What do you get when you cross a really bad signal filter with another really bad signal filter? It sounds like the start of an awful joke but, believe it or not, it might just be the question that dooms modern electronics as, maybe, just maybe, the answer to our ‘joke’ is “a better filter”.

To understand how this simple question might have any bearing at all it must be appreciated that evolution via selection does not apply just to nature. Darwin’s theories apply to any system in which properties of individuals are inherited from 'parents', mutations can occur between generations and there is some form of selection pressure. Unsurprisingly, we’ve not found many systems in nature which obey these strict criteria, and plants and animals are the only clear example. In nature traits are passed in the genes, mutations can occur and selection is via Natural Selection. At Sussex University however, Dr. Adrian Thompson wondered why we had to find such systems to observe them: why not build our own?

Theoretically, if he could make a system that had both heritable traits and the chance for mutations coupled with a method of controlling selection pressures, he would be able to produce whatever ‘offspring’ he wanted. Humans have used artificial selection before to do things like domesticate animals or breed new types of crop plant but this would be taking it to a whole new level.

Dr. Thompson decided to test the idea by creating a circuit that could distinguish between two different inputs, one at 1kHz and one at 10kHz. Depending on the input the circuit should output either zero or five volts. This sort of circuit is simple for an engineer using conventional methods of constructing electrical circuits. However, to ensure that the evolved circuit would have to take a unique approach he provided only a fraction of the number of components (a mere 100) compared to more conventional circuits, as well as deliberately providing no system clock. To enable this circuit to change he used chips that could change their internal logical structure for whatever task was assigned to them called Field Programmable Gate Array (FPGA) chips. Enough of these could become any device with the right programme loaded into them. He then set up each chip to have a random initial program and, having fed in the two audio codes, he programmed a computer to assess the effectiveness of the configurations. The best were allowed to intermingle and the process repeated with a few minor random mutations in the programme between generations.

And so the process went on and on, but within a few hundred generations the circuit was beginning to change function. By generation 220 it basically just replayed the input signal. By the time it had passed 600 it was starting to become sensitive to 1kHz, at about generation 1,500 it could tell the signals apart about half the time and finally, after about 4,000 generations, it settled on a program that differentiated between the two exactly as hoped. Spurred by this success, the Sussex team even managed to get it to tell the difference between the spoken commands “stop” and “go”.

This success alone seems rather remarkable, but there was more to come. Upon dissection of the final circuit it was found that only thirty-seven components were actually used, of which five weren’t even connected. The team couldn’t figure out how the system worked. What’s more, the program they evolved wasn’t effective when loaded onto another chip of 100 FPGAs. It seems that the evolution method is so efficient in its chip usage because instead of worrying about the internal logic of the chips - like a human engineer would - it takes advantage of the unique electromagnetic properties of the circuit. Perhaps the best example of this is that the evolved circuit ignored the conventional ‘on-off’ set up of the transistors and seemed to use each as an analogue device.

The implications of this are remarkable: circuits that are literally playing by a different rulebook to anything that we could build traditionally. Already the potential of evolved hardware is being taken seriously. NASA has used similar techniques to build antenna that are radically different from current designs. Not only that but the potential for circuits that can rearrange their functions to bypass damage is being hailed as the next generation method for spaceships vulnerable to radiation and stray debris. However, the other side of this is that soon we may be placing our trust in circuits grown to order where any unknown ‘gene’ might cause catastrophe. It might seem like a comfort to astronauts heading to Mars that their spaceship is more robust, but how comfortable would you be on a plane piloted by a chip no one understands? Like it or not, this is just the question you might have to ask yourself in the not too distant future.

Written by Andrew Maddox