Is Facebook jealous of Apple and starting to design its own chips?
To lead into the idea of chip design, Crothers points out that Facebook already makes its own servers. I think many people (including unfortunately a good chunk of tech writers) think that specing out a server platform is a complex technical problem, akin to some type of real design exercise like designing a chip. It’s not. In fact, building a server is not significantly more complicated than building your own PC from standard components and assembling it. That’s a task many embark on, often people without much electronics or engineering background. Mostly, it takes just motivation and access to the internet for tips. The fact is, it’s quite easy to go beyond the PC platform and even order up a tablet or cell phone from a basic reference design if you can front the cash for an order from a Chinese contract manufacturer. They are happy to pop your logo right on there and probably have a cousin up the road who can make those too.
Even with off-the-shelf IP blocks available for just about every type of function readily available at big silicon foundries, we are still not in the era where you could just tick these off from a menu and wait for the Fedex driver to show up with packaged devices.
Now I would have less issue with musings about Facebook designing its own chip if there had not been a comparison to Apple and its foray into integrated circuit development. In fairness, CNET’s Crothers did admit that chip design was “a more ambitious undertaking than building a server.” But let’s not forget that Apple is fundamentally a hardware company, despite the appeal of iOS and other user interfaces. As pointed out in an ongoing analysis of their applications processor design, Apple is becoming credible as a chip company. Apple deserves a lot of credit for ramping up its chip design as fast as it did, but it did not happen overnight. On the other hand, I hesitate to describe Facebook as even a software company. To me, they just have a web interface to a database. So, they program the database and the web, perhaps the odd app for Android or Apple.
To be more critical of the CNET post, it ran an image of a real silicon IC. Was it meant to confuse? I can’t say, but the effect is doubtless the same. Some people are going to think the quad core microprocessor graphic that ran in the article is Facebook’s invention. The caption says merely, “Chip die.” CNET editors probably should have specified the source for this image, rather than leaving the description in the link. The high aspect ratio layout with four cores in single file straddling the SRAM cache with GPU at the end of the die is Intel’s Sandy Bridge design. (Ivy Bridge is the architecture migrated to 22 nm process technology in the tick half of Intel’s famous tick-tock development cycle.)
I think there is a parable here if there is actually more to the story than I give Crothers credit for. Facebook and Zuckerberg would not be the first to experience wild success in an enterprise who then think they have all the answers or the CEO magic to create successes in other realms. I think the air becomes rarified when sitting on mountains of cash or receiving rock star media attention. They would not be the first company to be taken over by this kind of arrogance. Fooling beefy twin rowers might be easy. Building a design team and then a working chip, requires more finesse. You better have a real need. It’s not a hobby, and it doesn’t quite work the same way as an unstable beta software release with a few glitches. “Recoding” a design is going to take more cash and a lot more time in re-spin through the IC foundry.
Crothers speculates that perhaps the Facebook IC would be designed for a future tablet or smartphone. Let’s not forget that Apple didn’t design chips in advance of or parallel with its computer platforms. It started with the iPhone and iPad using off-the-shelf devices until it made sense to go beyond what mainstream chip vendors were offering.
But hey, if Mark Zuckerberg thinks chip design is his company’s next step, I have an idea for what to call the chip. Facebook becomes “FBIC.” And that name will be appropriate for more than just its similar pronunciation. I can bet that Facebook’s first silicon will be in dire need of circuit editing to fix errors, something that can be done on completed devices using focused ion beam (FIB) technology. My friends at Fibics can definitely help with that. Just ask for Pam.
Before I left for work this morning, I asked the closest Facebook power user at hand, my wife, “Did you know that Facebook is designing chips?” She responded, “What for?” Precisely.