Pixel 6 Will Ship With Tensor, Google’s First-Ever Custom-Built SoC

4 years in the making and smaller than a paperclip.

After a myriad of rumors and speculation, Google has finally unveiled its in-house custom processor for its imminent flagship Pixel device. While the rumors suggested that it was codenamed “Whitechapel”, that name did not carry through to the other side. Today we know that Google’s first custom SoC is called “Tensor” and it debuts on the upcoming Pixel 6.

Source: Google

Via a series of tweets, Google slowly panned back the curtain on not only Pixel 6, but also the Tensor SoC that powers it. Cleverly though, Google deviated from talking about on-paper specs like the clock speeds, process node, core count, etc, and instead focused on more layman terminology to market the chip.

Alright, “layman” may be a bit too harsh, Google chose to purposefully talk about AI (Artificial Intelligence) and machine learning (ML) to flaunt their chip. They mentioned real-word use cases where Tensor’s power will come in handy and improve your user experience. This kind of approach is very similar to how Apple does their marketing, where instead of giving you the nitty-gritty, they tell you how it can actually impact you in the real world.

Apple’s A12 Bionic chip in a reveal event Source: Apple

Google didn’t even tell us who manufactured the chip, but reports suggest it was Samsung. Being an SoC means that the chip houses both a CPU and a GPU but, once again, we know nothing about those. We do know one thing, however, and that is that Tensor will have a dedicated security chip at its core (no pun intended) named “Titan M2”, this will supposedly help fortify the security of Pixel 6. Google highlighted that Pixel 6 has the most layers of hardware security out of any phone and that is due to having the Titan M2 chip on-board Tensor.

And, that’s it. We don’t really know anything else. There are leaked specs, obviously, but we have to take those with a grain of salt. Google is so far touting Tensor’s AI and machine learning prowess significantly and making it the spotlight of not only their custom silicone but also their upcoming Pixel 6. 

What We Do Know

We did learn a couple of things, though, thanks to the chain of tweets that revealed Tensor to the world. There’s also a blogpost from Google, but that tells us even less than the tweets do. Starting from the top, Tensor enables even better speech recognition and camera post-processing in the Pixel 6 thanks to its, you guessed it, machine learning and AI chops.

According to Google, Tensor allows Pixel 6 to make “big leaps” in voice commands, translation, captioning, and dictation. All these are fields where Google already shines and has an upper hand over the likes of Apple, Amazon, and Microsoft, so it’ll be interesting to see how much of a jump in performance Tensor brings to all of these.  

Information from Media Outlets

The name “Tensor” might ring a few bells and that’s because it’s also the name of Google’s open-source platform called TensorFlow. Google disclosed a bunch of other crucial details to media publications as well that they didn’t outline in their tweets or blog. The first and perhaps the most obvious of those is that Tensor is an ARM chip, and it’s based on the company’s Tensor Processing Unit (TPU). 

More information from the publications tell us that Google has also rebuilt the microarchitecture of the chip to allow for better RAM utilization as now the memory can access the CPU more easily. This supposedly allows for better multi-tasking as more apps and processes can run simultaneously without the phone overheating or running out of processing power.

Apart from having the CPU, GPU, a 5G modem, and the Titan M2 security chip on Tensor, Google has also added dedicated co-processors to help with AI tasks. Stuff like real-time speech recognition can be made faster and more accurate thanks to the AI chip doing all the heavy-lifting, and taking the load off the CPU and GPU. You can look at this similarly to how Apple includes the “Neural Engine” on its A-series processors for more efficient AI.

Moreover, a bunch of Google’s processing algorithms have been baked into the silicone on a hardware level, enabling them to do things that were simply impossible before. And, that seems to be the gist of it, right? Google has always wanted to do so much with its Pixel line of phones but third-party hardware (more specifically silicone) has kept them constrained. We already know that Google is not fully satisfied with Qualcomm’s SoC offerings as they limit what Google can and what Google wants to do with image processing.

In the demo given to publications, Google officials showed video captured from the Pixel 6 and compared it to last year’s Pixel 5 and even the iPhone 12 Pro Max. They mentioned that things like Google’s real-time image-processing can’t be applied to video on a Snapdragon SoC, and an in-house custom chip finally unlocks this barrier. Not only that, but advancements made in speech recognition and real-time captioning by Google can also not be supported by Qualcomm’s chips, at least to the extent Google wants.

Why The Image Processing Stuff Matters

  • Google’s Pixel phones exhibit industry-leading photography despite having camera setups on the lower end of the spectrum. Compared to the competition’s 100 megapixel sensors and 5-camera setups, the Pixel should be getting eviscerated in comparisons due to its measly on-paper specs, yet it always pulls ahead and produce the best results. That’s enough to tell you how much computational photography plays a role into Google’s image-processing pipeline.
  • Google has gotten so good at it that it can produce results that counter the apex of camera hardware in phones, via their software. One place Google has struggled from the beginning, though, is video and Apple has always exceeded in that area. With the way things are looking right now, the switch to Tensor may finally allow Google to close this gap and inch over Apple’s lead in mobile videography.

Lastly, Google CEO Sundar Pichai said that Tensor was 4 years in the making and that it’s Google’s biggest innovation to date, building upon its over 2 decades of experience and expertise in the computing industry. On a side note, that paperclip looks gigantic next to Tensor.

https://twitter.com/sundarpichai/status/1422228336533676035

It will be very interesting to monitor this launch. Pixel 6 will likely get a proper, fully-fledged reveal in October at its own event, where we also hope to get more info on Tensor. Perhaps, an in-house chip is exactly what Google has needed for all these years to escape the confines of the midrange and battle it out in the big leagues. We already see how Apple benefits from having their own custom-built hardware and software, so it stands to reason that Google can replicate the same eloquence with Tensor and Pixel 6.

Huzaifa Haroon
Born and raised around computers, Huzaifa is an avid gamer and a Windows enthusiast. When he's not solving the mysteries of technology, you can find him writing about operating systems, striving to inform the curious.