substrate while the further protonation does not show distinct charge variation. This strongly suggest that CO2 adsorption is potential-dependent and optimizing the electrode potential is vital to achieve the efficient activated adsorption of CO2. We further identified a linear scaling relationship between the reaction free energy (ΔG) and the potential for key elementary steps of CO2RR and HER, of which the slope is adsorbate-specific and not as simple as 1 eV per Volt as suggested by the traditional Computational Hydrogen Electrode (CHE) model. The derived scaling relationship can reproduce the experimental onset potential (Uonset) of CO2RR, potential of the maximal CO2-to-CO Faraday Efficiency (FECO), and the potential where FECO = FEH2. This suggests that our state-of-the-art model could precisely interpret the activity and selectivity of CO2RR/HER on Fe-N4-C catalyst under different electrode potentials. In general, our study not only provides an innovative insight into the theoretical explanation of the origin of solvation effect from the perspective of charge transfer but also emphasizes the critical role of electrode potential on theoretical consideration of catalytic activity, which offers a profound understanding of the electrochemical environment and bridges the gap between theoretical predictions and experiment results.

Developing single-atom photocatalysts for selective conversion of CO2 to valuable fuel is of great attraction but remains challenging. In this work, ruthenium and copper single atoms are for the first time simultaneously incorporated into polymeric carbon nitride (PCN) through a simple preassembly-coprecipitation-pyrolysis process. The obtained PCN-RuCu sample exhibited much higher selectivity (95%) for CH4 production than the individual Ru or Cu decorated PCN during photocatalytic CO2 reduction under visible-light irradiation. The atomically dispersed Ru-N4 and Cu-N3 moieties were confirmed by spherical aberration-corrected electron microscopy and extended X-ray absorption fine structure spectroscopy. Density function theory (DFT) calculations revealed that the co-existence of Ru-N4 sites and Cu-N3 sites can effectively tune the electronic structure of PCN, making the Ru sites account for photogenerated electron-hole pairs and the Cu sites for CO2 hydrogenation. Moreover, the synergetic effect between Ru and Cu single atoms significantly promotes the consecutive hydrogenation processes of *CO species towards CH4 production. Our studies provide a new understanding of the mechanism for photocatalytic reduction of CO2 to CH4, and pave a new way to design photocatalysts for the selective production of solar fuels.

Modern electronics demand transistors with extremely high performance and energy efficiency. Charge-based transistors with conventional semiconductors experience substantial heat dissipation because of carrier scattering. Here, we demonstrate low-loss topological phase change transistors (TPCTs) based on tellurium, a Weyl semiconductor. By modulating the energy separation between the Fermi level and the Weyl point of tellurium through electrostatic gate modulation, the device exhibits topological phase change between Weyl (Chern number ≠ 0) and conventional (Chern number = 0) semiconductors. In the Weyl ON state, the device has low-loss transport characteristics due to the global topology of gauge fields against external perturbations; the OFF state exhibits trivial charge transport in the conventional phase by moving the Fermi level into the bandgap. The TPCTs show a high ON/OFF ratio (108) at low operation voltage (≤2 volts) and high ON-state conductance (39 mS/μm). Our studies provide alternative strategies for realizing ultralow power electronics.

Acquiring the fundamental understanding ofelectrochemical processes occurring at the complex electrode-liquid interface is a grand challenge in catalysis. Herein, to gaintheoretical insights into the experimentally observed potential-dependent activity and selectivity for the CO2reduction reaction(CO2RR) on the popular single-iron-atom catalyst, we performedab initio molecular dynamics (AIMD) simulation, constrained MDsampling, and thermodynamic integration to acquire the freeenergy profiles for the proton and electron transfer processes ofCO2at different potentials. We have demonstrated that theadsorption of CO2is significantly coupled with the electrontransfer from the substrate while the further protonation does notshow distinct charge variation. This strongly suggests that CO2adsorption is potential-dependent and optimizing the electrodepotential is vital to achieve the efficient activated adsorption of CO2. We further identified a linear scaling relationship between thereaction free energy (Delta G) and the potential for key elementary steps of CO2RR and HER, of which the slope is adsorbate-specificand not as simple as 1 eV per volt as suggested by the traditional computational hydrogen electrode (CHE) model. The derivedscaling relationship can reproduce the experimental onset potential (Uonset)ofCO2RR, potential of the maximal CO2-to-CO Faradayefficiency (FECO), and potential where FECO=FEH2. This suggests that our state-of-the-art model could precisely interpret theactivity and selectivity of CO2RR/HER on the Fe-N4-C catalyst under different electrode potentials. In general, our study not onlyprovides an innovative insight into the theoretical explanation of the origin of the solvation effect from the perspective of chargetransfer but also emphasizes the critical role of electrode potential in the theoretical consideration of catalytic activity, which offers aprofound understanding of the electrochemical environment and bridges the gap between theoretical predictions and experimentalresults