GlobalFoundries develops 7 nm and 10 nm technologies in-house.
- Thread starter mhagmann
- Start date
Yes, here is the backstory:
Samsung licensed a copy exact version of 14nm to GF so Apple could have a second source to meet capacity requirements for the iPhone 6s. That did not work out well for Apple so they had to split the A9x between Samsung 14nm and TSMC 16nm which are not compatible processes.
For the iPhone 7, Apple will use a variant of TSMC 16FFC. From what I hear today Apple will also use TSMC 10nm and 7nm so there is no need for a Samsung second source.
I have spoken to GF about this directly and strongly suggested that they forgo 10nm as it will be a very short node and go directly to 7nm. IBM has 7nm running in the lab so this is a strong possibility. Scott Jones (SemiWiki) did an interview with Gary Patton (VP GF) at the imec conference last week confirming that this is in fact true, GF will skip 10nm HVM and move quickly to 7nm. Hopefully the interview will be published this week.
Samsung licensed a copy exact version of 14nm to GF so Apple could have a second source to meet capacity requirements for the iPhone 6s. That did not work out well for Apple so they had to split the A9x between Samsung 14nm and TSMC 16nm which are not compatible processes.
For the iPhone 7, Apple will use a variant of TSMC 16FFC. From what I hear today Apple will also use TSMC 10nm and 7nm so there is no need for a Samsung second source.
I have spoken to GF about this directly and strongly suggested that they forgo 10nm as it will be a very short node and go directly to 7nm. IBM has 7nm running in the lab so this is a strong possibility. Scott Jones (SemiWiki) did an interview with Gary Patton (VP GF) at the imec conference last week confirming that this is in fact true, GF will skip 10nm HVM and move quickly to 7nm. Hopefully the interview will be published this week.
What's GF's expected timing for 7nm?
As GlobalFoundries approaches the 7-nm node "without licensing technologies from outside" I would suggest that they consider a more fundamental approach to metrology. Publications in the last few years show that 1 nm resolution in carrier profiling is marginal with scanning spreading resistance microscopy (SSRM) and the diamond nanoprobes distort the lattice of the semiconductor where the measurements are made. A rule of thumb in the Roadmaps has been that the resolution should be finer than 10% of the dimension for the respective node, which cannot be satisfied using SSRM at the sub 10-nm nodes. Furthermore, the effects of Random Dopant Fluctuations (RDF) are more intense at these nodes. I appreciate the magnitude of the difficulties and challenges, but I consider using SSRM at the 7-, 5-, and 2-nm nodes to be analogous to attempting to manufacture a car using a wooden ruler having only 1-inch markings.
My guess is 2020 for HVM but that is just a guess. It really depends on EUV and I'm a bit pessimistic right now.
Scotten Jones
Moderator
My guess is Global Foundries (GF) will introduce 7nm in late 2017 early 2018 to match up to TSMC's timing. Gary Patton has been very clear that GF was late starting work on 14nm and 10nm and they are pushing really hard to not be late on 7nm.
The GF 7nm process is designed based on what can be reasonably achieved with MP with EUV as a possible option once it is available.
I should note here that TSMC and GF 7nm processes are really 10nm processes analogous to Intel's 10nm process due in the same time frame.