news

Brain-inspired computer memory consumes less energy

A quantum phenomenon is key to a low-energy technology for electronic memory.

A new technology for fast, non-volatile computer memory uses a similar mechanism to the human brain, according to its developers at Paris-Sud University. This technology works with no power attached, but is faster than USB keys or CD memory retrieval.

Both current types of computer memory — volatile, which only retains data when it has power running through it, and non-volatile, which does not need power — store information in the form of an electrical charge. Volatile memory systems work fast, whereas non-volatile are slow. A newer version, called magnetic memory (STT-MRAM), is as fast as volatile memory but does not need applied energy. Unfortunately it’s also very expensive.

Rather than using charge, STT-MRAM stores data as magnetic orientation of electron spin, a quantum-mechanical phenomenon. Its basic units, called magnetic transfer junctions, are programmed by applying a voltage across their junctions. But if the voltage pulse is not long enough, the programming can be incorrect. AS this is a quantum phenomenon, there’s a degree of randomness proportional to the length of the pulse. Conventional electronic memory can’t tolerate randomness.

However, the Paris team has found a way to turn this to their advantage. Using MTJs like synapses (gaps between nerve junctions) in the human brain, they found that the more an individual MTJs is called upon to store a piece of data, the more likely it is that the piece of data will be recorded. To put this another way, after a number of repetitions, the system learns a function: like the way the brain learns a new task. Simulation studies suggest that systems of MTJs can perform  cognitive tasks like video or image analysis more efficiently, with less energy, than conventional memory.

The team explains its research in a paper in the journal IEEE Transaction on Biomedical Circuits and Systems.

Read more: http://www.theengineer.co.uk/news/brain-inspired-computer-memory-consumes-less-energy/1020279.article#ixzz3YhAgo1tH

Details

Category

engineering precision

Date

April 29, 2015

Author

Sally

social
Share This Project
Comment Form