Why It's a Good Idea for Google to Make Its Own Phone Chip

The best reason to buy a future Pixel phone might be Google’s custom system-on-chip.

|
Jun 14 2017, 4:30pm

Image: Google

A new report in Variety says that Google has poached an Apple veteran to work on its own SoC (system-on-chip). Sure enough, Manu Gulati's LinkedIn profile shows that, after being a chip architect at Apple for over seven years, he's now a Lead SoC Architect at Google.

This isn't the first time we've caught wind that Google might be developing its own mobile SoC for use in future devices. In fact, Google's own job postings are full of clues: Mobile SoC Architect, Mobile SoC Memory Architect, Mobile SoC CPU Architect, SoC Performance Architect… the list goes on and on. It seems obvious that Google is building its own SoC for mobile devices. It's a move that makes sense. Google now makes phones (the Pixel and Pixel XL), a tablet (the Pixel C), Chromecast streaming devices, the Google Home (its Amazon Echo-like "smart speaker"), and has produced Chromebook laptops in the past. Let's not forget about its sister company Nest and all the smart home gadgets it makes.

To be clear, Google already makes processors. Last year it debuted its TPU (Tensor Processing Unit), a card for cloud servers that is meant to help accelerate machine learning and AI. This year it announced a much faster and more flexible second-generation version. But these are nothing like mobile SoCs.

Google's new second-generation TPU for cloud computing. Image: Google

Mobile SoCs are a combination of many parts, all working in harmony to put an entire computer in your pocket. There's the general-purpose CPU cores, the GPU (graphics processing unit), the ISP (image signal processor, which processes data from the camera), the media encoder and decoder, the motion coprocessor that manages data from a wide variety of sensors, input-output interfaces, radios (cellular, Wi-Fi, Bluetooth), and more. It's all packaged together into a single chip with many layers, often with a hefty chunk of RAM. Making an SoC that is fast, full-featured, and power-efficient enough for a premium phone is one of those absurdly complicated tasks that only a handful of giant corporations can do (namely Apple, Qualcomm, and Samsung).

So why would Google want to take this on for its own Pixel phones (and, presumably, other mobile devices like tablets, Chromebooks, and watches)? For the same reasons Apple did over seven years ago.

It's all about control and cost

One of the biggest problems surrounding Android phones is that they're slow to receive OS updates, and you're lucky if your phone is updated for even two years. By contrast, Apple updates its devices for around five years. This is, in part, the fault of the SoC makers like Qualcomm. They don't want to write drivers and OS libraries for old processors they no longer sell; there's just no profit in it. But if you sell not just the chip but the whole device that uses it, if old devices are using the services you profit from, the whole cost-benefit equation changes. Suddenly, guaranteeing a long lifespan for your devices becomes a major selling point. Building its own SoC makes Google the one responsible for keeping drivers and libraries current with the latest Android OS, so it's easier to update old Pixel phones to the latest version of Android.

It also means Google gets to build the OS and silicon together, optimizing them according to its own vision of our mobile future. It can design the image signal processor (ISP) with exactly the right characteristics to really soup-up the computational photography used by the Pixel camera. It can incorporate a GPU aimed specifically at speeding up Daydream VR. It can build in a tiny mobile version of its TPU to speed up on-device machine learning functions and make them more battery-efficient.

Google launched the Pixel phone by saying it was meant to showcase "the best of Google." But it's difficult to make Google's technology run better when using an SoC from Qualcomm, which makes processors that have to meet the wide and varied needs of nearly every different phone maker except Apple. Making a phone that truly runs Google's technologies better than any other requires hardware optimized specifically for them.

Making a world-class mobile SoC is outrageously hard, to the point where even Intel has had a tough time of it. Then there are minefields like LTE patents to step through (which are currently making it impossible for Apple to avoid buying LTE radio chips from Qualcomm and Intel). We have no idea if Google will succeed in producing something as great as Apple, Qualcomm, or Samsung, but I doubt we'll find out soon. Building a top-tier SoC is a years-long effort that won't be ready in time for the Pixel successor this fall and maybe not in next year's, either. But it's the right move for a company that is increasingly making more consumer hardware.

Get six of our favorite Motherboard stories every day by signing up for our newsletter .