Sungjoo Lee

Stochastic neural computation is essential for probabilistic inference and brain‐inspired information processing. However, most hardware stochastic neurons generate statistically independent spikes and fail to reproduce the temporal correlations observed in biological neurons. We demonstrate an integrated hardware synapse‐neuron consisting of a FeFET synapse and a CulnP ₂ S ₆ neuron, where the neuron exhibits temporally correlated stochastic spiking. The spike probability is continuously tuned by synaptic weight through voltage modulation, enabling hardware stochastic neural computation. Electrical measurements reveal voltage‐dependent stochastic firing accompanied by leaky‐integration‐firing dynamics. Structural and polarization analyses reveal differences in Cu‐related atomic contrast and partially retained polarization states, suggesting a history‐dependent internal state. Autocorrelation and inter‐spike interval analyses reveal temporal dependencies between successive stochastic spikes, which are consistent with the evolution of this retained state, with the strongest correlation emerging in the intermediate stochastic regime. The proposed stochastic spiking neural network achieves an inference accuracy of approximately 90% and maintains over 87% accuracy under severe noise conditions, demonstrating the computational advantages of temporally correlated stochastic spiking for robust probabilistic neural computation. This work demonstrates a class of 2D ferroelectric neuromorphic hardware that inherently incorporates intrinsic temporal dynamics, providing a pathway toward next‐generation neuromorphic and probabilistic computing.

The rapid growth of data-intensive applications demands efficient data processing with reduced data movement. Logic-in-memory (LIM) and in-memory search, exemplified by ternary content-addressable memory (TCAM), offer promising solutions; however, their typical implementation in separate device modules leads to increased circuit complexity and limited cell-level functional density. Here, we demonstrate a split-gate van der Waals ferroelectric field-effect transistor (vdW FeFET) as a unified operational cell for multimode reconfigurable logic and ternary in-memory search. The architecture integrates two parallel split gates, employing MoS2 as the channel and ferroelectric CuInP2S6 as the gate dielectric. Polarity-dependent modulation by the split gates and the drain bias enables multimode reconfigurable logic operation, supporting six Boolean functions, including directional data transfer. The same device can be programmed for associative memory equivalent to a one-transistor-per-cell TCAM, where two spatially separated ferroelectric regions are programmed within one continuous MoS2 channel, allowing ternary states to be stored and match and mismatch responses to be encoded directly into the channel current. Device-calibrated statistical simulations confirm ranking-based row discrimination under realistic device variability. By linking reconfigurable logic, spatial ferroelectric programming, and current-domain associative search within a common split-gate device framework, this work presents a compact transistor platform for logic-search-coupled in-memory computing.

Organic mixed ionic–electronic conductors (OMIECs) combine facile volumetric ion transport with robust electronic conduction, yet enhancing ionic accessibility through polar side chains often forces polar moieties into the ordered domains that govern charge transport, largely broadening the energetic landscape and eroding carrier mobility. Cross-linking with polar networks can improve ion transport without altering conjugated backbones, but how it reshapes the ionic–electronic balance has remained unresolved. Here, we address this question using length-modulated nitrene-induced photo-cross-linkers (n-NIPS) in indacenodithiophene–benzothiadiazole (IDTBT), chosen as a high-mobility yet ionically nonconductive representative. We found that phase-selective partitioning confines the n-NIPS network to the amorphous free volume, providing predefined, length-tunable ionic pathways while preserving the ordered aggregates for charge pathways, confirmed by the preserved ultralow energetic disorder of IDTBT. Transient organic electrochemical transistor measurements, analyzed through a Chapman–Richards growth framework, resolve the effect into separable capacitive and kinetic contributions, revealing that cross-linking collapses the initial ion-injection barrier once the cross-linker exceeds the anion size and isolates electrical conduction from the progressive degradation seen in pristine films. Cyclic spectroelectrochemistry attributes this protection to mechanical suppression of microstructural rearrangement, which suppresses irreversible interchain disruption and widens the reversible doping window. This work provides a mechanistic design principle of disorder-free OMIECs applicable to neuromorphic and bioelectronic devices.

