Google's Tensor, And The Reason It Is One Of Google's 'Biggest Mobile Hardware Innovation'

Computers are smart, simply because they can process information in a way that is much faster than a human brain.

But the limitation is that, computers need to be programmed to do what they need to do. And computers cannot go beyond what is coded inside its algorithms. To make computers "smarter" in a way that they can think beyond their original programming, an AI is needed. With AI, computers can learn, and understand more things that what it's creators may not have thought.

And with the advancing of hardware, the resource-intensive AI processing can be squeezed down to smaller and smaller dimensions, without sacrificing unnecessary performance.

Google is among those companies that develop, promote, and extensively use AI in its various products.

With billions of users from around the world with different intentions, needs and habits, Google as a global company with a global presence, needs to ensure that its products can suit most if not all of its users' requirements.

And to help it accomplish that, Google uses AI and various other neural networks.

In 2021, Google announced the Pixel 6 phones, alongside the so-called 'Google Tensor'.

In a blog post, Google said that it built Tensor based on "how people use their phones today and how people will use them in the future."

And Google is calling it "biggest mobile hardware innovation in the history of the company.”

Google Tensor

The chip that is a result of 4 years of development, is smaller than the size of a paperclip.

According to Sundar Pichai in a tweet, the chip "builds off of our 2 decades of computing experience and it’s our biggest innovation."

As a start, Tensor is Google's first custom-built System on a Chip (SoC).

Google’s stated that its goal for building Tensor is to push what’s possible on smartphones, by bringing "AI breakthroughs directly to Pixel" and drive its vision of always-available technology.

Tensor was originally developed from Google’s hardware division believing that AI-backed smart features are how it can differentiate Pixel against competitors, while Google considers phones the "central control device of an ambient system.”

Instead of building a single co-processor to boost its AI models, like it once planned, Google opted to develop an entire chip instead, and design that chip to be optimized to its likings.

In other words, the Tensor chip is specifically designed to offer Google’s latest advances in AI directly, on mobile devices.

"We approached Tensor differently. Every aspect of Tensor was designed and optimized to run Google's ML models," Google Product Manager Monika Gupta said during Pixel 6 launch event. "This permeates our entire chip. We're fortunate to have great insights when it comes to ML, and built our chip based on where ML is heading, not where they are today."

As part of the presentation, Gupta said that mobile chips have not always been able to keep up with Google's advancements.

This is why Tensor exists in the first place.

Google created it to create new experiences it couldn't provide through previous third-party chips.

Another thing, Google built Tensor not with primarily win benchmarking scores.

According to Gupta, "peak CPU and GPU speeds look great in benchmarks, but they don't always reflect real-world user experiences."

Tensor as the first SoC Google built, consists of multiple parts, each with different purposes. For example, there are two main cores, with seven individual components.

At the Tensor Processing Unit (TPU), Google placed its AI ASIC, lies Google's integrated ML engine. It is custom designed by Google Research, for Google Research, said the company.

After that, there is what Google calls the Image Signal Processor (ISP). This component that uses "key algorithms directly in the silicon for power efficiency," is an accelerator capable of running the HDRNet algorithm to better and efficiently process Live HDR+ Video at 4K 60FPS.

The component also allows other computational photography and video features like Motion Mode in Google Camera. Action Pan blurs the background, while Long Exposure works on the subject.

It is also designed to run features like Google's Magic Erase, which provides a similar function to Photoshop AI-powered masking, improved camera features with the ability to fix blurry photos, and more.

The next component Context Hub to shift certain workload to the ultra-low power domain. This particular component powers the always-on display (AOD), Now Playing, and other "ambient experiences" to “run all the time, without draining your battery.”

On the security side, Tensor has a security core baked straight to it. The core is a CPU-based subsystem that is isolated from the application processor.

This particular chip is designed to process sensitive tasks and controls, and works with the dedicated Titan M2 security chip, which is not part of Tensor but Google touts as being resilient to advanced attacks like electromagnetic analysis, voltage glitching, and laser fault injection.

Tensor is also designed facilitate more accurate speech recognition, translate speech and text, and so forth.

As explained by Google, Assistant on Tensor is using the “most advanced speech recognition model ever released by Google” at, again, half the power. The high-quality ASR (automatic speech recognition) model is used to transcribe voice commands, as well as in long-running applications like Recorder and Live Caption “without quickly draining the battery.”

Furthermore, Google said that Tensor enables Live Translate to work on media like videos using on-device speech and translation models.

Google Tensor

All these components work together in conjunction to make up Tensor, with Google prioritizing "total performance and efficiency."

SoC consists of different parts. As chips become smaller, researches managed to cramp more things into a dimension that is smaller and smaller.

As a result, a phones' processor can become increasingly complex..

This is called heterogeneous computing.

And using the Google-branded Tensor, Google wants to bring Google's advances in AI to help excel heterogeneous computing tasks. And this is done by making the various parts of the SoC to work together through system-level decisions.

From the CPU, GPU, ISP, and its TPU, can be used when needed to accomplish certain task, without sacrificing too much resources.

"We ensured different subsystems inside Tensor work really well together, rather than optimizing individual elements for peak speeds."

"Google Tensor allows us to push the limits of helpfulness in a smartphone, taking it from a one-size-fits-all piece of hardware into a device that’s intelligent enough to respect and accommodate the different ways we use our phones," Google explained.

While Google said that Tensor is not originally built with performance in mind, the company claimed that the Tensor-powered Pixel 6 delivers an 80% gain in CPU performance if compared to the Pixel 5, and a massive 370% gain in GPU performance.

And on top of them all, by designing its own chip, Google can rely less on third-parties, spend less money by investing for the future, and creates a hardware foundation that can be further developed for years to come, all to Google's likings.

Read: Google Uses AI To Design Computer Chips, Because AI Is Faster Than Humans