Korean Scientists Create Low-Powered Smartphone AI Chip

Artificial Intelligence AI Brain AH

Researchers out of the South Korea Advanced Institute of Science and Technology (KAIST) and startup UX Factory may soon enter the burgeoning market for A.I. chipsets. That’s according to a February 27 announcement from the country’s Ministry of Science and ICT. Although the use of A.I. chips in smartphones is not a new thing, the team – led by professor Yoo Hoi-jun – was reportedly successful in creating a new A.I. semiconductor which they believe could be a boon to the smartphone industry.

The news about the innovation has been notably slim on fine details. The team is said to have successfully widened the scope of the A.I. to support machine learning in image-based object recognition, in addition to analyzing data that changes over time. Moreover, that accomplishment was made while reducing the amount of energy required by the chip by programming it to adjust its energy efficiency and its accuracy to individual circumstances. A byproduct of that, according to reports out of the region, is that deep learning can also be processed much more efficiently overall. As a proof of concept, the team developed an emotion-recognition system to test the A.I. It was ultimately able to recognize each of the seven facial expressions it was programmed to look for in human subjects via a smartphone camera before automatically displaying the appropriate emotion on-screen in real time.

If further research bears out across more testing of the A.I. chip’s capacity, the project’s partners believe it could handle learning a multitude of tasks. Those would range from object and emotion recognition to motion recognition and automatic translation but would also extend far beyond that. More importantly, if the claims about its efficiency and versatility turn out to be accurate, it’s feasible that the chips could drive a proliferation of A.I.-dedicated hardware in mobile devices. Other devices, namely those featuring HiSilicon’s Kirin 970 SoC, already have their own machine learning chipsets. However, the newly created semiconductor could turn out to be a better option than those that are currently available or could result in another alternative option for budget handset manufacturers. Better still, it could help drive other industries that are currently dependent on more established chipset manufacturers.