Learning, thinking machine on its way

Wei Lu, a computer engineering boffin researching at the University of Michigan, has connected a so-called memristor with conventional circuits to create a unit which can learn and remember, writes Science Daily. Lu is aiming to create a computer which is as powerful as a cat’s brain.

It would be able to make decisions and be far more multitasking than regular PCs.

Connecting two electronic circuits with a memristor emulates a process called “spike timing dependent plasticity.” It is believed it is the way a brain works – plasticity being the connection made between neurons firing up. Lu’s circuit acts like a synapse, which connect neurons in a brain.

The more synapses fire in a short amount of time, the more likely it becomes that information is passed on. Neurons learn to make new connections to one another through this process. The memristor can act that way as it acts in line with previous voltage.

Lu’s next step is to create a supercomputer the size of a two litre pack of juice. However, it will take a few years time until such a device can be built. His research project is being funded by DARPA and the National Science Foundation. It is not known if DARPA wants to the give the final product legs and a machine gun. 

The memristor, which combines memory with a resistor, was first predicted in 1971 by Leon Chua, yet Stanley Williams was the first person able to create one in 2008 at Hewlett Packard labs.

HP is currently researching how to stack memristors in order to make memory chips which will offer ten times as much density as flash chips. Memristors are immune to radiation and retain data when you switch your PC off. They also consume very low voltage and can be made at extremely small nano widths.

It can be expected HP will be filing truckloads of patents so Rambus can’t feed off of their memristor revenue once the technology hits the market in over five years time.