Posts Tagged ‘neural networks’

Engineers find ‘missing link’ of electronics

May 1, 2008

Paul Marks, NewScientist.com news service, 18:00 30 April 2008

Nanoscale circuits based on molecules used in sunscreen lotion have led to the discovery of the “missing link” of electronics engineering – a previously mythical device known as a “memristor”.

First predicted in 1971, the memristor could help develop denser memory chips or even electronic circuits that mimic the synapses of the human brain, says Stan Williams who made the discovery with colleagues at Hewlett-Packard’s lab in Palo Alto, California.
===========================================

Some cool stuff here. Basically all electronics boil down to resistors (current flows through and some energy is given off as heat), capacitors (charge is temporarily stored), and inductors (current flows through and some energy is stored in a magnetic field). In 1971, Leon Chua at UC Berkeley predicted that there should be a 4th hardware device that remembers the last voltage applied to it and how long it was applied. Now someone has made such a device.

Memristors are cool because they mimic properties of neurons. In neural networks of cells or in computer simulations, each neuron has multiple inputs but only one output. So there are lots of On and Off signals coming in (chemical and/or electrical signals in cells), but the neuron has to take all that info and the either fire or not. Basically, it has to reach a voltage threshold before it will fire. Memristors work the same way. The consequences are that you can store information in the resistance of the device, and you can have passive computer memory (no power needed to maintain it).

Even cooler, you can envision a solid state brain. To model neurons now, we have to program them into computers, thus slowing the process down. Making the model out of solid state parts means that it could work much faster. From a plot standpoint, that means AI without having to have massive supercomputers. And of course that leads to the Singularity, which is overdone.

How about making solid state brain “blanks”. Then we can download our mind into the blank when we die (or before) and live on as cyborgs. For inspiration, see Robocop or read David Brin’s Kiln People.

Advertisements