While electrical pulses convey impulses along neurons, the cells communicate with each other and with other cells such as muscles by releasing chemical messengers. These neurotransmitters are released from one side of a cell junction, or synapse, and picked up by receptors on the other side, triggering another electrical pulse.

Since synapses are typically around 50 nanometres across, and each chemical puff contains just a few thousand molecules, building an artificial synapse is a huge challenge. But Mark Peterman and Harvey Fishman at Stanford University in California are getting close. They told a biophysics conference in Texas earlier in March that they have created four "artificial synapses" on a silicon chip one centimetre square.

To cells on the surface of the device, the artificial synapse is simply a hole in the silicon. But each hole opens into a pipeline etched into a plastic layer on the back of the chip, connected at both ends to a reservoir of neurotransmitter. When an electric field is applied, the neurotransmitter is pumped through the pipeline, and a little of it squeezes out of the hole, stimulating nearby cells (see graphic).

At 5000 nanometres wide, these artificial synapses are closer in size to a whole cell than a real synapse, but even so, the pair have fine-tuned the device so that it can stimulate just one cell in the layer above the chip.

Neural prosthetics

The ultimate ambition is to develop neural prosthetics - implanted devices that can interact with our nervous system. Devices that use electrical stimulation are already commonplace, such as the cochlear implants that partially restore hearing.

But electric pulses stimulate nerve cells indiscriminately. Different neurotransmitters, in contrast, have different effects on a given cell. What is more, a single neurotransmitter can affect one cell in one way and a different cell in another. The neurotransmitter used by retinal cells, for example, turns some cells on and others off.

It means that devices that use neurotransmitters could interact with cells in more subtle and precise ways. Biomedical engineer Gerald Loeb from the University of Southern California speculates that powerful devices could be produced by combining chemical and electrical stimulation in one implant.

But there are still formidable obstacles. How densely can you pack the synapses in when each needs its own pipeline? How do you stop the pipelines from clogging with immune cells when the device is implanted? And how often would you need to refill the neurotransmitter reservoir?

To answer this last question, Peterman estimates that a thousand artificial synapses firing a thousand times a second would need as little as half a millilitre of fluid to function for 250 years. The other problems may prove tougher. "But it's early days," Loeb points out.

In the meantime, the most immediate application of the technique could be in tissue research. Drugs could be delivered to individual cells in a tissue sample to see how this affects the entire system.