Silicon retina mimics biology for a clearer view

20 October 2006
NewScientist.com news service
Tom Simonite

A silicon chip that faithfully mimics the neural circuitry of a real retina could lead to better bionic eyes for those with vision loss, researchers claim.

About 700,000 people in the developed world are diagnosed with age-related macular degeneration each year, and 1.5 million people worldwide suffer from a disease called retinitis pigmentosa. In both of these diseases, retinal cells, which convert light into nerve impulses at the back of the eye, gradually die.

Most artificial retinas connect an external camera to an implant behind the eye via a computer (see 'Bionic' eye may help reverse blindness). The new silicon chip created by Kareem Zaghloul at the University of Pennsylvania, US, and colleague Kwabena Boahen at Stanford University, also in the US, could remove the need for a camera and external computer altogether.

The circuit was built with the mammalian retina as its blueprint. The chip contains light sensors and circuitry that functions in much the same way as nerves in a real retina – they automatically filter the mass of visual data collected by the eye to leave only what the brain uses to build a picture of the world.
Fully implanted

"It has potential as a neuroprosthetic that can be fully implanted," Zaghloul told New Scientist. The chip could be embedded directly into the eye and connected to the nerves that carry signals to the brain's visual cortex.

To make the chip, the team first created a model of how light-sensitive neurons and other nerve cells in the retina connect to process light. They made a silicon version using manufacturing techniques already employed in the computer chip industry.

Their chip measures 3.5 x 3.3 millimetres and contains 5760 silicon phototransistors, which take the place of light-sensitive neurons in a living retina. These are connected up to 3600 transistors, which mimic the nerve cells that process light information and pass it on to the brain for higher processing. There are 13 different types of transistor, each with slightly different performance, mimicking different types of actual nerve cells.

"It does a good job with some of the functions a real retina performs," says Zaghloul. For example, the retina chip is able to automatically adjust to variations in light intensity and contrast. More impressively, says Patrick Deganeer, a neurobionics expert at Imperial College London, UK, it also deals with movement in the same way as a living retina.
Changing scene

The mammalian brain only receives new information from the eyes when something in a scene changes. This cuts down on the volume of information sent to the brain but is enough for it to work out what is happening in the world.

The retina chip performs in the same way. The lowest image (right) shows how this allows it to extract useful data from a moving face.

As well as having the potential to help humans with damaged vision, future versions of the retina chip could help robots too, adds Deganeer. "If you can perform more processing in hardware at the front end you reduce demand on your main processor, and could cut power consumption a lot," he explains.

Zaghloul and Boahen are currently concentrating on reducing the size and power consumption of the retina chip before considering clinical trials.

Journal reference: Journal of Neural Engineering (vol 3, p 257)