Integrating diverse optical functionalities, including ultraviolet‐visible multispectral and polarization‐resolved sensing, offers a pathway toward neuromorphic visual systems that operate beyond human‐centric limits. Conventional machine vision systems lack access to additional optical dimensions, such as ultraviolet and polarization, which encode critical information for robust perception in complex environments. However, integrating these functionalities into a single device remains challenging. Here, a WS ₂ / α ‐In ₂ Se ₃ ferroelectric van der Waals heterostructure is developed that enables simultaneous multispectral (365–660 nm) and polarization‐resolved sensing, exhibiting a high dichroic ratio of ∼9.7. The device demonstrates synaptic behaviors, including paired‐pulse facilitation of up to ∼142%, and short‐term‐to‐long‐term memory transition. These anisotropic in‐memory synaptic characteristics arise from the interplay between crystal anisotropy, ferroelectric polarization dynamics, and light‐induced interfacial carrier modulation. Photo‐induced memory current‐based real‐time multispectral and polarization‐resolved imaging enables spatial reconstruction across both spectral and polarization domains. Furthermore, a neural‐network‐based RGB+UV multimodal inference framework achieves 93% recognition accuracy, while a polarization‐encoded, synaptic‐weight‐driven edge extraction approach reaches 97% accuracy. These results establish a device‐level platform for multidimensional optoelectronic neuromorphic vision, offering new opportunities for advanced intelligent sensing systems.

Neuromorphic computing requires memory elements that can emulate both nonvolatile and volatile operation. However, achieving reconfigurability without trade-offs in performance or device fatigue remains a critical challenge. Conventional electrically driven memristors face difficulties in maintaining repetitive volatile states while preserving their nonvolatile state, which poses a major challenge for achieving reliable reconfigurability. Here, we report a surface acoustic wave (SAW)-stimulated memristor using a two-dimensional (2D) transition metal dichalcogenide monolayer MoS2 that achieves reconfigurable memory operation in a reversible volatile mode enabled by acousto-electric modulation. SAW excitation provides contactless, strain-based control of the 2D material conductance for memristor operation, allowing dynamic and nondestructive volatile behavior without deteriorating the electrically programmed nonvolatile state. We leverage these tunable temporal dynamics to realize a SAW-driven memristive reservoir based on 2D monolayer MoS2, achieving 96.1% accuracy in a character classification task.

Conductive two-dimensional covalent organic frameworks (2D COFs) possess intrinsic π-conjugation and long-range order, yet their electrical performance in thin-film devices remains constrained by grain boundaries and amorphous regions that interrupt charge transport. Here we report a simple and scalable strategy to overcome these limitations by bridging polycrystalline COF domains with molecularly dispersed conjugated polymers (CPs). Guided by electronic alignment, geometric compatibility, and chain-length criteria, we identify COF-CP combinations that exhibit markedly enhanced conductivity when assembled into heterostructures. By demonstrating that electrical property enhancement occurs only below the CP crystallization threshold, we show that bridging CPs require short-range ordered or near-amorphous configurations to effectively span COF grains and establish continuous transport pathways. This approach is compatible with wafer-scale fabrication and enables ppb-level NO₂ sensing by coupling COF porosity with CP-mediated charge transport. Our results establish a rational polymer-bridging strategy for COF-based electronic materials and identify key design parameters for its extension to other COF-CP systems.

Achieving both steep‐slope switching and robust non‐volatile memory within a single transistor would enable new device concepts for low‐power and logic‐in‐memory architectures. Conventional negative‐capacitance FETs (NC‐FETs) are designed to suppress ferroelectric bistability to enable hysteresis‐free sub‐thermionic switching, whereas ferroelectric FETs (Fe‐FETs) exploit bistable polarization to provide non‐volatile memory, leaving the two operating regimes largely separate in practice. Here, we report a van der Waals ferroelectric negative‐capacitance transistor (FeNC‐FET) that simultaneously realizes stabilized negative capacitance and intrinsic bistable polarization by employing a CIPS/h‐BN/α‐In ₂ Se ₃ trilayer gate stack. Landau–Khalatnikov analysis and polarization–voltage measurements confirm complementary ferroelectric roles in which CIPS provides static negative curvature, h‐BN enables charge compensation, and α‐In ₂ Se ₃ supplies non‐volatile polarization. This cooperative mechanism enables steep sub‐threshold swings of 35 mV/dec (forward) and 51 mV/dec (reverse), a ∼3 V memory window, long retention (>10 ⁴ s), and endurance exceeding 2500 cycles. The device further performs AND, OR, and majority logic‐in‐memory operations using 10 µs pulses with clear ON/OFF separation. These results establish the FeNC‐FET as a compact platform that combines steep‐slope switching with non‐volatile programmability in a single device, enabling low‐bias readout, short‐pulse programming, and robust logic‐in‐memory operation.

