On Wednesday of SEMICON West I got to sit down with Gary Patton, CTO of GlobalFoundries and get an update on what has been going on with them.
Gary started the interview by pointing out that it has now been a year since the GlobalFoundries purchase of many of IBM's semiconductor assets and they have hit every commitment they made. They had a black eye from the ramp up of 28nm in Dresden, they canceled 20nm and had to license 14nm from Samsung. Last year they said they would qualify 14nm at the beginning of this year and they did. They now have a ton of tape-outs in-line, they are in production on multiple parts and yields are world class. The 14nm process they are running now will also provide a baseline for 7nm development.
Gary confirmed GlobalFoundries will not be offering a 10nm process. They believe it will be short lived node and don't see the value proposition in it (authors note, at 20nm TSMC is really the only foundry that offered it and they quickly transitioned to 16nm, many believe the 10nm to 7nm transition will be similar).
For 7nm they are designing the process to be done with optical but they are also positioned to introduce EUV when it is ready. In Gary's opinion, the mistake that was made at 20nm was it added multi patterning but it didn't provide much scaling so the value proposition wasn't there. 7nm is designed to optimize cost and scaling. He wouldn't comment specifically on 7nm timing but he would say he thought they would be competitive with other foundries. They will have a base 7nm technology and they are also looking at further performance kickers to bring in later plus preparing for EUV.
In terms of EUV the IBM 3300 EUV tool is at the advanced patterning center in Albany and GlobalFoundries will put their first tool there in partnership with NY State and CNSE. Prior to 20nm you might have 8 to 10 - 1x metal layers. With multi pattering you can only afford a small number of 1x layers.
In terms of 22FDX (their FDSOI process), progress is right on target. The 0.5 process design kit (PDK) was released in the second quarter as planned. They said they would get high 128Mb SRAM yield this year and they got it mid-year. They have invested heavily on IP this year - Cadence, Synopsis, Mentor, ARM, Invecas (design). They now have a pretty large foundational library of IP. With 22FDX they can operate at 0.4 volts and can get to 1pA per micron of leakage for IOT and they are working on integrating RF. You can have much higher Ft and Fmax than a FinFET technology and he thinks that will be important for 5G. They are on track for risk production at the end of this year and volume production next year. FDSOI is optimized for cost and power and really designed for IOT and the mobile space, for large chips and high performance, you need the performance of FinFETs. FDSOI is also easy to design for. FinFET design at 14nm is 3x the cost of a 28nm design according to Gartner and 7nm is expected to be another 3x.
In terms of embedded memory, embedded Flash works down to 28nm. GlobalFoundries has a collaboration with Everspin on MRAM and they will include it on the 22FDX platform. They are also looking at it for 14nm and beyond.
The IBM acquisition benefits have really played out as expected. 350 engineers have moved up to Malta to help with development and they still collaborate with Samsung on development as well. Gary has ramped up RF development, he controls the R&D budget. Burlington is an RF leader and they are using it to strengthen RF in Dresden, Singapore and Malta. They are number one in the ASIC space and they have moved it onto 14LPP as well as using it to provide foundry IP. They are already quoting ASICs on 7nm. They got silicon photonics IP, trusted foundry status and they have completed the certifications. IBM had a fairly large 2.5D and 3D program and they combined it with Global' s program. They were first to do TSV and 32nm for the Micron Hybrid Memory Cube. IBM had no focus on wireless, the RF group was really on the outside. They now have one of their best RF development guys in Dresden leading the implementation of RF on the 22FDX technology.