The human eye can see in a dimly lit room and, moments later, look directly at a bright sky — a photodetection range of more than 160 decibels. Conventional cameras and machine vision systems require complex circuitry, algorithms, and power to approach even a fraction of that range. A team at the University of Electronic Science and Technology of China (UESTC) and Penn State University has now built a device that does it intrinsically, using nothing more than a composite of titanium dioxide and a conductive polymer.
The device is a two-terminal photomemristor — a light-sensitive resistor with memory — made from a thin film of anatase-phase TiO₂ and the conductive polymer PEDOT:PSS, sandwiched between transparent and metal electrodes. At just 0.5 mm across, it mimics the retina’s ability to adapt to vastly different light levels using a single physical mechanism: water.
How it works
When light hits the TiO₂ layer, it generates a photocurrent that passes through the PEDOT:PSS. Under bright illumination, the photothermal effect heats the device (from approximately 24.2 to 32.7 degrees Celsius at 320 mW/cm²), causing water molecules adsorbed on the polymer surface to desorb. This reduces the concentration of charge-carrying H₃O⁺ ions, decreasing conductivity and suppressing the photoresponse. Under dim light, water re-adsorbs, conductivity increases, and the photoresponse is amplified.
The result is an intrinsic, materials-level adaptation that requires no auxiliary circuits or real-time algorithms. The device’s photosensitivity score drops from +2,527 percent under normal light to -33.7 percent under extreme brightness — effectively mimicking the pupil constriction and rhodopsin bleaching that protect the human retina.
The adaptation time is shorter than the human eye’s.
Demonstrated performance
The researchers built a 4×4 photomemristor array and connected it to an artificial neural network with spike-timing-dependent plasticity. The system achieved 91.3 percent accuracy in recognizing images under dynamically changing mixed-light conditions — glare, dim shadows, alternating brightness — without any algorithmic compensation for lighting.
This compares favorably with conventional machine vision systems, which degrade significantly under lighting changes and require separate sensor calibration, HDR processing, and computational white-balance correction — all of which consume additional power and hardware.
The biomimetic advantage
The human retina achieves its extraordinary dynamic range through photopigment bleaching and regeneration — a constant cycle of chemical depletion and renewal that sets the sensitivity level moment by moment. The TiO₂/PEDOT:PSS photomemristor achieves the same effect through water adsorption and desorption, a reversible physical process that operates at the material level.
“Full vision adaptation in mixed-light conditions enabled by dynamic water adsorption/desorption,” the paper is titled, published in Nature Communications (DOI: 10.1038/s41467-026-73217-7). The device also demonstrates paired-pulse facilitation (PPF) index up to 3.18 at 10 ms intervals — a hallmark of synaptic plasticity.
Next steps
Lead author Jia Zhu of UESTC said the team plans to optimize the material formula and fabrication process, integrate a fully packaged array with signal readout modules and adaptive control units, and ultimately build a functional bionic artificial eye prototype. The researchers believe the same dynamic regulation strategy could be extended to other oxide-polymer composite memristor systems, potentially establishing a universal design principle for high-performance light-adaptive neuromorphic devices.
Potential applications include autonomous driving (resisting headlight glare while identifying pedestrians in dim conditions), humanoid robots, outdoor surveillance, aerial reconnaissance, and portable vision systems where power and hardware volume are constrained.
Sources
1. Physics World, “Machine vision begins to work like the human eye” (8 July 2026). https://physicsworld.com/a/machine-vision-begins-to-work-like-the-human-eye/
2. Zhu, J. et al., “Full vision adaptation in mixed-light conditions enabled by dynamic water adsorption/desorption”, Nature Communications 17, 4965 (2026). DOI: 10.1038/s41467-026-73217-7