Neuromorphics engineers are professional plagiarists, freely borrowing their ideas from nature's bag of computational tricks. But the flow is not one way. By building silicon models of animal brains, researchers can also learn about biology.

Many neuromorphics engineers pay lip service to this idea, but researchers at the Institute of Neuroinformatics in Zurich are putting it to work. Neuroscientist Kevan Martin and electronics engineer Shih-Chii Liu are collaborating on a project to investigate how our brains process visual images.

Visual clues: Shih-Chii Liu has designed an analog chip that models some of the brain processes involved in image analysis.

The project centres on a device made by Liu called the cortical chip, an analog silicon circuit based on part of the human visual system. Our brains analyse images using a series of different areas of the cerebral cortex. The first of these, known as 'V1', performs low-level processing such as detecting motion and the orientation of lines.

Liu's chip attempts to model two of V1's six layers. It uses a network of artificial neurons connected to mimic the links between layers 4 and 6. Layer 4 is interesting because it receives many of the inputs to V1. To make the chip as realistic as possible, the input to the chip is also taken from biology. Martin has recorded the activity of neurons in part of the cat brain involved in sending signals to layer 4 of V1. Liu feeds these recordings straight into her chip.

Liu can adjust the connections and properties of the artificial neurons to see how they affect the performance of the whole network. By tweaking specific parameters — such as the number of links between layers 4 and 6 — Liu generates insights that Martin can relate to neurophysiological studies. “The number of questions we have to answer is overwhelming,” says Martin.

For example, Martin is interested in how different neurons in layer 4 form specific 'receptive fields' — in other words, why each responds to a different area of space within the eye's field of view. Each neuron's receptive field is thought to be controlled by inputs from another brain area known as the thalamus. But Martin suspects that there is more to it than that. “Fifty per cent of all connections to layer 4 neurons come from layer 6,” he points out. “So what does layer 6 do? With this circuit, we can adjust the layer 6 synapses and use real data to find out.”