Microsoft has, for the first time, provided a look at the hardware inside its Hololens augmented reality headset, at the Imec Technology Forum in Belgium. Although there have been teardowns of the dev kit versions, which began to be shipped in March, they lacked context or commentary from the designers.
The Holographic Processing Unit (HPU) sits at the heart of the device. It takes inputs from sensors on the Hololens (including a miniature Kinect depth sensor, which can track gestures at 1m and map rooms at longer distances, and an inertia measurement unit). The HPU also accelerates algorithms that track the user’s environment, movement and gestures, and displays AR images.
Essentially, the HPU is a highly customised digital signal processing (DSP) array, which runs at less than 10W. It contains a number of cores from US-based Tensilica, which are optimised to run the Hololens-specific instructions. Each typically has its own unique organisation of related memory units.
Ilan Spillinger, VP of HoloLens and silicon at Microsoft, said that the HPU accelerates “new style algorithms that need special local memories and a unique memory architecture, not a typical level 1-2-3 cache.”
Hololens uses a 14nm Intel Cherry Trail processor, with embedded graphics, running Windows 10. The dual-sided motherboard uses 64GB of flash storage and 2GB of external memory, split between the HPU and SoC.
The optics subassembly includes an inertial measurement unit, which is tailored to meet very specific requirements. The optics use gratings developed and manufactured by Microsoft. They support a wide inter-pupil distance and adjust for users wearing contact lenses or glasses. The display itself is an LCoS type.
I got to try the Hololens at SID and although it is clearly too big and heavy at the moment, the performance is excellent, especially in the way that the virtual world is mapped to the real world. It clearly sets a benchmark in terms of performance for consumer devices. (BR)