Abstract

With computers and the internet accounting for more than 10% of global electricity production, engineers and computer scientists are attempting to design more energy efficient computers and they are turning to the brain for inspiration. But how energy efficient is the brain, how are its neurons and circuits designed to improve energy efficiency, and can these neural designs be usefully implemented by engineers?

Starting with the first measurements of a neuron’s energy efficiency, made almost twenty years ago in the Zoology Department, I will present principles of neural design that make a brain more efficient. These principles include computing directly using analogue circuits (as opposed to digital), computing with chemical reactions, and using parallel circuits designed for specific tasks. In addition, like all efficient devices, the brain matches the energy it uses to the task in hand. This often boils down to doing things as inaccurately and slowly as possible, where “the as possible” is “just good enough to survive”. The brain’s remarkable ability to adapt its components depends upon its “winning technology”, molecular and cell biology. Emulating molecular systems that have been honed over more than a billion years of evolution will be a major challenge for future generations of engineers.