A new energy-efficient computer chip that can do powerful artificial intelligence (AI) tasks on smartphones and Internet of Things (IoT) devices has been developed by researchers. The tiny chip called Eyeriss developed at the Massachusetts Institute of Technology (MIT) would integrate things called "neural networks" modeled after the human brain into mobile AI.

Like Us on Facebook

MIT introduced its new chip at the International Solid-State Circuits Conference in San Francisco. The chip contains 168 cores and requires no central memory bank.

Eyeriss has many key features. It has its own memory, compresses processed data, and contains a special circuit to maximize each core's workload.  

Neural networks are usually inserted in a mobile device's graphics processing unit (GPU). The reason is a mobile GPU can have up to 200 processing units/cores for building a network.

However, MIT's game-changing chip is 10 times more efficient than standard mobile GPUs, according to Tech Times. This would allow handsets to run AI algorithms instead of uploading or downloading data from the web. The latter is the current method for processing AI data.

Vivienne Sze is a professor at the MIT Department of Electrical Engineering and Computer Science. She explained that neural networks are now usually run on powerful GPUs, but her research team's chips would also provide faster processing and better privacy.

Using the AI tech on mobile and IoT devices would also allow them to function without a WiFi or Bluetooth connection.

MIT's chip could be useful in the development of future IoT tech. This would allow connected devices including smartphones and tablets to only upload final results to the Internet.

Such new hardware could change how AI interacts with the net. It could be a big plus for connected devices and mainly for ones with slow Internet connections.

In recent years AI technology has started to change people's day-to-day lives. Applications include Facebook's algorithms to see important news on timelines, and Google learning what people like to create targeted ads.

The big problem that researches are solving is that machine learning requires tons of processing power and energy. Last year Qualcomm unveiled a machine learning platform built into a Snapdragon processor that could power smartphone AI, according to Android Headlines.   

Here's a video on the Internet of Things: