Sampson Wilcox, Research Laboratory of Electronics
New optical processor developed by MIT performs tasks in mere nanosecute. Team also developed an optical neuronal network architecture for signal processing.
MIT researchers have created a new AI hardware accelerator designed specifically for the Wireless signal processing.
The new optical processor performs tasks in mere nanosecute, advances, performing machine learning activities Light speed.
The photonic chip works about 100 times faster than the best digital alternatives Currently available and reaches about 95% accuracy in signal classification.
“There are many applications that would become possible with EDGE devices capable of analyzing wireless signals. What we have presented in our article can open many possibilities for inference in real time and reliable. This work is the beginning of something that can have a significant impact,” says Dirk Englund, a professor at the MIT Electrical Engineering and Computer Science Department and author of published in Science This month.
Researchers developed an optical neuronal network architecture specifically for signal processing, which they called Optical neuronal network of multiplicative frequency analog transformation.
“We were able to integrate 10,000 neurons into a single device and perform the necessary multiplications at once,” explains Ronald Davis, the main author.
“The longer you measure, the greater the accuracy obtained. As MAFT-ONN performs inferences in nanoseconds, there is virtually no speed to gain accuracy,” he adds.
In the future, researchers plan to use multiplexagem To perform more calculations and expand the maft-on, and apply its discoveries to more complex Deep Learning architectures.
