You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
We already knew this of course since the Facebook domain is all over SemiWiki content. They are very interested in EDA, IP, AI, ASICs, Security, ARM, and FPGAs for what it is worth. Yet another fabless systems company to compete with the traditional chip companies. Get ready for another semiconductor industry transformation: IDM -> Fabless-> Fabless Systems Companies, absolutely.
Facebook Inc. is building a team to design its own semiconductors, adding to a trend among technology companies to supply themselves and lower their dependence on chipmakers such as Intel Corp. and Qualcomm Inc., according to job listings and people familiar with the matter.The social media company is seeking to hire a manager to build an “end-to-end SoC/ASIC, firmware and driver development organization,” according to a job listing on its corporate website, indicating the effort is still in its early stages.
A bit late to the party. Why do you think FB constitutes a transformation? Look at Apple, Baidu, Google, Alibaba, AWS, all those have actual chips. I would actually be surprised if the team at FB is really just starting now.
PyTorch (developed by FB) has gained quite a bit of momentum in the last year, and is now the favored python library for deep learning with many practitioners preferring it over TensorFlow. My guess is Facebook is planning to build on that success and build custom processors for PyTorch to compete with Google's TPUs for TensorFlow.
Again, late to the party. Take a look around, ML processors are sprouting everywhere. And, while PyTorch is nice, thinking it has wide market share is optimistic. Most groups have access to tools from NVidia (including at AWS) and Google, which also represent the silicon they can use. Those are the ones to catch. And since everyone sees $$$ in this, there are lots of entrants in the race.
PyTorch absolutely has and is continuing to gain marketshare on TensorFlow. There is a large scale shift happening right now in Deep Learning away from TF and towards PyTorch. They may have started a couple years behind, but I'd say FB has caught up to Google on the deep learning software side, and can easily catch up on the hardware side, since TPUs don't really have a ton of functionality quite yet.
