How Tensor SoC for Pixel matters to Google and AI

Google Tensor SoC processors will run on Pixel 6 and Pixel 6 Pro smartphones coming later this year, Google said in a Monday blog. The announcement led  to diverse industry reactions: one analyst predicted Google Tensor could be licensed to other smartphone OEMs.

For background, Google Pixel has not so far gained broad customer support, hitting less than 1% of the global market in any quarter since it debuted in 2016. That apparently was  OK by Google at first, which initially sold the phone as a “pure” Android device devoid of clutter and add-ons seen in phone by other Android makers.

The U.S. has been the biggest Pixel market with no higher than 3% share, according to IDC, which marked 2019 as the first year that Google seriously ramped up sales to 20 or so countries.

The use of Tensor in coming Pixels won’t necessarily hurt Qualcomm, according to analyst Patrick Moorhead at Moor Insights & Strategy, as Qualcomm has supplied processors for Pixel and nearly every other Android phone. While Google hasn’t officially disclosed so, he believes Google is using an Exynos CPU for the Tensor SoC to lower its overall costs.

 Samsung and Apple already make their own processors for their branded phones. Samsung is the top smartphone maker globally, and Xiaomi came in second last quarter, with Apple third, according to a recent IDG tally.

RELATED: Smartphones are surging again but are still behind 2015

Android, in various forms, runs in about 85% of the globe’s smartphones, while iOS on iPhone runs in nearly all the remaining 15%.

Ambitions beyond Pixel

Could Google have ambitions beyond Pixel with Tensor chips? Some analysts think so.

“Owning an SoC [like Tensor] doesn’t necessary equate to [smartphone] sales,” said IDC analyst Ryan Reith via email. “But if you look at what others have done in the past who have their own SoC such as Samsung, Apple and Huawei, all of those companies have been market leaders at the high end of the smartphone space. Time will tell if Google can make the next Pixel more successful than its predecessors, but I believe it takes a lot more than its own SoC to do so.”

What will matter more is if Google puts solid marketing into the next round of Pixel, Reith believes. “Consumers need to understand why they are paying a premium and the average consumer doesn’t understand or perhaps care about speeds and feeds,” he added.

Instead, Reith believes that the longtail effect of Tensor is what matters,  for both other Android smartphones and Chromebooks.  “Owning their own silicon is a more important puzzle piece to Google’s future than it is the next round of Pixels,” Reith said. “Tensor is a good runway to other ventures down the road, meaning the ability to license the SoC to other OEMs that are more proven at shipping large volumes of smartphones, PC’s and tablets…Google would have to make it compelling enough for others to license from them, which is not impossible but not the easiest task.”

Leonard Lee, managing director of NeXt Curve, said Tensor on Google’s smartphones “will mean feature differentiation starting with computation photography, butit will undoubtedly be just the beginning. It’s a seemingly logical reaction to the significant advancement Apple’s M1 represents for mobile silicon out of the smartphone device category.”

“It is easy to imagine that Google has interest in driving current and future smartphone silicon innovations into their Pixel Chromebooks, tablets, constrained edge computing environments and eventually into their data center where efficiency concerns and the need for application-specific compute continue to grow unabated,” Lee added.

Lee  said Google could also face antitrust claims if it tries to sell Tensor to other Android device makers.

Google has been building up its Tensor tech for cloud artificial intelligence (AI) and machine learning (ML) for at least a couple of years. “I suspect the cloud chips and phone chips are very different, but I suspect the core technology is very similar and tuned to the needs of Google algorithms and frameworks, especially TensorFlow and Tensor Flow light,” said Jack Gold, an analyst at J. Gold Associates.

Original Tensor chips for Google Cloud “were supposed to be optimized accelerators that operated at a better power efficiency than what was available in the market,” Gold added. “If you are running huge cloud data centers, power is a really big deal.”

Gold and others hit on a central point about Tensor SoCs for smartphones, which is about smartphones having the ability to process AI in a bespoke manner.  Google’s Rick Osterloh hit on this point when he remarked, “As more and more features are powered by AI and ML, it’s not simply about adding more computing resources, it’s about using ML to unlock specific experiences for our Pixel users.”

Tensor will be used in Pixel to run computational photography and will have a security core with Titan M2 for the most hardware security in any phone, Osterloh added.

Tensor will make Pixel phones the “fastest, smartest and most secure Pixel phones yet,” Google touted in an online promotion.

Gold said the AI capabilities of Tensor are key.  “Google is now saying that AI/ML functions are achieving a critical mass in smartphone designs for cameras, video processing, audio and speech recognition, and user interface and that it needs to make the Tensor capabilities a key part of smartphone designs,” Gold said.

Qualcomm also has an advanced Ai/ML processor embedded in its Snapdragon chips used widely in phones and Arm has generically added AI acceleration in its designs.

“This move by Google should not be seen as an un-endorsement of the Qualcomm chips they have used in the past,” Gold said. “Instead, it’s a positioning by Google that they have a specific vision of what AI can do in smartphones and elsewhere and they are building an optimized SoC for their purposes. It’s always more effective to build a specifically optimized accelerator for your own systems and code.

“This is an acknowledgement by Google, similar to Apple and Samsung, that AI/ML will be a core ingredient of smartphones and other mobile devices going forward, perhaps even more important than in PCs,” Gold said.

RELATED: Google’s Pixel 6 phones to run Tensor Soc, replacing Qualcomm chips