Chongfan Technology

News

All Categories >

Home page

2025

Beijing Institute of Technology Review: Infrared Detectors for In-Memory Sensing and Computing

Author:

In-memory sensing and computing (ISC) devices integrate the functions of sensors, memory, and processors, offering advantages such as low power consumption, high bandwidth, and zero latency. They are particularly well-suited for simulating synaptic behaviors in biological neural networks. As the pace of digital transformation accelerates, the demand for efficient information-processing technologies continues to grow, and in-memory sensing and computing devices are demonstrating tremendous potential in the fields of artificial intelligence (AI), machine learning, and edge computing. In recent years, with continuous advancements in infrared detector technology, infrared ISC devices have also ushered in new opportunities for development.

According to Mems Consulting, a research team from Beijing University of Technology has published a review article titled “Advances in Infrared Detectors for In-Memory Sensing and Computing” in the journal Photonics. This review focuses on recent research achievements in infrared ISC devices. The research team first introduces the working principles and performance metrics of ISC devices, then discusses transistor- and memristor-based devices with infrared-band responsiveness, and finally provides a forward-looking perspective on the development of this field. This review offers valuable reference for future research in this area and aims to promote the continued development and innovation of infrared ISC devices.

1. Operating Principle and Performance Metrics of ISC Devices

In-memory sensing and computing (ISC) devices function similarly to the way neurons and synapses operate in biological neural networks—particularly in terms of their highly efficient information-processing capabilities. In the human brain, neurons are interconnected via synapses to form complex networks that enable tasks such as learning, memory, and computation. These processes exhibit high parallelism and remarkable energy efficiency. In neuromorphic computing, artificial synaptic devices mimic the signal-transmission processes of biological synapses through specific electronic or optoelectronic properties. Such devices typically consist of two main types of components: two-terminal memristors and three-terminal transistors. These two structures have distinct operational mechanisms and characteristics.

Plasticity, retention characteristics, switching ratio, durability, power consumption, and linearity are all critical metrics for evaluating the performance of synaptic electronic devices. Together, these parameters determine the reliability and effectiveness of the devices in practical applications, laying the foundation for designing high-performance neuromorphic computing systems.

2 Development of Infrared ISC Devices

This chapter focuses on ISC devices that respond in the infrared wavelength range. Based on their core structures, these devices are categorized into “phototransistors” and “photoresistive memristors.” The research team, drawing on the latest case studies, elaborates on the technological breakthroughs achieved in these fields.

2.1 Optoelectronic Transistor for Storage and Computation

2.1.1 Heterojunction Transistor

Heterojunction transistors leverage the heterojunction structure formed between different materials to achieve photoelectric response and storage functionality. These transistors typically consist of two or more materials. Under illumination, photogenerated electron-hole pairs are generated and separated at the heterojunction, thereby altering the conductivity of the materials. This structural design not only enables efficient photoelectric conversion but also allows precise control over the device’s conductive state by adjusting the intensity and duration of light exposure, thus realizing both storage and computing functions.

Transition metal dichalcogenides (TMDs), with the general formula MX₂ (where M = molybdenum or tungsten; X = sulfur or selenium), have attracted considerable attention due to their outstanding performance in infrared detection. The bandgap of these materials ranges from approximately 1.4 eV to 2.0 eV, covering the visible to near-infrared (NIR) spectral region—a range that is crucial for infrared detection. Transition metal dichalcogenides also exhibit a strong excitonic effect, with exciton binding energies reaching tens of millielectronvolts, which significantly enhances their efficiency in absorbing and emitting light in the infrared wavelength range. Moreover, the bandgap of transition metal dichalcogenides can be tuned via strain engineering. Experiments have shown that a strain of just 1% can induce a bandgap shift of about 300 meV. This strain sensitivity provides an effective means of tailoring the spectral response of detectors. Consequently, transition metal dichalcogenides hold great promise for infrared detection applications thanks to their tunable bandgaps, strong excitonic effects, and controllable strain responsiveness.

