Mid-infrared cameras are typically associated with military night vision, but their importance now extends far beyond seeing in darkness. They are used to monitor industrial emissions, track heat loss in buildings, guide autonomous vehicles and detect objects hidden by smoke, haze or poor illumination. Unlike conventional cameras, which record reflected visible light, infrared systems measure thermal radiation emitted by objects. The challenge is that modern sensors must do more than collect images. They increasingly need to interpret thermal scenes at the point of detection, identifying edges, suppressing backgrounds and highlighting targets before vast streams of raw data reach an external processor. A research team has now developed a programmable photothermoelectric detector based on a suspended, nanometer-thick membrane of strontium titanate, or SrTiO₃, a perovskite oxide that could help move this kind of image processing directly into the sensor.
The device tackles one of the central limitations of thermal detection: the compromise between sensitivity and speed. Conventional thermal detectors can function at room temperature and respond across broad wavelength ranges, but their operation depends on heating and cooling a physical structure. If the active material has a large thermal mass, it takes longer to change temperature, limiting the detector’s response time. The researchers addressed this problem by reducing the SrTiO₃ thermoelectric channel to nanometer-scale thickness and suspending it above the substrate. With less material to heat, the membrane can respond more quickly to incoming radiation. At the same time, suspending the channel reduces heat leakage into the underlying substrate, helping maintain the temperature difference required to generate a measurable electrical signal.
The detector operates through the photothermoelectric effect. When infrared radiation is absorbed, one region of the suspended channel becomes warmer than another, producing a lateral temperature gradient. In a thermoelectric material, that gradient drives charge carriers and generates a voltage known as the Seebeck voltage. The magnitude of this voltage depends not only on the temperature difference but also on the Seebeck coefficient, which describes how efficiently a material converts a thermal gradient into electrical potential. The ultrathin suspended architecture is therefore important for two reasons: it lowers the thermal mass and preserves the lateral temperature gradients that would otherwise be weakened by heat spreading. According to the study, the resulting detector produces a broadband, zero-bias photovoltage response extending from visible wavelengths to approximately 10 micrometers, covering a substantial portion of the mid-infrared range without requiring an externally applied operating voltage.
Speed is one of the most striking characteristics reported for the device. The detector reaches a response time of about 10 microseconds, a substantial improvement over bulk SrTiO₃ photothermoelectric devices discussed by the researchers. The reported response is more than 10,000 times faster than those bulk counterparts. Such a rapid response matters because thermal cameras often generate large quantities of data, and slow pixels can limit frame rates or blur rapidly changing scenes. A fast detector can respond to transient thermal signatures, moving objects and modulated infrared sources while reducing the need for extensive correction after image acquisition. The performance also demonstrates how engineering the geometry of a material can be as important as selecting the material itself: by isolating a very small active volume, the researchers altered the balance between heat capacity, thermal transport and electrical readout.
The team then added a second layer of functionality by placing two local bottom gates beneath the suspended channel. These gates do not simply switch the detector on or off. Instead, they modify the electrical properties of separate regions of the thermoelectric channel. When opposite voltages are applied to the two gates, the researchers can create a controlled contrast in the local Seebeck coefficient across the membrane. Under illumination, the same thermal gradient can therefore produce signals with different strengths or even opposite polarities. In practical terms, the detector becomes programmable: its response can be electronically reconfigured rather than permanently determined during fabrication. This capability is especially significant for infrared image processing, where different tasks may require a sensor to emphasize bright thermal objects, detect temperature differences or identify rapid spatial changes.
The researchers report a responsivity of approximately 50 volts per watt under infrared illumination. Responsivity measures the electrical output generated for a given amount of incident optical power, and a high value is useful when weak thermal signals must be distinguished from noise. The dual-gate design also produced 40 experimentally measured stable response states. These states represent different combinations of response magnitude and polarity that can be selected through gate voltages. Instead of sending every pixel’s unprocessed signal to a computer and applying all transformations digitally, an array of such detectors could perform some mathematical operations as the image is being captured. This approach, often called in-sensor or near-sensor computing, can reduce data movement, lower energy consumption and accelerate the extraction of relevant features from thermal scenes.
To explore that possibility, the team fabricated a 3-by-3 array of suspended SrTiO₃ detectors. Although small, the array allowed the researchers to examine whether the programmable behavior could be reproduced across multiple devices rather than demonstrated only in a single laboratory structure. The array showed repeatable switching between positive and negative responses when the gate voltages were changed. This polarity control enables the detectors to act as weighted elements in spatial filtering operations. In image processing, a convolutional kernel assigns different positive or negative weights to neighboring pixels and combines their signals to reveal patterns. By programming the response of individual infrared detectors, the researchers configured the array to approximate operations including inversion, difference-of-Gaussian filtering and Laplacian processing.
These operations are widely used to expose features that may be difficult to see in an untreated image. An inverse transformation can reverse the visual representation of thermal intensity. A difference-of-Gaussian filter compares information at different spatial scales and can help isolate objects from slowly varying backgrounds. A Laplacian operation emphasizes sharp changes, such as the boundary between a warm target and a cooler environment. Performing such transformations through the detector’s physical response could allow a thermal imaging system to produce feature-enhanced outputs before data are digitized or transferred to a processor. The researchers also demonstrated attention-guided enhancement of thermal targets using experimentally measured responsivity states. The result suggests that programmable thermoelectric pixels might eventually support sensing systems that assign greater importance to selected regions of a scene.
The study’s demonstration should nevertheless be interpreted as a step toward intelligent infrared electronics rather than a complete autonomous vision platform. The attention-guided thermal enhancement relied on dataset-driven, off-chip reconstruction, meaning that the measured detector responses were used in processing performed outside the device. The experiment therefore does not yet represent a fully integrated, real-time recognition system operating entirely on the sensor chip. Further development will be needed to increase array size, improve uniformity between pixels, address readout circuitry and establish long-term stability under repeated thermal cycling. Researchers will also need to examine noise, fabrication tolerances, power requirements and the ability of the suspended membranes to withstand practical operating environments.
Even with those limitations, the work identifies suspended nanometer-thick SrTiO₃ as a promising platform for reconfigurable infrared sensing. The material combines a broad spectral response with thermoelectric operation at zero bias, while the suspended geometry provides the thermal isolation needed for fast detection. More importantly, electrostatic control of the Seebeck coefficient gives the detector a programmable electrical personality: it can be tuned to respond positively, negatively or with different strengths depending on the task. If the approach can be scaled into larger, reliable arrays, future infrared cameras could perform portions of image enhancement and feature extraction as radiation is converted into electrical signals. That could reduce the volume of information sent to external processors and help create faster, more energy-efficient systems for night vision, autonomous navigation, environmental monitoring and other applications where seeing heat is only the beginning.
Subject of Research: Suspended nanometer-thick SrTiO₃ photothermoelectric detectors for programmable mid-infrared sensing and in-sensor image preprocessing.
Web References: https://doi.org/10.1093/nsr/nwag499
References: National Science Review, DOI: 10.1093/nsr/nwag499.
Image Credits: © Science China Press
Keywords
mid-infrared imaging, infrared detectors, photothermoelectric effect, strontium titanate, SrTiO₃, thermoelectric materials, Seebeck coefficient, nanomaterials, suspended membranes, in-sensor computing, infrared image processing, perovskite oxides, thermal cameras, programmable electronics
