FEM simulation of the RESET process of a phase change memory cell (IBM)
IBM has used phase change memory to build network of 913 neurons with 165,000 inter-connections, reports the MIT Technology Review.
The connections change as the chip processes incoming data, altering how the virtual neurons influence one another, allowing the device to learn. can recognise handwritten numbers.
The connections mimic the human brain’s synapses to produce so-called neurotrophic ICs which are thought to be the route to getting computers to understand video or learn from experience.
Previous efforts at using phase-change memory to build neuromorphic systems have been modest, with 100 synapses or less, says IBM researcher Geoff Burr.
After being shown 5,000 labelled images of handwritten digits from a standardised data set, the researchers’ chip could recognize handwritten digits it had never seen before with an accuracy of 82 percent.
from News http://www.electronicsweekly.com/news/research/device-rd/ibm-uses-phase-change-build-neuromorphic-ic-2014-12/
via Yuichun
沒有留言:
張貼留言