Facebook has hired one of Google’s lead chip developers, Shahriar Rabii, to help the social network in its ongoing effort to design its own silicon, according to a report from Bloomberg. Facebook hopped on the chip-developing bandwagon earlier this year, when they started to build a team that could design custom chips to power server and consumer hardware. Rabii’s new role at Facebook will be as a vice president and head of silicon, according to an updated Linkedin bio.
It’s a move that’s on trend with other tech giants, many of which are bringing chip design in-house rather than relying on big-name suppliers like Intel and Qualcomm. Apple has been creating its own custom processors for iOS devices for nearly a decade, and it has designed custom single-purpose chips for artificial intelligence and other tasks in recent years. The iPhone maker is also reportedly planning on using its own chips to replace the Intel processors for their Mac computers by 2020. Earlier this year, Amazon reportedly embarked on a new initiative to design its own chips, specifically to help power AI features for its Echo line of smart speakers.
Facebook may use custom chips for AI training or smart speakers
Google produces their own custom Visual Core chips for the Pixel smartphones — and Rabii, Facebook’s recent hire, had previously led the team that developed them. In his new role, Rabii isn’t likely to be developing chips for Facebook-branded smartphones, but the company is working on several types of hardware that could use a custom processor.
Earlier this year, Facebook-owned Oculus VR launched the standalone Oculus Go virtual reality headset that currently relies on a Qualcomm-branded chip. Future models may use custom Facebook chips instead. The company is also reportedly developing its own series of Echo Show-like smart speakers with AI features, and a custom chip may give Facebook a competitive advantage in the home.
The custom chips could also be used to better train the AI algorithms that Facebook has patrolling its site for hate speech, fake accounts, and potentially dangerous content. Right now, the company uses modified third-party GPUs from companies like Nvidia. Designing its own AI training servers with proprietary chips, as Google does with its Tensor Processing Units, could help with the very tricky problem of using AI instead of human eyes to police its ever-growing platform.