The "cognitive computers" built with these chips are not programmed the same way today's traditional computers are. These computers, instead, learn from their experiences, find correlations, develop assumptions, remember - and learn - results, thus imitating the plasticity of the human brain, IBM promises.
Two prototype chips have been manufactured and are being tested, said the American group. Both were burned with an ultimate line of 45 nanometers on silicon on insulator, and contain the equivalent of 256 "neurons".
IBM is testing two types of architectures for these chips: one with 262,144 synapses (zone of interaction between nerve cells) programmable, the other with 65,536 synapses for learning. The long term goal of IBM is to build a complex component which contains 10 million "neurons" but it's still very far from the human brain, which includes 100 billion. The goal is to get at creating 100,000 billion synapses in a space less than 2 liters, while using one kilowatt of electricity, said the American group.
In terms of applications ,a "cognitive"computer is for example capable of triggering a tsunami alert by analyzing data from different types of marine sensors, compiling data on temperature, pressure and wave height, explains IBM. It could also help retailers manage their inventories fresh products through its sense of "smell".
For the second phase of the project, called Synapse, IBM has enlisted the help of several renowned universities such as Columbia, Cornell, the University of California and that of Wisconsin. The project also benefits from a funding of $ 21 million (14.6 million) from Darpa, the agency that funds high-tech projects in the field of defense.
IBM has already designed Watson, a computer with the ability to learn from mistakes, and a surge capacity of semantic analysis which makes it much more effective than traditional software in the association of ideas.
Source: lemonde.fr
No comments:
Post a Comment