Tellurium nanowire transistors switch between boosting and suppressing their light response through voltage alone, enabling retina-like image preprocessing that cleans up noisy pictures at the sensor.
(Nanowerk Spotlight) Camera chips already turn light into electrical signals, but most of the hard work happens later. After a sensor captures an image, the data typically travels to separate processors that clean it up, sharpen contrast, and reduce noise before any recognition system can make sense of it. That pipeline works, but it can be wasteful. Raw visual data is huge and moving it back and forth costs energy and time.
Engineers have tried to shrink that distance by pushing computation closer to where light first becomes electricity. Biology offers a strong hint that this should be possible. The human retina does not act like a passive film. It performs early filtering and contrast shaping before signals ever reach the brain, using circuits that can either boost or suppress responses depending on context.
The retina manages this trick through two types of cells that respond to light in opposite ways. ON bipolar cells become more active when light increases. OFF bipolar cells become less active. Working together, they sharpen edges and suppress irrelevant background information before any signal leaves the eye. Engineers trying to recreate this behavior in silicon have faced a persistent obstacle.
Getting a single device to flip between boosting and suppressing its electrical output typically requires shining two different colors of light on it, perhaps ultraviolet to inhibit and infrared to excite. That works in the lab, but aligning multiple light sources adds bulk, cost, and complexity that limits practical use.
A team at Sungkyunkwan University and Chung-Ang University in South Korea has found a simpler path. In a paper published in Advanced Functional Materials (“Retina‐Mimetic In‐Sensor Visual Processing via Bidirectional Photoconductivity in Encapsulated Tellurium Nanowire Devices”), they describe transistors made from tellurium nanowires that switch between excitation and inhibition under a single color of light. The toggle is electrical: change the voltage on the device’s gate terminal, and the same ultraviolet beam that previously boosted current will now suppress it. No extra optics required.
Tellurium makes this possible because of its unusual atomic structure. The element’s atoms form corkscrew-shaped chains that stick together through weak forces, encouraging the material to grow into long, thin wires rather than flat sheets or bulk crystals. These nanowires absorb light across a broad range of wavelengths, from infrared at 850 nm down to ultraviolet at 405 nm, thanks to an optical bandgap of 1.37 eV.
a) Hierarchical structure of the human retina and representative responses of photoreceptors and bipolar cells under light and dark conditions. b) Signal modulation mechanisms of ON- and OFF-type bipolar cells under light illumination (excitation in ON bipolar cells via mGluR6 and inhibition in OFF bipolar cells via AMPA). c) Schematic architecture of the retina-mimetic in-sensor neuromorphic device (RIND). d) Tunable transition between positive and negative photoconductivity (PPC/NPC) governed by gate-bias polarity under identical light irradiation. e) Demonstration of in-sensor image preprocessing enabled by gate-controlled photoconductivity in neuromorphic transistors. f) Comparison of recognition efficiency between the noisy image and the pre-processed image via an artificial neural network (ANN). g) Block diagrams comparing overall imaging architectures between conventional approaches and the RIND-based processing system. (Image: Reproduced with permission from Wiley-VCH Verlag) (click on image to enlarge)
Researchers had previously noticed that under certain conditions, shining light on tellurium reduces rather than increases its electrical conductivity. Yet nobody had harnessed this inverse response for a working neuromorphic device.
The breakthrough came from adding a protective shell. The team synthesized tellurium nanowires in solution, deposited them onto a chip with indium tin oxide electrodes, and wrapped the structure in a 20 nm layer of aluminum oxide using atomic layer deposition.
Without that coating, bare nanowires behaved erratically. Their threshold voltage, which determines when the transistor switches on, drifted without pattern. Oxygen and moisture from the air degraded performance further. The aluminum oxide solved both problems. Threshold voltage variation shrank dramatically, and the devices developed a clean, reliable switching characteristic.
The encapsulation also created the conditions for bidirectional light response. At the boundary between tellurium and aluminum oxide, electrons can become trapped at defect sites. When ultraviolet light at 405 nm hits the nanowires, it knocks electrons loose from atoms, creating mobile electron-hole pairs.
What happens next depends on the gate voltage. Apply a positive voltage, and holes flow freely through the channel. Current rises. Apply a negative voltage, and the physics shifts. Electrons pile up at those interface traps. They then recombine with holes, pulling mobile carriers out of the channel. Current drops below its level in darkness.
Same light, opposite outcomes, controlled entirely by voltage.
The researchers showed that their devices also learn from experience. Sequences of combined light and voltage pulses induced gradual changes in conductivity. Positive gate pulses paired with light caused current to climb with each repetition. Negative gate pulses caused it to fall. The more pulses applied, the larger the cumulative shift. Longer pulses and brighter light amplified the effect.
This kind of use-dependent strengthening and weakening, called synaptic plasticity, is how biological neurons encode memory. The tellurium devices replicate it in hardware.
To demonstrate practical value, the team ran an image-cleanup test. They selected 16,000 handwritten digits from the standard MNIST dataset, a benchmark collection used throughout machine learning research. They then corrupted each image with random noise. Every pixel’s brightness became a voltage fed to the device under 30 mW cm⁻² ultraviolet illumination.
Activating negative photoconductivity across all pixels suppressed the noisy background. Selectively activating positive photoconductivity on brighter pixels amplified the digit strokes.
The improvement was measurable. For images with moderate noise, background suppression alone pushed recognition accuracy to roughly 93% after training, nearly matching performance on clean images. Heavy noise posed a harder test. Recognition collapsed when static drowned out the digits. But combining global suppression with selective amplification recovered accuracy to approximately 69%. The sensor itself performed contrast enhancement before any external processor intervened.
Single-wavelength operation eliminates an entire layer of engineering overhead. Systems relying on multiple light sources must keep them aligned, powered, and synchronized. A device needing only one color of light and a tunable voltage fits more naturally into compact, low-power platforms. Adjusting gate voltage is fast, precise, and requires no moving parts.
This work points toward vision hardware that participates in perception rather than passively recording it. Conventional cameras capture everything and let downstream systems sort it out. A retina-inspired sensor could reject noise and sharpen contrast at the source, cutting the data load before it ever hits a wire. The tellurium nanowire transistor now provides a material platform for building such systems, one thin aluminum oxide shell at a time.
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.
