A layered graphene and chromium oxychloride device functions as a broadband optical synapse, detecting visible to deep infrared light and learning image orientation through built-in memory.
(Nanowerk Spotlight) Infrared light reveals information that ordinary vision cannot capture. It carries signatures of heat, chemical composition, and movement that support technologies from medical imaging to autonomous navigation. Infrared detectors capture signals yet rely on separate processors to make sense of them, an arrangement that is efficient for computation but inefficient for perception.
The brain offers a different model because it interprets sensory input where it is received and adjusts its responses as patterns change. Translating that principle into hardware is the central aim of neuromorphic engineering.
Attempts to design light-sensitive components that can both detect and learn have followed many directions. Early devices stored the trace of an optical signal through charge traps or structural defects inside semiconductors. These could mimic short-term memory but lacked precision and degraded over time. Other systems depended on materials sensitive only to visible wavelengths, leaving the infrared range, which is essential for thermal and environmental sensing, poorly addressed. Black phosphorus and related compounds responded well to infrared light but decayed quickly in air. The field needed a material system that could sense infrared light, modify its response with experience, and remain stable under normal conditions.
The rise of atomically thin materials made that goal more realistic. When researchers stack these layers into what are known as van der Waals heterostructures, they can control how electrons move across interfaces with atomic-level accuracy. This method has already advanced the design of photodetectors and transistors and now provides a path toward components that combine sensing, memory, and computation in one unit.
Graphene, a single layer of carbon atoms arranged in a honeycomb lattice, is central to this direction. It absorbs light across a wide range of wavelengths and conducts electrons efficiently. When it is paired with certain insulating crystals, graphene can turn light pulses into electrical changes that persist, which is an electronic form of learning. The challenge has been to extend this ability into the infrared region without losing stability or control.
A study published in Advanced Science (“A Van Der Waals Broadband Infrared Optical Synapse Enabling Orientation Detection”) addresses this problem. It presents a graphene and chromium oxychloride structure that reproduces synapse-like behavior from visible to deep infrared wavelengths. The device records, adapts, and classifies optical signals, showing how perception and learning can coexist in a single compact element.
The structure consists of two thin graphene layers separated by a crystal of chromium oxychloride that serves as a tunneling barrier. The stack is enclosed in a protective layer of hexagonal boron nitride to prevent degradation. When a small voltage is applied, electrons can tunnel through the chromium oxychloride layer even in the dark, which does not occur in equivalent devices built with metal contacts. Microscopy confirms that the chromium oxychloride layer is uniform and about sixteen nanometers thick, ensuring that the observed effects are not caused by defects or shorts.
The structure of the device and synaptic plasticity under 1550 nm laser optical stimulation. a) Schematic diagram of the few-layer graphene/CrOCl/few-layer graphene optoelectronic device, which is fully encapsulated in hBN. b) Optical image of the device. few-layer graphene and CrOCl are highlighted with black and yellow dashed lines. Inset shows the AFM morphology of the CrOCl, with a thickness of 16 nm. c) Cross-sectional TEM image of the device. Scale bar is 5 nm. d) Enlarged cross-sectional STEM-HAADF image in c showing the lattice structure of CrOCl. Scale bar is 1 nm. (Image: Adapted from DOI:10.1002/advs.202507530, CC BY) (click on image to enlarge)
To show that the device can act as an optical synapse, the researchers measured its response to light pulses at 1550 nanometers, a wavelength used in telecommunications. When they illuminated the device with a train of identical pulses, the electrical current increased gradually with each pulse.
This phenomenon, known as spike-number-dependent plasticity, mirrors how biological synapses strengthen with repeated stimulation. After sixty pulses, the current became about sixteen times higher than at the start, demonstrating that the device retained a record of previous exposure.
The team also explored how the response changed when the frequency of the pulses varied. Faster pulse sequences produced stronger current changes, a behavior known as frequency-dependent plasticity. The response scaled linearly with both the intensity and duration of each pulse, which allows the device to represent light strength in a controlled and predictable way.
A second experiment tested paired-pulse facilitation, a hallmark of short-term memory in neurons. When two light pulses arrived in close succession, the second produced a larger current than the first. At 1550 nanometers the facilitation index reached about 184 percent, confirming that the device can imitate the adaptive response of a biological synapse.
