How is that for a cryptic title? It is GlobalFoundries’ rallying cry. Their big focus is to bring up 14nm during 2015 in fab 8 in Malta NY.
Last week I chatted to Shubhankar Basu. He is the senior product line manager for leading edge technologies. Not surprisingly, that means he is currently leading the 14nm FinFET business. He told me that they are well on track to achieving “14 in 15”.
It is almost exactly a year since GlobalFoundries announced that it was licensing Samsung’s 14nm process and would implement it in fab 8. The same process also runs in Samsung’s fab in Austin as well as in Korea. The two companies are running identical processes with a single PDK. It is possible to tapeout 14nm designs to either or both suppliers.
There are actually two 14nm processes, LPE (E is for early) and LPP. GF plans on this being a long-lasting node since they are seeing lots of demand. In fact they have kicked part of their technology development center out of the fab to create more space to enable a bigger and faster ramp. In the future there will also be additional 14nm derivatives based on the same underlying process.
GF is running just a single part in LPE. LPE completed qualification in January of this year and has reached the yield levels for a volume ramp of this product starting this month.
GF could not say who the customer for this single product is, but there are rumors all over the net that it is nVidia. But nVidia isn’t admitting to it either:When asked about this during a conference call with investors and financial analysts[in February], Jen-Hsun Huang, CEO of Nvidia, neither confirmed nor denied the plan.
The second process, LPP is also on-track. It has already reached yields as good as LPE for SRAM after less than 3 months. Qualification should be early in Q3 this year, with volume ramp in Q1 2016. They have been (and will continue) to run LPP shuttles. Almost all of these have been over-subscribed.
There is a core of common IP. GF have added some additional capabilities to their 14nm LPP offering:
- additional IP blocks
- 2.5D and 3D TSV-based packaging
- advanced DFM
- advanced EDA reference flows
There are many designs taping out this year in LPP. Note that these are “first source” designs where GF is the only fab (or is the first source). These designs are in mobility, and some high-performance computing large-die customers.
I asked Shubhankar about 10nm. He said that they were currently doing internal development and he couldn’t comment about any possible co-operation with Samsung at that node.
So I switched topic to IBM. Since the deal to “buy” IBM’s semiconductor business hasn’t officially closed, there is a limit to what he could say (in fact, having been through mergers myself, I know there is even a limit to how much the two companies can even work together). He would say that since GF is IBM’s exclusive foundry for the next 10 years then obviously 14LPP will be part of the offering. GF is also inheriting IBM’s ASIC business, some of which over time should migrate to LPP as a next generation process.
So the catchphrase is “14 in 15”, with 14nm LPE ramping to volume, and 14nm LPP qualified and ready to ramp, all in 2015.