The new coprocessor will be used to implement deep neural networks (DNNs) and will run continuously off the HoloLens battery. It will coexist with the HoloLens' main processor - the Holographic Processing Unit (HPU), which processes information coming from all of the unit's on-board sensors - to help analyze data without using the cloud.
This will allow for faster processing and greater mobility for the unit, says the company, as well as for future additional augmented- and mixed-reality features. According to reports, the coprocessor was designed to handle complex voice and gesture interactions for the next-generation HoloLens, and enable the unit to do things like being able to recognize objects in the line of sight.
The company is designing the AI coprocessor in house. Microsoft's CTO, Kevin Scott, told Bloomberg, "We really do need custom silicon to help power some of the scenarios and applications that we are building."
Microsoft says the coprocessor will be able to "natively and flexibly" implement DNNs, and support a wide variety of layer types, fully programmable by Microsoft. At the unveiling of the new chip at computer vision expo CVPR 2017, an early version of the new HoloLens HPU was demonstrated running live code implementing hand segmentation.
HoloLens 2 does not currently have a scheduled release date, but is likely expected to be generally available sometime in 2019.
AI wars heat up with new Microsoft research lab
VR tech in manufacturing: One-time "Sword of Damocles" has prosperous future
Microsoft acquires deep learning startup
Apple launches machine learning research site
AI's 'human side' is focus of new Google initiative