Thermal–energy harvesting, which converts heat into electrical signals, offers a promising route toward sustainable energy utilization. However, achieving efficient and stable thermal–electrical conversion in low‐dimensional materials remains challenging. Although few‐layer Tellurene (Te) exhibits pyroelectricity owing to intrinsic broken symmetry, this effect is confined to the trilayer limit, ‐restricting its practical applications. Here, we demonstrate that Te nanowires (Te NWs) overcome this limitation through edge‐induced symmetry breaking, enabling robust and persistent pyroelectric polarization. Under photothermal excitation, Te NWs exhibit a synergistic pyroelectric–bolometric coupling, wherein photothermal modulation of polarization and resistance cooperatively enhances the thermally driven photocurrent. Without external bias, the device operates in a fully self‐driven mode, exhibiting excitatory synaptic responses governed solely by thermally driven charge dynamics. With biases, the response becomes bidirectional, emulating both excitatory and inhibitory synaptic characteristics. Leveraging these dual mechanisms, a Te NW‐based artificial vision system achieves 97% digit recognition accuracy through light‐driven learning without external electrical bias, demonstrating energy‐efficient, self‐stimulated neuromorphic functionality. This work establishes Te NWs as a promising material platform that bridges thermal–energy harvesting, bioinspired sensing, and self‐stimulated neuromorphic computing.

Achieving high-quality source/drain contacts in two-dimensional (2D) semiconductors remains challenging due to Fermi-level pinning induced by metal-induced gap states (MIGS). Here, we demonstrate an atomically sharp Hf₂C/HfSe₂ edge contact formed via a laterally directed chemical conversion, driven by catalyst-assisted hydrodeselenization and carbonization under epitaxial alignment. Classical and ab initio molecular dynamics elucidate the atomistic mechanism of heteroepitaxial interface formation. Scanning tunneling microscopy and spectroscopy measurements confirm suppressed MIGS at the heteroepitaxial interface, indicating an electronically transparent junction. Electrical measurements reveal a near-zero Schottky barrier height ( ≈ 5 meV) and reduced contact resistance ( ≈ 475 Ω·μm) when compared with previously reported 2D edge-contact systems. When co-integrated with a van der Waals-integrated native high-κ HfO₂ gate dielectric within a single HfSe₂ channel, the devices exhibit a subthreshold swing of 62 mV/dec and on-state current density of 920 μA/μm. This integrated platform establishes a scalable design framework that couples contact and gate-stack engineering for next-generation 2D logic technologies.

The emergence of ultralarge-scale hardware systems for artificial intelligence is driving demand for high-performance heterogeneous integration. At the heart of these systems lies the maximization of computational capability through high data bandwidth, necessitating interconnects that either increase the number of links between tiers and chips or enhance the data transfer rate of each link. Monolithic three-dimensional (M3D) integration, particularly with two-dimensional (2D) materials, offers ultradense intertier vias and multifunctional devices within back-end-of-line-compatible processes, enabling compact vertical stacking of logic and memory. A critical challenge in this architecture is thermal management, requiring cross-layer electro-thermal analysis and codesign with integrated power regulation. In parallel, photonic integrated circuits provide low-latency, energy-efficient interchip communication by overcoming the traditional bandwidth limitation imposed by electrical signal loss, and their advantages become increasingly significant as the communication distance increases. Emerging concepts, including spectrally tunable 2D photodetectors and vertically stacked microlight-emitting-diode-photodiode transceivers, further enhance scalability by eliminating reliance on external lasers. This Review article highlights the convergence of M3D integration, 2D materials, and photonic interconnects, while outlining challenges of material compatibility, process scalability, and system-level codesign that must be addressed to realize a unified framework for next-generation computing and communication systems beyond conventional Si scaling.