[ad_1]
As artificial intelligence is becoming more sophisticated, it has inspired new efforts to develop computers whose physical architecture mimics the human brain. An approach, called reservoir computation, allows hardware devices to perform the higher dimensional computations required by emerging artificial intelligence. A new device highlights the potential of extremely small mechanical systems to perform these calculations.
A group of researchers from the Université de Sherbrooke in Quebec City, Canada, reports the construction of the first tank computing device built with a microelectromechanical system (MEMS). Posted in Journal of Applied Physicsfrom AIP Publishing, the neural network exploits the non-linear dynamics of a silicon beam at the microscopic scale to perform its calculations. The group's work aims to create devices that can simultaneously act as a sensor and computer, using a fraction of the energy that a normal computer would use.
The article appears in a special section of the journal devoted to "Physical News and Materials for Neuromorphic Computation," which highlights new developments in physics and materials physics research that are promising for the development of "neuromorphic" systems. On a very large scale tomorrow, the calculation will exceed the limits of current semiconductors.
"These types of calculations are normally done only by software and computers can be inefficient," said Guillaume Dion, author of the newspaper. "Today, many sensors are built with MEMS. Devices like ours would be an ideal technology to blur the boundaries between sensors and computers. "
The device relies on the nonlinear dynamics of how the silicon beam, of a width 20 times thinner than a human hair, oscillates in space. The results of this oscillation are used to build a virtual neural network that projects the input signal into the higher dimensional space required for neural network computing.
During demonstrations, the system was able to switch between different common reference tasks for neural networks with relative relative ease, including classifying spoken sounds and processing binary models with an accuracy of 78.2%. 99.9% respectively.
"This tiny silicon beam can accomplish very different tasks," said Julien Sylvestre, another author of the newspaper. "It's surprisingly easy to adjust it to recognize the words."
Mr. Sylvestre said he and his colleagues were looking to explore increasingly complex calculations using the silicon beam device, hoping to develop small-sized robot sensors and controllers. energy efficient.
Source:
American Institute of Physics. .
Source link