2.1.2 Floating-Gate Transistor

Floating-gate transistors leverage a floating-gate structure to achieve non-volatile storage functionality. These transistors store electrical charge by embedding an isolated floating-gate layer beneath the gate electrode. When an external voltage is applied, electrons can be injected from the source into the floating-gate layer and trapped by the isolation layer, thereby altering the conductivity of the channel. For example, a floating-gate transistor based on a BP/Al₂O₃/WSe₂/h-BN structure generates a non-volatile photocurrent under illumination, enabling it to perform storage functions. Thanks to their ability to retain state information even after power is turned off, floating-gate transistors are an ideal choice for low-power storage applications.

a) Schematic diagram illustrating the working principle of BP-PPT programmed using optical pulses; (b) Microscopic image of the reconfigurable BP photodiode; (c) The storage structure—comprising MoS₂ (channel material), h-BN (tunneling layer), and graphene (floating-gate layer)—on the left half, and the optoelectronic detection structure formed by the BP/MoS₂ heterojunction—on the right half—jointly constitute a non-volatile sensor-in-memory computing device; (d) Real-time testing of conductance configurations and photoresponses under different conductance states.

2.2 Optoelectronic Memristors for Storage and Computation

2.2.1 Photon-Electron Coupled Photoresistive Memristor

The basic structure of a photon-electron-coupled optoelectronic memristor comprises two or more electrodes and a light-sensitive active material layer. When the device is illuminated, the energy of photons excites electrons in the active material, generating electron-hole pairs. The migration of these charge carriers alters the material’s conductivity, thereby influencing the memristor’s resistance state. In some oxide-semiconductor-based optoelectronic memristors, illumination induces the ionization and deionization of oxygen vacancies—a process analogous to long-term potentiation (LTP) and long-term depression (LTD) observed in biological neuronal synapses, offering the potential to mimic complex neural network behaviors. Beyond the ionization and deionization processes, defect traps within the optoelectronic memristor can capture and release photogenerated charge carriers; this mechanism is crucial for achieving both short-term potentiation and long-term potentiation. Moreover, by constructing heterojunctions—such as p-n junctions or type-II heterojunctions—potential wells can be formed at the interface, effectively suppressing the recombination of photogenerated charge carriers. As a result, the device can maintain more persistent changes in conductivity following optical stimulation, further enabling the emulation of dynamic synaptic behaviors.

2.2.2 Conductive-filament memristor

The core operating principle of conductive filament-based memristors lies in storing data by forming a conductive filament between the electrodes. Such a memristor consists of two electrodes and an intermediate storage medium layer, typically made of electrochemically active materials like metal oxides or sulfides. In the unprogrammed state, the memristor exhibits a high resistance, representing “0”; whereas in the programmed state, applying a high voltage induces ion migration within the material, thereby forming a conductive path—known as a conductive filament—between the electrodes. At this point, the resistance drops, representing “1”.

The formation of conductive filaments is analogous to the electrochemical metallization process: When a sufficiently high voltage is applied, the electric field drives ion migration, causing these ions to accumulate between the electrodes and form a conductive pathway. By precisely controlling the voltage and current, this process can be accurately regulated, thereby adjusting the resistance state of the memristor and enabling multi-level storage. To erase the stored data (i.e., restore the device to a high-resistance state), a lower voltage or a brief current pulse can be applied to break the conductive filament.

The non-volatility of conductive-filament memristors means that they can retain their resistance state even after power is turned off, which is crucial for data preservation. Moreover, this type of memristor features a simple structure, low power consumption, high operating speed, and excellent scalability, making it an ideal choice for applications such as neuromorphic computing systems.

