Count Microsoft among the companies preparing to build specialized chips for artificial intelligence (AI): The next version of the company’s HoloLens augmented reality headset will come with a chip capable of complex AI computation, revealed Microsoft Research VP Harryn Shum at a computer vision conference Sunday. This will make it possible to improve hand tracking on the device, as well as run object recognition and other computer vision tasks.
Cloud-based server farms are getting very good at using artificial intelligence for advanced image recognition, something that Google and Facebook use to recognize persons and even objects in images uploaded to their services. But for HoloLens to work without delays, tasks like these have to be solved locally — which is why the company is putting AI straight into the headset itself.
“We’re in the business of making untethered mixed reality devices,” explained HoloLens director of science Marc Pollefeys in a blog post. “We put the battery on your head, in addition to the compute, the sensors, and the display.”
The current version of HoloLens already features a custom processor, dubbed the Holographic Processing Unit, that makes use of all of the device’s sensor data of features like head tracking. The next version of this chip will incorporate an artificial intelligence co-processor, Shum said.
Microsoft isn’t the only company building custom chips for AI and similar tasks. Apple has long been building its own chips for the iPhone and the iPad, and is increasingly optimizing them for on-device image recognition tasks. And Google recently hired one of Apple’s key chip architects to build custom chips for its own phones, which will likely incorporate machine learning as well.