The team from Massachusetts Institute of Technology (MIT) has built an energy-friendly chip that performs powerful artificial intelligence (AI) tasks, which is specially designed to implement neural networks.
According to MIT researchers this chip is 10 times as efficient as a mobile GPU (Graphics Processing Unit), that is, it could enable mobile devices to run powerful AI algorithms locally, rather than uploading data to the Internet for processing.
Vivienne Sze, the Emanuel E. Landsman Career Development Assistant Professor in MIT’s Department of Electrical Engineering and Computer Science whose group developed the new chip, says that currently most of the networks are pretty complicated due to their high-power GPUs.
Sze states the reason why it’s better for devices to operate locally rather than via the Internet: your cell phone still powerfully operates even if you don’t have a WI-Fi connection nearby required to process large amount of information. You have much better privacy of your information, and another reason is the avoidance of any transmission latency, so that you can react much faster to certain applications.
MIT researches have presented the new chip at the “International Solid State Circuits Conference” in San Francisco.
The new chip is called “Eyeriss,” and according to MIT – “its key to efficiency is to minimize the frequency with which cores need to exchange data with distant memory banks, an operation that consumes a good deal of time and energy.”
Moreover, many of the cores in a GPU share a single, large memory bank while each of the Eyeris’s cores has its own memory. Thus, the chip is capable of compressing data before sending it to individual cores.
The Eyeriss chip has 168 cores, which could communicate directly with each other, so that if they need to share data between each other, they don’t have to route it through main memory.
“This is essential in a convolutional neural network, in which so many nodes are processing the same data.”
Neural nets, also known under the name “deep learning,” were widely studied in the early days of AI, however, by the 1970s, the researchers had thrown this field out of favour.
Sze says the deep learning is useful for many applications, such as speech and object recognition, and face detection.