Tuesday was the GlobalFoundries Technology Conference GTC. GF announced earlier in the month that they are dropping 7nm and are focusing all their effort on differentiated processes, in particular FDX (see my post GLOBALFOUNDRIES Drops 7nm to Focus on Other Geometries ). In March, Thomas Caulfield was appointed GF's CEO. Ironically, as the manager of fab 8, he was in charge of ramping GF's 7nm efforts that were scheduled to be in volume production in 2019. So dropping 7nm was, in a sense, killing his own baby. GF calls it "the pivot" (or "the pivot to relevance") and it even has its own logo. At GTC, Tom explained the pivot. History He started with the inside view of the history of the company. The timeline is in the picture above. The story starts with AMD. If you've ever worked for a semiconductor company, you know that the top priority is to fill the fab. Most of the cost of a fab does not depend on how many wafers are being run, the costs are either fixed (depreciation) or semi-fixed (wages). The problem for a company like AMD was that the economic volume required to fill a fab keeps going up at each process generation (sometimes, this is called Moore's Second Law, the cost of a semiconductor fab doubles every 4 years). Eventually the economics don't make sense since one company doesn't have enough scale to fill and pay for the fab. GF was spun out of AMD in 2009 when that happened and AMD felt that it no longer had enough scale on its own to be an IDM. Of course, spinning out their manufacturing didn't immediately solve the scale problem. GF had fab 1 in Dresden Germany. It broke ground on Fab 8 in New York. In 2010, it acquired Chartered Semiconductor, at the time the #3 foundry, with fabs in Singapore. In 2012, fab 8 delivered its first silicon. In 2014, fab 8 launched 14nm FinFET, initially trying to develop its own process and then using a process that it had licensed from Samsung. In 2015, GF acquired the semiconductor business of IBM, who made a similar decision to AMD that it no longer had enough scale on its own. In 2016, Fab 8 delivered 14nm FinFET and ramped it to volume. In 2017, fab 1 delivered 22FDX, the 22nm version of FD-SOI that GF had licensed from ST Microelectronics, who had developed the original concept. Over that ten years, GF had increased its capacity 8X and increased its revenue 6X. Tom Becomes CEO When Tom became CEO, he did what many CEOs do and made a tour of GF's customers, partially to introduce himself, partially to find out their care-abouts. At the time, GF was investing about 85% of its R&D dollars and 95% of its capital equipment budget in 7nm (and below). Tom found that his customers split into three groups: People who were never going to use 7nm or below. People who intended to use 7nm but for whom GF was not relevant. People GF was already engaged with at 7nm. The view of the first category was that GF's investments were not aligned with their needs since it was mostly going on a process roadmap they would not use. The second group were the big 7nm potential customers, where GF was irrelevant since it was always going to be too late, and with too little capacity. The third group, which included IBM and AMD among others, were worried whether GF was financially strong enough to fund a dual technology roadmap, and also that GF would not be big enough to be more than ~30% of their capacity for the long haul. The second challenge, after relevance, was financial sustainability. They would need another $3B to get to 12 Kwpm capacity (thousands of wafer-starts per month). This is "very modest" in Tom's words. A competitive 7nm is more like 50 Kwpm. So they would need to raise money, but the financial return was not there to make that attractive. Further, it was a problem for customers, since they need could see that the company didn't look financially sustainable, and they often had production requirements that would need manufacturing over 5-10 years. They were worried GF might not be there for them in the long-term. A third aspect, more of an opportunity, was that Moore's Law was slowing. It is not a technology issue but an economic statement. Leading edge is not the only game in town. The above pie charts show that in 2018 about 12% of the market is 7nm and below, growing nicely to about 20% of the market in 2022. But above 12nm grows from $56B today to $65B in 2022. It is simply not true that "everyone" is rushing to the leading edge. Another series of meetings Tom had were on a European tour where he met with Infineon, NXP, and STM. All had been on the leading edge until the economics didn't work for them anymore, so they switched to markets where they were relevant. Now they are doing so well that they are investing in capacity in Europe. Tom left Europe knowing that these were role models for GF. The Pivot So GF decided to pivot. Part of the pivot is easy, just stop doing the things that you are not going to do anymore, in this case 7nm (and below) process development, and the work on EUV that went with it. Since that was 85% of their R&D and 95% of their capex, that freed up a lot of investment dollars to do other things instead. Their processes are divided into four general categories: FinFET, FDX, RF, and Power/AMS. Four Offerings FinFET: this is targeted at the computing segment, but not at the most aggressive edge which will clearly move to 7nm. GF have an 1-155 performance boost over through SRAM Vmin, and 15% improved circuit density through library enhancements. GF have taken the 14nm process that they licensed from Samsung and developed 12nm version(s) of it (in-house, not jointly with Samsung). RF segment is aimed at front-end modules for 4G and 5G, with 8SW for LTE and sub-6 GHz, and 45RFSOI for mmWave. Plus see FDX below, that has RF capability. For Power/AMS they have 55 BCDLite as the industry's smallest BCD technology, optmized for digital-heavy mobility applications. Finally, FDX is aimed at IoT, intelligent X (where X can be home, city, factory, cars). This has the lowest active and standby power (using reverse back bias), high density logic, unmatched on-chip RF performance for always-on connectivity and the general move from LAN to wireless. The Fab Footprint The fab footprint has 5 fabs (or a fab complex in the case of Singapore): Burlington, Vermont Malta, New York Eash Fishkill, New York Dresden, Germany Singapore As Bami Bastani (SVP of the biz units) said in his presentation: Dresden is the home of FDX. Malta is the home of FinFET. Singapore is the home of Power/AMS (and embedded memory and sensors). East Fishkill and Burlington are home of RF. But RF is SIngapore too, BCD is in Dresden too, and more is coming up in Malta. Summary For many power-sensitive applications, Tom believes that FDX is the leading edge. Plus, it has the ability to add RF (which you can't generally do with FinFET). This covers a lot of the market since so much is not "tethered" in the sense that it is battery-powered and wirelessly networked. For applications like that, where power needs to be aggressively low, and cost considerations make on-chip RF a requirement, then FDX is very attractive. As Tom said wrapping up: We will grow, but we need to grow in a way that produces cash flow. We will continue to invest over $1B in R&D. We now have a healthy financial model. But we have to double down in relevant places and be successful in differentiated technology. Later that day, Tom and the other GF executives had a panel with the press and analysts in which they went into more detail. I will cover that next week. Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