Self-Powered Infrared 2D Nb8PtSe20 Sensors for Nighttime UAV Perception

Technical Summary: Self-Powered Infrared 2D Nb8PtSe20 Sensors for Nighttime UAV Perception

Problem Statement
Nighttime unmanned aerial vehicle (UAV) missions, such as search and rescue and border surveillance, require reliable perception in low-visibility conditions where visible-light imaging fails. While long-wave infrared (LWIR, 8–14 μm) sensing offers passive, all-weather capabilities, current high-performance LWIR detectors (e.g., HgCdTe, InSb) typically require cryogenic cooling and external bias, resulting in high size, weight, and power (SWaP) penalties unsuitable for compact UAV payloads. Furthermore, conventional von Neumann architectures separate sensing and computation; raw image data must be transmitted to a processor for feature extraction and fusion, incurring significant latency and energy costs. This bottleneck is particularly acute for detecting small, partially occluded targets where subtle local gradients are critical. There is a lack of a single-material, two-dimensional (2D) photodetector that simultaneously offers room-temperature, self-powered broadband operation (visible to LWIR), high responsivity, and the capability for in-sensor convolutional computing.

Methodology
The authors report the synthesis and characterization of a novel zero-bandgap 2D van der Waals material, Nb8PtSe20.

• Material Synthesis & Structure: High-quality single crystals were grown via high-temperature solid-state reactions. The material crystallizes in the monoclinic space group C2/m, featuring a stepped layered arrangement of [Nb8PtSe20]∞ chains. A key structural feature is the expansion of Se–Nb six-membered rings via Pt–Se bonding, creating the largest hexagonal unit-cell area (~15 Ų) reported for a 2D material to date. This structural distortion breaks the six-fold rotational symmetry found in materials like graphene, inducing strong in-plane anisotropy.
• Device Fabrication: Metal–Nb8PtSe20–metal pixel devices were fabricated via mechanical exfoliation and micro/nanofabrication. The devices were tested under zero-bias (self-powered) and biased conditions.
• Optoelectronic Characterization: The team evaluated the material's response across a broad spectrum (671 nm to 10.6 μm) using scanning photocurrent microscopy (SPCM), Raman spectroscopy, and blackbody radiation sources. They analyzed the photocurrent generation mechanism, specifically distinguishing between photovoltaic, photoconductive, and photothermoelectric (PTE) effects.
• In-Sensor Computing & System Integration: Leveraging the position- and intensity-dependent bidirectional photocurrent, the authors demonstrated in-sensor convolution. They mapped illumination positions to convolution kernel weights to perform denoising, edge detection, and sharpening. Finally, a 3×3 sensor array was integrated with a multimodal Vision Transformer (ViT) to evaluate nighttime small-object detection performance on the RGBTDronePerson dataset.

Key Contributions

1. Discovery of Nb8PtSe20: Identification of a zero-bandgap 2D metallic material with a unique expanded hexagonal lattice structure and pronounced in-plane anisotropy.
2. High-Performance Self-Powered Detection: Demonstration of room-temperature, self-powered photodetection spanning visible (671 nm) to LWIR (10.6 μm). The device achieves a zero-bias responsivity of 3.0 A W⁻¹ at 10.6 μm, rising to 100.6 A W⁻¹ at 0.5 V bias, representing the highest reported values for uncooled 2D LWIR photodetectors.
3. Mechanism Elucidation: Confirmation that the photocurrent is generated via the photothermoelectric (PTE) effect, driven by localized heating at electrode interfaces and the Seebeck effect, rather than photovoltaic or photoconductive mechanisms. This mechanism enables linear, position-dependent bipolar photocurrents essential for analog computing.
4. In-Sensor Convolution: Realization of hardware-efficient convolution operations directly at the sensor level. The device's ability to encode kernel weights through illumination position allows for real-time image preprocessing (denoising, edge detection) without data transfer to a central processor.
5. System-Level Validation: Successful integration of the sensor array into a multimodal vision system, achieving superior performance in nighttime UAV small-object detection compared to traditional architectures.

Results

• Structural Validation: XRD, STEM, and Raman spectroscopy confirmed phase purity, stoichiometry (Nb:Pt:Se ≈ 8:1:20), and strong vibrational anisotropy consistent with the C2h point group.
• Optoelectronic Performance: The device exhibited metallic transport behavior and linear I–V characteristics. Responsivity was measured at 6.6 A W⁻¹ (671 nm), 1.3 A W⁻¹ (1064 nm), and 3.0 A W⁻¹ (10.6 μm) under zero bias. The response time was competitive (rise times of 128–165 ms), and the device showed excellent stability over six months.
• Thermal Imaging Fidelity: Using the linear relationship between photocurrent and incident blackbody radiation (I ∝ T⁴), the authors reconstructed thermal images with a Structural Similarity Index (SSIM) of 0.97, confirming high-fidelity thermal mapping.
• Convolutional Processing: Experimental results matched simulations for in-sensor denoising and edge detection, validating the use of illumination position as a tunable convolution kernel.
• Detection Accuracy: When paired with a multimodal ViT, the Nb8PtSe20-based system achieved a 50.76% mean average precision (mAP) on the RGBTDronePerson dataset for nighttime small-object detection. This outperformed architectures without in-sensor convolution (45.82%) and single-modality networks (23.00%).

Significance
The paper establishes Nb8PtSe20 as a versatile platform for broadband optoelectronics, specifically addressing the SWaP constraints of UAV payloads. By combining high LWIR responsivity with self-powered operation and in-sensor computing, the work demonstrates a pathway toward compact, hardware-efficient intelligent vision systems capable of operating in low-visibility environments. The ability to perform analog convolution directly at the sensor level reduces data movement and latency, offering a practical solution for real-time perception tasks where traditional von Neumann systems are limited. The findings suggest that zero-bandgap 2D metallic materials can overcome the limitations of semiconducting 2D materials in the LWIR regime, enabling a new generation of multimodal, low-power sensors for autonomous aerial platforms.