Apple develops silicon by itself and for itself, and the latest iPhones featured lots of its handiwork. On Tuesday, the company boasted of its A11 processor, a slab of silicon with 4.3 billion transistors.
The chip contains six CPU cores and three GPU cores, which Apple built for the three iPhone 8 models announced at its new headquarters in Silicon Valley. Apple has arranged for the new chip, also called the Bionic, to take advantage of the advanced camera and crown of sensors in the premium iPhone X.
The company said that the CPU contains two performance cores 25% faster and four efficiency cores 70% faster than its previous generation of A10 processors. It is also 70% faster at multithreading, in which a computer chip chews on small sets of code at different stages of execution.
The custom GPU runs around 30% faster than the A10, the company said. It musters the same graphics performance while using half the battery power to organize images, compose photos, and enhance augmented reality apps, which overlay digital images on scenes seen through the smartphone.
Phil Schiller, Apple’s senior vice president of worldwide marketing, said the “neural engine” embedded in the silicon enables the iPhone X to recognize faces. The Face ID program reads faces to unlock the screen and authorize mobile payments. It also runs software that normally needs to be trained in data centers and then recalled from the cloud.
These algorithms boil inside the neural engine, which uses two processor cores to perform 600 billion operations per second without burning through battery life. The silicon is a product of Apple’s chip division led by Johny Srouji, senior vice president of hardware technologies.
Using custom silicon for machine learning, a personal assistant like Siri could answer questions faster and make augmented reality games more responsive. Keeping data on the smartphone preserves privacy and save battery life that would be used to wirelessly connect to the cloud.
In June, the company released a set of programming tools called CoreML, which helps developers infuse applications with machine learning code. The tools could be a natural fit with the Bionic chip, which will run “specific machine learning algorithms,” including Face ID and another feature called Animoji, which analyze facial expression and recreate them on animated pandas, robots, and unicorns.
Other companies are also trying to inject machine learning into gadgets. Huawei has built what it calls a neural processing unit to sort photos, translate text, and understand speech in smartphones. Qualcomm, as a stepping stone to its own custom silicon, released tools that inject apps with machine learning.
Arm is also rethinking its basic chip technology used as a starting point for Apple and Qualcomm. The technology, called Dynamiq, allows them to create chips with separate cores for tasks like speech recognition while reserving others for low power. It can also assign the ideal cores for particular applications.
The neural engine is only one of Apple’s unique silicon etchings. For the new iPhones, it also built an image processor and video encoder from scratch, so that the cameras better estimate lighting conditions and autofocus faster in low light, as well as provide motion analysis to optimize videos.
Apple revealed the custom GPU after it vowed in April to phase out graphics licensed from Imagination Technologies in its smartphones, watches, and tablets within the next two years. Imagination tried to curb the financial fallout by putting other business units up for sale. But it eventually gave up.
Imagination is now trying to sell itself completely and Apple was viewed as an unlikely but possible destination. The company had used Imagination’s graphics since the first iPhone came out a decade ago, and it poached engineers from the firm’s ranks. But after Tuesday’s announcement, an acquisition seems unlikely.
Developing silicon is nothing new for Apple, one of the world’s largest chip makers due to its smartphone sales. It has been speculated that its chip ambitions are why Apple’s research and development bills have doubled to $11 billion since 2013. The A10 can rival Intel’s desktop processor in single-core benchmark tests.
Updated September 15th, 2017: This article was updated with additional information about the neural engine embedded in Apple's latest processor. The title has also been updated to reflect the changes.