What if a camera pixel could see color, store the signal, and generate a fresh encryption key in the same exposure cycle? This paper builds that idea from plasmonic RRAM devices. In-sensor computing aims to bring data storage and processing capabilities to sensors.

The idea is to handle image information closer to where it is detected. In-sensor computing is an emerging field that aims to bring data storage and processing capabilities directly to sensors, combining storage and processing at the sensing site.

The device uses optoelectronic RRAM arrays with semitransparent noble-metal electrodes for in-sensor color image cryptography. Silver and gold are selected as top electrodes to respond to blue and red light, respectively. Few-layer hexagonal boron nitride forms the resistive-switching layer, while an inert gold film serves as the bottom electrode.

Color images are captured by comparing low-resistance-state currents during the set process under dark and illuminated conditions, while reconfigurable PUF keys are harvested in every exposure cycle from high-resistance-state resistance. Different absorption peaks let the devices mimic biological dichromatic cone cells, and the plasmonic effect facilitates rupture of the conductive filament.

Together, the arrays capture color images and generate reconfigurable PUF keys for in-sensor color image cryptography. Figure one introduces the gold–hexagonal boron nitride–gold optoelectronic RRAM array, its layered structure, and a photograph of the fabricated device.

The electrical plots compare dark and six-hundred-nanometer illumination during switching: light changes the low-resistance-state current, produces distinct device-to-device resistance distributions, and alters both SET and RESET behavior. The readout in panel h confirms that these differences persist after irradiation, supporting light-controlled filament growth and dissolution rather than photocurrent alone.

The gold device is a three-by-five array made from a few-layer hexagonal boron nitride sheet between a semitransparent top gold electrode and an inert bottom gold electrode. Each device is five by ten square micrometers. The h-BN layer, top gold electrode, and bottom gold electrode are approximately seven, five, and twenty nanometers thick.

The top-electrode thickness was optimized for the plasmonic effect while maintaining resistance switching. Under dark conditions, the device shows stable asymmetric bipolar resistance switching, but its reset current is more than ten to the power of minus two amperes, much higher than the ten to the power of minus five ampere compliance current.

Under light irradiation, the current-voltage curve becomes symmetric and the reset current drops dramatically to ten to the power of minus five amperes. This indicates that light affects conductive-filament growth dynamics. That behavior makes low-resistance-state resistance useful for recording image information, while the high-resistance state remains independent of external light stimulation.

The high-resistance resistance has an intrinsic random distribution from ten to the fourth to ten to the eighth ohms, providing a source for reconfigurable PUF keys. Figure two lays out the Ag/h-BN/Au optoelectronic RRAM, using a five-nanometer Ag top electrode, roughly seven-nanometer h-BN, and a twenty-nanometer Au bottom electrode.

The current–voltage curves show switching in darkness and under four-hundred-nanometer illumination, while the device distributions compare low-resistance states across devices. Panel f further reports RESET current for Ag/h-BN/Au and Au/h-BN/Au under dark and illuminated conditions, showing how light and electrode material influence switching behavior.

Figure three links the device architecture to dichromatic color vision: the array combines Ag/h-BN/Au and Au/h-BN/Au RRAM pixels, representing two cone-cell types tuned to about four hundred and six hundred nanometers. The extinction spectra show distinct optical features, including a three-hundred-seventy-nanometer peak for the Ag structure and a five-hundred-fifteen-nanometer peak for the Au structure.

The figure therefore motivates wavelength-selective optoelectronic memory elements for color perception. The plasmonic silver and gold arrays can mimic two types of cone cells in dichromatic mammals by responding to characteristic wavelengths. The extinction spectra show pronounced absorption peaks at three hundred seventy nanometers for silver and five hundred fifteen nanometers for gold.

Absorption at four hundred nanometers is higher than fifty percent for the silver structure, while absorption at six hundred nanometers is higher than fifty percent for the gold structure, carrying purple and yellow color information. These wavelengths are close to the human visual-system cone-cell absorption peaks of four hundred twenty and five hundred fifty-eight nanometers.

