Facebook is assembling a team to design its own semiconductors, joining other tech titans that have recently invested in building their own chips, according to job listings and sources who spoke with Bloomberg.
One job posting on Facebook’s corporate website highlighted by the news agency is for a manager to create an “end-to-end SoC/ASIC, firmware and driver development organization,” suggesting the initiative is still in the early stages.
It wasn’t immediately clear what specific application Facebook might have in mind for the chip; Facebook declined to comment.
By developing its own chip, the Menlo Park, California-based company would join other tech giants hoping to reduce their dependence on chipmakers such as Intel and Qualcomm.
Apple, for example, builds its own chips, which makes it better able to control the features it releases, as well as its timeline for introducing new devices. Apple now makes its own application processors that act as the brains of its mobile devices, a Bluetooth chip to quickly link its AirPods to its iPhones, and security chips that protect personal data and biometrics from hacking attempts.
Google developed chips in-house to process AI apps, boasting that they are 15 to 30 times faster than today’s conventional CPUs and GPUs. Unlike Apple, Google’s chips aren’t in your devices, but they are hard at work in Google’s data centers, delivering search results, identifying your friends in your photos, translating text, drafting Gmail message replies and screening out email spam.
Facebook chip job post details
The job descriptions fit into two categories: chips for artificial intelligence, also called machine learning, and chips for Facebook’s Oculus virtual reality headsets.
Some of the jobs are for Oculus-related development of SOC processors, the system-on-a-chip design popular in smartphones and other energy-efficient devices. SOCs pack modules for handling multiple jobs — processing, network communication, graphics, and other chores — onto a single slice of silicon.
One job is an SOC architect in Oculus’ computer vision team, but there are also several firmware engineer positions for writing low-level SOC software “to support development of state-of-the-art SOC with dedicated hardware accelerators (HWA) designed for computer vision and machine learning.”
AI is a hot area in tech, and Facebook is a major player, using it for things like face recognition for photos posted to the social network. But with processor progress slowing, some companies are turning to custom processors called ASICs (application-specific integrated circuits) or to adaptable chips called FPGAs (field programmable gate arrays) that can be tuned to speed up specific jobs. Google is taking the ASIC approach with its tensor processing unit for AI, now in its second generation, while Microsoft’s AI program is employing FPGAs. That’s for use in data centers packed with servers, but ASICs also are widely used in smaller devices.
Facebook at this stage looks interested in both approaches, judging by one job posting for an ASIC and FPGA designer. The job posting includes AI and machine learning as one duty, but FPGAs also are useful for simulating new chips, including ASICs in development.
Another job post, for a managerial position at Facebook, is for managing both SOC and ASIC work.
