The spider and the fly

Researchers at the Center for Neuromorphic Systems Engineering at the California Institute of Technology are mimicking spiders and flies to improve the performance of electronic systems.

Visual sensors found in some living organisms appear to rely on the principle of using mechanical movement of some part of the visual system to increase the amount of information acquired.

Jumping spiders, for example, are known to acquire visual data by sweeping an essentially linear retina back and forth perpendicularly to its larger dimension, while slowly rotating the retina in its own plane. The human visual system also exhibits minute vibrations of an amplitude close to the smallest intercone separation. If the eyes are immobilised, images fade in a matter of seconds. In addition to preventing such fade-out, it has been postulated that these movements may be used to diminish the resolution limits imposed by the intercone spacing.

Now, researchers at the Center for Neuromorphic Systems Engineering at the California Institute of Technology aim to see if they can apply similar principles in the design of a novel CMOS visual sensor based system that may also offer similar benefits to its biological counterpart.

Specifically, the Caltech researchers propose to improve the resolution of imagers with focal plane processing by employing continuous low-amplitude vibrations on the order of pixel spacing along the optical path. Spatial intensity gradients turn into temporal intensity fluctuations which are detected and processed by every pixel independently from the others. This approach enhances resolution by allowing each pixel to gather information in time, and virtually eliminates fixed-pattern noise by permitting the detection of local features without interpixel comparisons.

To test out the theory, the developers have have designed and implemented an IC which incorporates this visual sensing principle. It consists of an array of 32 by 32 pixels with local temporal signal processing, and a novel non-arbitrated address-event communication scheme providing timing guarantees on external signals for easy interfacing with off-the-shelf digital components. Two methods have also been developed for generating the required mechanical vibrations.

Other researchers in the Department are mimicking other biological systems to improve the performance of electronic systems. One group of researchers, for example is developing a surface micromachined 2D angular velocity sensor – also known as gyroscope – by using a detection principle inspired by the fly’s haltere system. They expect that the new sensor will tolerate a higher noise level than previous designs for detecting the direction of the axis of rotation, thereby enabling a significant reduction of supply voltage and power consumption.

Another group have built a guidance system for a robot with an analog VLSI motion sensor based on the visual system of the fly.