Crash-proof computer created by London researchers

Computer "mimics nature" to improve fault tolerance and automatically correct errors

Stewart Mitchell
18 Feb 2013

Researchers at University College London are working on a computer that can repair itself to prevent crashes – instantly recovering and fixing corrupted data.

In a report in the New Scientist, the researchers explain that their computer is based on the chaos of the natural world, which marks a significant break from the linear way in which conventional computers work through sets of instructions.

Instead of working through inputs from each program running on a PC to reach a goal, the "systemic" self-repairing computer mimics the way nature reacts to challenges.

"Its processes are distributed, decentralised and probabilistic. And they are fault tolerant, able to heal themselves,” said UCL computer scientist Peter Bentley. "A computer should be able to do that."

If one system becomes corrupted the computer can access another clean copy to repair its own code

The computer combines its instructions with the data it receives so that it can adapt the instructions to match changing circumstances, and hives data sets off to separate "systems" within the computer.

The computer can also use environmental data, linking the temperature outside with how to react if the conditions are too hot. Once it has calculated how to react to such a scenario, the results are divided between separate systems within the computer, where they are treated as individual elements.

The distributed nature of the systems is key to the project, the scientists say, because the computer "contains multiple copies of its instructions distributed across its many systems, so if one system becomes corrupted the computer can access another clean copy to repair its own code".

The result is that instead of crashing and rendering a screen of death, the system accesses the data from another of its self-contained systems to perform the operation, and then goes back and corrects the corrupt data.

Read more about: