Google Tensor Chip: Pixel 6's Custom Design

Google's Tensor Chip and the Pixel 6 Series

Prior to their official release, the Google Pixel 6 and Pixel 6 Pro have been exceptionally well-documented. In a market often characterized by similar Android devices, Google has implemented distinctive design choices, particularly concerning the processors powering these phones.

For the first time, Google is offering a smartphone driven by a System on a Chip (SoC) developed internally. This new processor is named Tensor.

Introducing the Tensor SoC

Google initially discussed Tensor earlier this year. It serves as the engine for all the phone’s on-device artificial intelligence functionalities. Essentially, it’s Google’s dedicated AI/ML accelerator, integrated with standard Arm CPU and GPU cores, alongside the new Titan 2 security core.

Google asserts that Tensor delivers up to 80% improved performance compared to the chips found in the Pixel 5. While the Pixel 5 offered a smooth user experience, it was positioned as a mid-range device.

Preliminary benchmark results suggest Tensor’s performance is comparable to contemporary Snapdragon mobile chips from Qualcomm. However, these benchmarks don't fully account for Google’s specialized AI/ML cores.

The impact of these cores on the responsiveness of the Pixel 6’s cameras and its advanced computational photography capabilities cannot be accurately measured by conventional benchmarks.

Tensor's Core Architecture

Available information indicates that Tensor incorporates two Arm Cortex-X1 cores, representing Arm’s top-tier performance-focused mobile designs. In contrast, the Snapdragon 888 utilizes only one such core.

Like most modern SoCs, Tensor also includes lower-power cores designed for energy efficiency. Reports suggest a combination of older A76-based cores and the newer, highly efficient A55 cores contribute to the Pixel 6’s projected 24-hour battery life.

Google has deliberately refrained from disclosing specific details about these core configurations, prioritizing the emphasis on the system’s AI capabilities.

AI and Machine Learning Capabilities

The chip also features a low-power AI “Context Hub” which enables several always-on machine learning features.

Throughout today’s announcement, Google’s Rick Osterloh highlighted the importance of these AI-driven experiences, including live translations and the phone’s enhanced photo and video features.

According to Monika Gupta, Google’s senior director for Google Silicon, “With Google Tensor, we’re unlocking amazing new experiences that require state of the art ML — including Motion Mode, Face Unblur, Speech enhancement mode for videos and applying HDRnet to videos.”

Gupta further explained that Google Tensor allows for a more personalized smartphone experience, moving away from a generic hardware approach towards a device that intelligently adapts to individual user needs.

During the event, it was also emphasized that the chip has been under development for several years. All design decisions were made with the goal of optimizing the performance of the AI features.