Different wavelengths induce significant differences in low-resistance-state conductance, showing color selectivity. Two standalone silver and gold devices can be integrated into one pixel responding to four hundred and six hundred nanometers. The different reset currents also enable color perception, and tuning the plasmonic characteristic wavelength of the top metal electrode could enable trichromatic and tetrachromatic color visions.

Figure four combines electrode morphology, simulations, and switching mechanisms. SEM and AFM reveal nanoparticle-textured ultrathin gold and silver films, while the simulations show local electric-field enhancement and heat generation in the h-BN gap.

The final panels propose how light-generated hot electrons affect filament formation and rupture: in the gold device they interact with boron ions, whereas in the silver device they inhibit or facilitate silver oxidation during SET and RESET. Instead of using reset voltage or low-resistance-state resistance as PUF keys, the device harvests high-resistance-state resistance because it varies more significantly and decouples PUF generation from image capture.

High-resistance values range from ten to the fourth to ten to the eighth ohms in gold devices and from ten to the sixth to ten to the eighth ohms in silver devices. A double-binary PUF key is generated by comparing both devices in one pixel with a ten-megaohm reference resistance.

Both above the reference gives bits eleven, both below gives bits zero zero, and only one above gives bits zero one or one zero depending on the device. Figure five d shows the obtained high-resistance state, and the device can reconfigure that resistance in the next cycle, as shown in Figure five e.

To quantitatively evaluate the PUF key's performance, the study characterizes its randomness and irreproducibility using three measures: uniformity, uniqueness, and stability. The reported Hamming-weight uniformity is zero point five, uniqueness is zero point five, and stability is zero point nine zero three, close to the stated ideal values.

Figure five shows a complete in-sensor cryptography workflow: the array captures color using Ag and Au RRAM devices under four-hundred or six-hundred nanometer light, then resets them to the high-resistance state for PUF key generation. Panels b and c show the resistance distributions, while d and e illustrate harvesting HRS values for reconfigurable keys.

In panel f, the captured color image is converted into double-binary data, combined with the PUF key for encryption, and then recovered through decryption. The schematic illustrates color image capture, image encryption, and decryption, using a three-by-three optoelectronic RRAM array as the worked example.

Figure five f schematically illustrates color image capture, image encryption, and decryption as one integrated workflow for the three-by-three optoelectronic RRAM array. Each pixel contains two RRAM devices with silver and gold top electrodes, and the array is exposed to a letter Z image with binary colors near four hundred and six hundred nanometers.

The low-resistance-state current is compared with dark current to obtain photocurrent and color, and the captured color image is converted into a double-binary bit array. Negligible photocurrent from both devices maps to zero zero, obvious photocurrent from both maps to eleven, and photocurrent from only one maps to zero one or one zero according to the responding device.

The device is then reset to the high-resistance state, whose resistance becomes the PUF key. The Vernam algorithm encrypts and decrypts images through an exclusive OR operation between the captured image and the PUF key. The device array can therefore capture the color image and generate the PUF key simultaneously.

The authors also propose an experimental exclusive OR scheme using circuit-level parallel connection of two optoelectronic RRAM devices with a series load resistance. The prototype demonstrates the concept, while future large-area h-BN grown by chemical vapor deposition could support wafer-scale RRAM crossbar arrays, using CMOS-compatible processes with high scalability.

The fabricated h-BN plasmonic optoelectronic RRAM uses silver and gold top electrodes for in-sensor color image cryptography. Light enhances conductive-filament annihilation and induces optically controlled resistance switching for image capture and color recognition.

The photoresponse comes from the plasmonic effect of the noble-metal top electrodes, with characteristic wavelengths of four hundred and six hundred nanometers for silver and gold, respectively. Reconfigurable double-binary PUF keys come from comparing high-resistance-state resistance with a reference resistance, using the device's random variation.

The demonstrated cryptography converts image intensity and color into a double-binary bit array. The device structure and pixel design provide a route to encrypt and decrypt both image intensity and color information. The broader implication is that in-sensor cryptography can bridge the sensory terminal with security hardware and enhance security protocols.

The device array combines color-selective sensing with reconfigurable double-binary PUF keys, demonstrating in-sensor image cryptography that encodes image intensity and color before transmission.