Wednesday, May 29, 2024

Artificial electronic retina can recognize handwritten numbers

Neuromorphic vision sensors have been extremely beneficial in developing energy-efficient intelligent systems for robotics and privacy-preserving security applications. There is an extreme need for devices to mimic the retina’s photoreceptors that encode the light illumination into a sequence of spikes to develop such sensors.

KAUST researchers have built an artificial electronic retina that can “see” in a similar way to the human vision system and can recognize handwritten digits. They have designed and fabricated an array of perovskite-based flexible photoreceptors that detect the visible light intensity via a change in electrical capacitance, mimicking the behavior of the eye’s rod retina cells.

For their photoreceptors, the KAUST team used a hybrid material of perovskite nanocrystals embedded in the ferroelectric terpolymer. Perovskite is very efficient at absorbing light and is already of great interest in solar cell research, while terpolymer has a high dielectric constant. This hybrid material is sandwiched between a bottom aluminum electrode and a patterned top electrode of indium tin oxide to form a pixelated array of miniature light-sensitive metal-insulator-metal capacitors. The array is made on a thin substrate of polyimide that lets the device flex and bends into whatever shape is needed, including a hemispherical shape mimicking the human eye.

The photoreceptor array was connected to an electronic CMOS-sensing circuit and a spiking neural network with 100 output neurons. In tests with a 4×4 array, the LED illumination of different visible colors indicate that the optical response of the array mimics the response of the human eye with maximum sensitivity to green light. In other tests, the artificial electronic retina was able to recognize handwritten numbers with an accuracy of around 70%. The photoreceptors are also found to be highly stable, with no change in response even after being stored for 129 weeks in ambient conditions.

“The ultimate goal of our research in this area is to develop efficient neuromorphic vision sensors to build efficient cameras for computer vision applications,” explained Salama. “Existing systems use photodetectors that require power for their operation and thus consume a lot of energy, even on standby. In contrast, our proposed photoreceptors are capacitive devices that don’t consume static power for their operation.”

In future work, the team plans to build larger arrays of photoreceptors, optimize the interface circuit design and employ a multilayered neural network to improve the accuracy of the recognition functionality.