Microsoft Research is hard at work developing HoloLens 2. And continue to be the world’s first–and still only fully self-contained holographic computer. The self-contained nature of HoloLens means it needs to run on the battery Microsoft incorporates into the headset. So efficiency is key.
To improve HoloLens 2’s capabilities while retaining that efficiency, Microsoft decided it needed to design its own artificial intelligence chip. This AI coprocessor will form part of the next custom multiprocessor that powers HoloLens called the Holographic Processing Unit (HPU).
Deep Neural Networks (DNN)
Microsoft’s desire to keep HoloLens self-contained means relying on the cloud isn’t acceptable. So it’s replacing that need with a dedicated AI chip in HPU 2.0.
In order for HoloLens to work, it needs to quickly and accurately recognize objects in the world around it. That requires Deep Neural Networks (DNN), which right now work best in the cloud running on custom processor/memory architectures.
Moving AI to run locally not only increases efficiency. It also allows for higher performance, as there’s no need to pull data from the internet. HoloLens 2 will be able to implement its own DNNs. Developers will have full access to program the AI chip.
For end users, it should mean an obvious improvement in the computer vision capabilities on offer by HoloLens 2, which translates to better experiences.
We have no idea when HoloLens 2 will launch or how much it will cost. The first version only went out to developers in March 2016, with a consumer version expected at some point this year after pre-orders opened last October.