MIT develops new super power-efficient AI chip for smartphones
A research team from the Massachusetts Institute of Technology (MIT) has successfully developed a new chip that is designed to allow for faster artificial intelligence processing in mobile devices (especially smartphones and tablets), while at the same time significantly lowering the power consumption by up to 95 percent.
One can measure the effectiveness of artificial intelligence solutions by how efficiently they run neural networks. For those not familiar with the term, neural networks are basically interconnected data processors that interact with each other. By processing information together, these processors can handle tons of data and execute loads of tasks. Some of today’s cool artificial intelligence capabilities, such as speech recognition, image manipulation, and facial recognition, are made possible through neural networks.
But it goes without saying that the larger the amount of data and tasks that a neural network processes, the more power it will need to be able to continue running. This is why the MIT researchers’ new AI chip is truly a marvel -- it can get away with superfast analysis of data without requiring too much power. And when they say superfast, they mean about three to seven times faster than the usual speeds.
How did they pull this off? Well, MIT’s research crew did it by finding a way to simplify the machine learning algorithms in neural networks into a single point -- the dot product. This way, the processing of information is concentrated in just one area, instead of the data being passed around from memory to the nodes (and vice versa). With the new AI processor, dot products for multiple nodes can be computed with a single step, as opposed to transporting the raw data for every node between the memory and the chip.
The world of artificial intelligence has been busy lately with new developments. Earlier this week, ARM had officially unveiled a couple of new AI chips -- specifically, a machine learning processor that is capable of handling typical artificial intelligence tasks, and a next generation object detection chip that comes with the ability to detect human faces and even human gestures and actions, in videos that are rendered in very high res quality (even hitting full HD levels) and running at 60 frames per second. Meanwhile, AI continues to be embraced by today’s mobile brands -- the latest to join in on the party is LG, which is planning to unleash its own AI suite at the upcoming Mobile World Congress event (kicking off this February 26th at the beautiful city of Barcelona in Spain).