Extending these behaviors into the infrared has been a central challenge in optoelectronic research. Most thin-film devices show such plasticity only in visible light because infrared photons carry lower energy, making it harder to generate long-lived charge carriers.
The graphene–chromium oxychloride synapse overcomes that limitation. It exhibits stable plasticity across a wide range of wavelengths, including 520, 1064, 1400, 1550, and 2000 nanometers. Across this spectrum, the response trends remain consistent, showing that the same mechanism governs behavior from green light to deep infrared.
Thickness studies revealed that the effect is robust across devices with chromium oxychloride barriers between 2.5 and 40 nanometers. Although the absolute dark current decreases as the barrier becomes thicker, the synaptic behavior remains nearly unchanged. This indicates that the interfacial coupling between graphene and chromium oxychloride, rather than bulk properties of the barrier, drives the effect. Devices maintained stable operation after many cycles and months of testing, an essential feature for practical systems.
The physical explanation lies in how light alters the tunneling barrier between graphene and chromium oxychloride. Graphene has no band gap and therefore absorbs light over a wide spectral range, producing free electrons and holes when illuminated. Some of these photoexcited electrons transfer into the adjacent chromium oxychloride layer due to strong coupling at the interface. The transferred charges do not move freely but form a spatial pattern that changes the local electric potential.
This pattern temporarily lowers the energy barrier that electrons must cross, increasing the current. When the light is removed, the charges gradually disperse, and the current returns to its initial level. The delayed relaxation provides a form of memory, similar to the lingering activity of a synapse after stimulation.
This explanation is supported by theory and experiment. Quantum mechanical calculations known as density functional theory, used to describe how electrons behave in materials, show that charge transfer from graphene shifts the conduction band of chromium oxychloride to lower energy, making tunneling easier. Measurements of surface potential using a scanning probe method called Kelvin probe force microscopy reveal that illumination reduces the work function of the chromium oxychloride layer.
This confirms that light exposure modifies the barrier. Control experiments using hexagonal boron nitride in place of chromium oxychloride show ordinary photoresponse but no synaptic behavior, establishing that the observed effect arises specifically from the graphene–chromium oxychloride interface.
To demonstrate practical use, the researchers integrated the device into a computational framework known as reservoir computing. In this system, time-dependent signals are fed into a nonlinear element whose internal dynamics encode information, while a small output layer learns to classify the resulting patterns. The optical synapse provides the nonlinear dynamics because its electrical response depends on the timing and history of light pulses.
The researchers encoded simple black and white images representing different orientations into sequences of light pulses at 1550 nanometers and measured the corresponding electrical outputs. These signals were then used to train a lightweight artificial network to recognize image orientation. The system achieved about 98 percent accuracy after sixty learning cycles. The same approach was extended to a set of eight mouse-shaped images representing motion in different directions, with comparable accuracy and stability.
These demonstrations show that the device can both sense and process infrared information without additional circuitry or memory units. The work establishes several advances. It extends synaptic optical response from visible light to the deep infrared range of 2000 nanometers. It achieves this through interfacial charge coupling rather than unstable defect states. Its nearly linear response to light intensity simplifies encoding of optical information. It tolerates variations in material thickness and maintains stable performance over time. Together, these qualities make it a strong candidate for next-generation sensors that combine detection and interpretation in one platform.
Although this study focuses on single devices rather than full arrays, it provides a clear foundation for future integrated infrared vision systems. A sensor that can learn directly from infrared light could reduce the energy and time cost of processing images in low-light environments. It could also support autonomous systems that must interpret visual cues in real time, such as robots, satellites, and driver-assistance technologies.
By combining broadband light absorption with controlled interfacial charge transfer, the graphene and chromium oxychloride junction demonstrates that sensing and learning can occur within a single, compact structure. It marks a practical step toward neuromorphic optical systems that operate across the full spectrum of light visible and invisible, and that move computation closer to where perception begins.
For authors and communications departmentsclick to open
Lay summary
Prefilled posts
Nanowerk Newsletter
Get our Nanotechnology Spotlight updates to your inbox!
Thank you!
You have successfully joined our subscriber list.
Become a Spotlight guest author! Join our large and growing group of guest contributors. Have you just published a scientific paper or have other exciting developments to share with the nanotechnology community? Here is how to publish on nanowerk.com.