3 Infrared Neural Network

Building on previous research into infrared artificial synapses based on two-terminal memristors and three-terminal transistors, recent studies have further integrated sensing, storage, and information-processing functions using optoelectronic neuromorphic devices to emulate biological visual systems. Corresponding to the ability of visible-light visual systems to perceive and recognize colors through wavelength information, infrared-optical visual systems can be developed to identify temperature information, which can serve as a basis for distinguishing between objects. As a key technology in the field of human-computer interaction, gesture recognition has received extensive research and application in recent years. By combining infrared detection signals with motion-detection devices or image-recognition systems, it is possible to achieve highly efficient detection of object-motion information.

Moreover, by leveraging memristors to simulate the memory function of neurons, artificial visual systems can achieve long-term or short-term memory for detected images. Such infrared sensing systems fill a gap in nighttime image acquisition and memory capabilities, enabling robust memory and processing of feature information about objects in low-light conditions.

4 Conclusion and Outlook

In summary, current research is advancing the development of ISC devices along several key directions: multi-band response, material innovation, neuromorphic computing, and integration compatibility. These studies not only broaden the application scope of these devices but also enhance their performance, making them better aligned with practical needs. This review summarizes the high-performance infrared ISC devices mentioned in this article and highlights the progress made in this field.

Moreover, these devices are increasingly being designed to respond to optical signals spanning the ultraviolet to near-infrared and even mid-infrared spectral ranges. This enhanced multi-band responsiveness indicates that researchers are striving to expand the applicability of these devices to meet a broader range of practical application needs. Researchers are also continuously exploring new material systems. Two-dimensional materials such as molybdenum disulfide (MoS₂), indium selenide (In₂Se₃), and black phosphorus have become research hotspots due to their unique physical properties. While significantly boosting photoelectric conversion efficiency, these materials also exhibit exceptional mechanical flexibility and compatibility with other materials. To achieve miniaturization and integration of devices, researchers take into account compatibility with existing manufacturing processes already during the design phase.

However, current research still has certain limitations. For example, many classical neuromorphic device architectures that perform exceptionally well in the visible-light spectrum have yet to be applied to the infrared regime—such as photon-ion coupling, phase transitions, ferroelectric memristors, or transistors. Researchers could explore leveraging optical signals to control ion migration, facilitate transitions between different material phases (e.g., crystalline and amorphous states), or modulate the polarization state of ferroelectric materials, thereby developing optoelectronic synaptic devices with infrared responsiveness. If these operational mechanisms can be successfully introduced into infrared devices and their performance optimized, it might lead to groundbreaking technological advances.

Moreover, existing research has primarily focused on broadband-responsive devices, leaving insufficient exploration of specific spectral bands such as the mid-wave infrared. Materials like tellurene (Te) exhibit high photocurrent densities across a broad spectrum; however, materials with excellent infrared-band response—such as mercury telluride—still require further investigation in ISC devices.

On the other hand, existing studies typically evaluate the learning or computational performance of devices using specific application scenarios and simple neural networks. However, these test scenarios rarely take into account the real-world application requirements of infrared detection, such as night vision, remote sensing monitoring, medical diagnosis, and industrial inspection. Only a handful of studies have provided simulations tailored to infrared-band applications, such as nighttime ship docking, gesture recognition, or vein detection. Therefore, designing dedicated test platforms and application scenarios that are specifically suited to the characteristics of infrared detection will become an essential component of future research. This will help better assess the actual performance and applicability of ISC devices in infrared detection, thereby promoting their practical implementation.

Overall, although significant progress has been made in areas such as multi-band response, material innovation, neuromorphic computing, and integration compatibility, there are still limitations that need to be overcome before infrared-band ISC devices can achieve widespread adoption. By drawing on the operational mechanisms of conventional neuromorphic devices, conducting in-depth research into new materials suitable for the mid-wave infrared band, and designing application scenarios that meet practical needs, we can expect further advancements in this field and ultimately achieve major breakthroughs in the application of ISC devices for infrared detection.

None

High-performance III-V family infrared detectors

None

High-performance III-V family infrared detectors