The idea is that with this new chip inside the main “Holographic Processing Unit,” the next-generation HoloLens will be able to do things like recognize objects in the line of sight, but its battery won’t be exhausted too quickly, because the main chip won’t need to do the extra computing work.
“We really do need custom silicon to help power some of the scenarios and applications that we are building,” as Microsoft’s chief technology officer, Kevin Scott, told Bloomberg.
In the past few years, as AI has become trendy, many app makers have shunted the necessary underlying computations, like recognizing the words that people say, to servers in remote data centers. So chipmakers like Intel and Nvidia have begun adjusting their silicon accordingly, and Alphabet has even designed its own chips.
But for some applications, it’s not desirable or realistic to expect a robust internet connection that can provide a backbone for those connections from a local mobile device to a cloud. Earlier this year a Google executive said he expects mobile device makers to come up with new processors for AI. And there are indications that Apple is preparing to unveil an AI chip for the iPhone.
Now it turns out Microsoft is also working on optimizing AI on lower-powered devices. The HoloLens is by no means today a mass-market device — sales were said to be in the thousands. But whatever the company learns from the affair could surely be valuable for the development of future products — although they probably won’t be phones.
Microsoft worked with Cadence and Taiwan Semiconductor Manufacturing Co. on the original special processor for the HoloLens, which worked with a standard onboard Intel chip. Cadence, Microsoft and TSMC did not immediately respond to requests for comment.