It sounds like this custom Google SoC-powered Pixel is really going to happen. Echoing reports from about a year ago, 9to5Google is reporting that the Pixel 6 is expected to ship with Google’s custom “Whitechapel” SoC instead of a Qualcomm Snapdragon chip.
The report says “Google refers to this chip as ‘GS101,’ with ‘GS’ potentially being short for ‘Google Silicon.'” It also notes that chip will be shared across the two Google phones that are currently in development, the Pixel 6 and something like a “Pixel 5a 5G.” 9to5 says it has viewed documentation that points to Samsung’s SLSI division (Team Exynos) being involved, which lines up with the earlier report from Axios saying the chip is “designed in cooperation with Samsung” and should be built on Samsung’s 5nm foundry lines. 9to5Google says the chip “will have some commonalities with Samsung Exynos, including software components.”
XDA Developers says it can corroborate the report, saying, “According to our source, it seems the SoC will feature a 3 cluster setup with a TPU (Tensor Processing Unit). Google also refers to its next Pixel devices as ‘dauntless-equipped phones,’ which we believe refers to them having an integrated Titan M security chip (code-named ‘Citadel’).” A “3 cluster setup” would be something like how the Snapdragon 888 works, which has three CPU core sizes: a single large ARM X1 core for big single-threaded workloads, three medium Cortex A78 cores for multicore work, and four Cortex A55 cores for background work.
The Pixel 6 should be out sometime in Q4 2021, and Pixel phones always heavily, heavily leak before they launch. So I’m sure we’ll see more of this thing soon.
Reasonable expectations from Whitechapel
It’s easy to get overhyped about Google’s first in-house smartphone SoC—”Google is ready to take on Apple!” the headlines will no-doubt scream. The fact of the matter, though, is that Apple is a $ 2 trillion hardware company, and the iPhone is its biggest product, while Google is an advertising company with a hardware division as a small side project. Whitechapel will give Google more control over its smartphone hardware, but Google’s custom chips in the past have not exactly set the world on fire, and therefore it’s reasonable to temper expectations for the company’s first-generation SoC.
Google’s consumer hardware team has already shipped several custom chips, and I don’t know if you could call any of them world-beaters:
- The Pixel Visual Core in the Pixel 2 and 3 was a custom camera co-processor created with the help of Intel. The Visual Core helped with HDR+ processing, but Google was able to accomplish the same image quality on the Pixel 3a, which didn’t have the chip.
- The Pixel Neural Core in the Pixel 4 was spun out of the company’s Tensor Processing Unit (TPU) AI accelerator efforts and had a similar job doing camera and AI voice recognition work. It was unimportant enough to just cut from the Pixel 5 entirely.
- There was the air-gesture detection chip, Project Soli, on the Pixel 4. This was a radar-on-a-chip concept that Google originally pitched as capable of detecting “sub millimeter motions of your fingers,” but by the time it was commercialized, it could only detect big, arm-waving gestures. The feature still exists today in the new Nest Hub, for sleep tracking, but it was not good enough to make the jump to the Pixel 5.
- The company’s Titan M Security Chip works as the secure element in some Pixel phones. Google says this makes the Pixel phones more secure, though a roughly equivalent secure element also comes with a Qualcomm chip, or at least, the company has never demonstrated a tangible difference.
I think the biggest benefit we’ll see from a Google SoC is an expanded update timeline. Android updates go a lot smoother when you get support from the SoC manufacturer, but Qualcomm abandons all its chips after the three-year mark for major updates. This lack of support makes updates significantly harder than they need to be, and today that’s where Google draws the line at updates. With Qualcomm out of the way, there are no excuses for Google to not match Apple’s five-year iPhone update policy. With a custom SoC, Google will totally control how long it can update devices.
Currently, Google is in the embarrassing position of offering less support for its devices than Samsung, which is now up to three years of major updates (Qualcomm’s maximum) and four years of security updates, while Google only offers one year less of security updates. It’s a weird position for Google to be in, which previously was leading the ecosystem in hardware support. Maybe Google didn’t immediately match Samsung because it’s waiting for the Pixel 6 launch, where it will announce dramatically longer support timelines thanks to its own chip?
Actually competing in the SoC business is tough
Beyond easier updates, I don’t know that we can expect much from Whitechapel. Lots of Android manufacturers made their own chips now, with varying levels of success. Samsung has the Exynos line. Huawei has its HiSilicon chips. Xiaomi made the Surge S1 SoC back in 2017, recently launched the Surge C1 camera chip in the Xiaomi Mi Mix Fold, and it has an investment in a silicon designer. Oppo is working on developing in-house chips, too. None of the existing efforts has been able to significantly beat Qualcomm, and most of these companies (other than Huawei) still choose Qualcomm over their own chips for important devices. Everyone, even Qualcomm, is relying on the same company, ARM, for its CPU designs, so there’s not much room for difference between them. When everyone is using off-the-shelf ARM CPU designs the major areas of differentiation left are the GPU and modem, two areas Qualcomm excels at, so it gets picked up for most major devices.
The companies that take hardware seriously do their best to separate themselves from ARM’s baseline CPU designs, choosing instead to design their own cores based on the ARM instruction set. Apple dominates mobile CPU performance thanks to its acquisition of an entire semiconductor company, PA Semi, back in 2008. Qualcomm is doing its best to catch up, buying Nuvia, a chip-design company founded by some of those ex-Apple chip designers, and it plans to ship its internally designed CPUs in 2022. Google has made a few chip design hires, but those are split between the separate hardware and server teams, and they pale in comparison to buying an entire company. When even Qualcomm isn’t currently shipping custom chips, I don’t see any way Google uses anything over the off-the-shelf ARM CPU designs.
Google’s GPU and modem solutions will be an area of great interest. There aren’t a lot of GPU designs to go around. Qualcomm has its own Adreno division, which it purchased years ago from ATI. Samsung has a deal with AMD for its future GPUs, but I doubt that would be up for grabs in its Google partnership. If this chip is really Exynos-adjacent, Samsung and many other also-ran SoC vendors go with off-the-shelf ARM Mali GPUs, which are generally not competitive with what Qualcomm puts out. Samsung signed that AMD partnership for a reason!
Imagining Google’s SoC having an onboard modem is a challenge. You generally don’t get to integrate a modem into your SoC unless you own the modem design, and Google doesn’t own any modem IP. Samsung has produced chips with onboard 5G modems, but they generally don’t come to the US, so a Samsung modem would require both sharing the design to Google and bringing it to the US for the first time. Qualcomm is, of course, the king of strong-arming companies with its modem IP and keeping competitors out of the US, and it’s also generally a leader in modem technologies like 5G. Apple has managed up to now with separate cellular modems—today the iPhone 12 comes with a discreet Qualcomm modem for 5G, which is probably the most likely option for Google. Apple also bought Intel’s modem division for a billion dollars, indicating it’s working toward onboard modem tech.
Along with the usual CPU/GPU/modem options, Google could also include some camera and AI special sauce in the form of some kind of co-processor (hopefully we’ll also get the Pixel’s first camera sensor upgrade in four years). Google will also probably include a Titan security chip. Even if it did, I can’t imagine these making a huge difference compared to something like shipping with a low-quality GPU or modem. Google has never demonstrated a strong end-user benefit from its custom silicon in the past, just a whole lot of hype.
It’s hard to be bullish on Google’s SoC future when the company doesn’t seem to be making the big-money acquisitions and licensing deals that Apple, Qualcomm, and Samsung are making. But at least it’s a start.