Yesterday, we got a really good presentation from ARM’s Physical IP division and we heard some incredible details about the company’s developments in that part of their business. We also heard things about their upcoming A9 replacement, the A12 as well as got to hear from one of their biggest partners, AllWinner. This session, however, was strictly about their physical IP and their relationships with the respective fabs that would be making those chips for their customers.
The presentation itself was filled with plenty of foundry information and different info about current and trailing process nodes and the improvements that ARM has made even on older process nodes. However, as things tend to be, the most interest is in the leading process nodes. And it just so happens that ARM is working closely with four different fabs to implement their FinFET manufacturing strategy. Interestingly enough, three of the four fabs are actually members of the common platform alliance, which honestly isn’t much of a surprise but is something to consider when looking at these fabs. Those fabs are IBM, Global Foundries and Samsung. With TSMC being the 800 pound gorilla, but not a member of the Common Platform Alliance.
ARM has worked closely with all of their fabs to test FinFETs on their different process nodes in order to make sure that they were prepared for those nodes. One of the first test chips was in fact an ARM A7 chip that they tested at Samsung, which they were able to successfully manufacture. This was done as a partnership between ARM, Samsung, and their Common partner, Cadence.
In addition to the A7 chip being fabbed on the 14nm FinFet process, ARM worked with TSMC to test the Cortex-A57 on TSMC’s 16nm FinFET process, which actually makes the A57 the first ARMv8 based chip to be produced on a 16nm FinFET process. Once again, ARM worked with TSMC and Cadence to achieve this goal, which shows how important their partnerships with these companies really are.
Moving on from there, ARM talked more about their current and trailing nodes. However, they got back to their leading nodes when they started to talk about R&D and production. As you can see n the slide below, they talked about 32/28nm as well as 20nm, 16/14nm and even 10nm. With the most interesting news coming from ARM that they had taped out 10nm FinFETs which I believe is something that even Intel hasn’t done yet (correct me if I’m wrong). This is definitely a pretty big deal for ARM and their partners, but in the grande scheme of things, it most likely means very little to nothing for the overall progression of new nodes. Especially when you consider that we still have to move to 2 or 3 nodes prior to reaching 10nm.
After that, as they wrapped up the presentation they said something that I found incredibly ballsy and lofty. The company claims that they have FinFET leadership. While I can totally understand that ARM is very excited about where they are with their fab partners and licensees, I think that the company is going to need to give us some serious explanation about why they are claiming the title of FinFET leadership. Last time I checked, ARM still isn’t shipping an FinFET products, while their competitor Intel has been shipping them in significant volumes for quite some time now. We have reached out to ARM for them to clarify this statement, but have yet to hear back.