Optical Sensor Mimicking the Human Eye Could Sharpen Neuromorphic ComputingDecember 16, 2020 by Luke James
A new type of optical sensor developed at Oregon State University can closely mimic the human eye’s ability to perceive changes in its visual field, researchers claim.
Researchers at Oregon State University (OSU) claim to have made a new type of optical sensor. Build from perovskite, the sensor can allegedly closely mimic the human eye’s ability to sense and perceive changes in its visual field. The researchers claim this could potentially buoy fields such as artificial intelligence (AI), robotics, and automotive because it may yield better image recognition technologies.
Input frame (left) and the voltage output from the simulated retinomorphic array based on the input (right). Image used courtesy of Applies Physics Letters
In research published in Applied Physical Letters, the OSU scientists describe how they were able to fabricate a simple photosensitive capacitor and characterize its response to physical stimuli.
A Bilayer Dielectric Structure
The sensor’s structure is based on a bilayer dielectric. The bottom dielectric is made from silicon dioxide (SiO2) and is expected to be highly insulating and irresponsive to light.
In contrast, the top dielectric, which is the prototypical perovskite made from methylammonium lead iodide (MAPbI3), is expected to have a large photoconductive response and a dielectric constant that “changes significantly” when illuminated. These properties make MAPbI3 a useful candidate for a dielectric medium.
Architecture of the Eye-Like Device
In their design, the bottom electrode is composed of highly-doped silicon that acts as both a substrate and an electrode. Meanwhile, the top electrode is made from 15nm of gold deposited via thermal evaporation.
The gold layer is thin enough to be partially transparent while still retaining its conductive properties. The researchers acknowledge that while the gold will retain its conductive properties, it will exhibit a large contact resistance.
A cross-sectional diagram of the photosensitive capacitor used in this study as a retinomorphic sensor. Image used courtesy of Applies Physics Letters
When the sensor is placed in series with an external resistance, the voltage dropped across the resistor temporarily spikes as the capacitor charges and discharges before returning to its equilibrium value. The result is a sensor that only spikes in response to changes in illumination, otherwise emitting zero other output.
This architecture makes the sensor very different from conventional technologies, which are designed for sequential processing. In a camera, for example, images are scanned across a two-dimensional array of sensors at a set frequency, each generating a signal with an amplitude that varies with the intensity of the light it receives.
This means that a static image will lead to an almost constant output voltage from the sensor. In contrast, the OSU team’s retinomorphic sensor stays "quiet" under static conditions and only registers a signal when it detects a change in illumination, quickly reverting to its quiet state.
A Match for Neuromorphic Computers?
Previous attempts by researchers to build this type of device, a "retinomorphic sensor," have relied on the use of software or complex hardware, says John Labram, an assistant professor of electrical engineering and computer science at OSU. In this study, however, the sensor’s operation is inherent in its fundamental design.
Example of a pixel circuit for a retinomorphic imager. Image used courtesy of Caltech
"You can think of it as a single-pixel doing something that would currently require a microprocessor," said Labram, who reckons that the new sensor could be a match for neuromorphic computing.
While the researchers are currently only able to detect one sensor at a time, they claim to have measured several devices and developed a numerical model that can replicate their behavior. This allowed the team to simulate an array of retinomorphic sensors to predict how a retinomorphic video camera would respond to input stimulus.