Samsung Foundry had an intimate gathering recently for 200 customers and partners that I missed, but I know several people who attended. This event was a precursor to #53DAC where Samsung has the largest foundry presence. I was able to clarify what I had heard via a phone call with Kelvin Low so here is my version of what is important:
Samsung is all in on the foundry business
Samsung is opening up their 200mm fabs, internal IP, design methodologies (IE: low power), and related services (packaging) to foundry customers. To me this is a definitive statement as to their foundry commitment. Samsung is not however going into the captive ASIC business like TSMC (GUC), UMC (Faraday), GlobalFoundries (Invacas), and SMIC (Brite Semiconductor). Samsung could easily buy an established ASIC supplier like eSilicon, Open-Silicon, or Verisilicon, but Samsung is choosing to not compete with their ASIC partners, which makes complete sense since the other foundries do. I would bet Samung will get a much larger share of the ASIC business in the not too distant future (it’s a safe bet since I have already asked my ASIC friends about this).
Samsung Foundry is continuing to focus on 28nm FD-SOI
I saw this at the FD-SOI symposium where Kelvin presented “28FDS – Industry’s first mass produced FDSOI technology for IoT era, with single platform benefits.” Unfortunately the slides are not up yet, I will let you know when they are posted. For China, Samsung is FD-SOI enabling their ASIC partners which is a great strategy, Verisilicon for example is very active in China.
Key FD-SOI take aways from the Symposium:
- Variability lower than bulk
- No reliability concerns – all WLR and PLR completed
- No FD-SOI specific in-line defect generation and systematic failure
- Proven performance benefits on silicon
28FDS commercial products are in production
- Technology deployed in actual products
- 12 tapeouts in 2015 and >10 tapeouts so far in 2016
Full foundry support from design to manufacturing
- Samsung Foundry supports foundation and basic IP
- Other IP by 3rd party vendors (ARM, Synopsys, etc…)
- Regular MPWs available for design validation
28FDS will be a long-lived node
- Derivative offerings including RF and eNVM
- Increase reach into new markets (Auto, IoT, Industrial, etc…)
Samsung Foundry is offering a low cost version of 14nm
This was not surprising at all given the TSMC 16FFC announcement last year but I am told that Samsung Foundry LPC (cost down version) offers process simplifications (less masks) without compromising performance. LPC is also PDK compatible with LPP for seamless design migration. Thus far Samsung has shipped more than .5M 14nm wafers making them the largest FinFET foundry share holder today and that’s a fact.
Samsung Foundry 10nm will be in production by the end of 2016
Samsung is approaching 10nm differently than TSMC. Rather than doing a quick node transition from 10nm to 7nm, Samsung will focus on 10nm as a full node by building out different versions targeted at multiple markets. According to Samsung a “true” 10nm can be done using double patterning thus saving the cost of triple or quad patterning. Samsung does use triple patterning on one of the metal layers but still allows bidirectional routing which is easier to design to.
Samsung Foundry 7nm will use EUV for cost reduction
As I was told at SPIE, Samsung will use EUV for 7nm logic before using EUV for memory. An executive from ASML EUV (Dr. Hans Meiling) even presented at the Samsung event to bring everyone up to date. Given that Samsung 10nm will be a full node, delaying 7nm until 2020 (EUV ETA) should not be a problem.
Bottom line: Samsung is showing significant foundry leadership skills again with FD-SOI and FinFETs. Not only does this greatly benefit the fabless semiconductor ecosystem by giving us more innovative foundry choices, it also benefits the semiconductor industry by continuing to push the cost per gate to affordable levels.