At the recent GlobalFoundries Technology Conference GTC there was a press and analyst lunch. Tom Caulfield (CEO), Gary Patton (CTO), and Bami Bastani (SVP of the business units) answered questions, Tom opened the proceedings by giving a little more detail covering some of what he already covered in his keynote earlier (see my post GTC: GlobalFoundries Pivots for more details). Tom said that GF started on 7nm about 2-3 years ago but the conditions to go to the leading edge have really changed. All businesses need to regularly challenge their business models. Tom actually talked about Gordon Moore and Andy Grove taking the decision to pivot and move Intel from being a memory company, where they were losing money, to put it on the path to what it has become today. I actually told the same story in my piece when Andy Grove passed away (see Andy Grove: RIP ). Tom pointed out, too, that it is actually easier to do as a private company since there is no need to deal with hundreds of thousands of shareholders. Once they took the decision to stop work on 7nm, they have had 60 day task force to work out what they will do instead. The Pivot and AMD Someone asked when they decided to do the pivot. Tom said that he started as CEO on March 9th, and it is easier for new management to make decisions. But it's not as if this sort of thing arises over a single quarter, they take a long time. They felt they needed to make strategic decisions based on their customers. Their customers were "apathetic" (Tom's word) about what they were investing in. So from March to mid-June they knew they had to make the decision, and then figure out what to do instead going forward. They were not spending enough to keep leadership in other areas, such as RF (where they are #1). The next question was about what markets they were going to give up. The computing market. Tom said they have a great business today in CPU and GPU, but they are pretty much forced to go to the next node. Some companies are doing more of a mixed model with just the core itself on the bleeding edge, and then assembling the system using advanced 2.5D packaging. Tom said that packaging is really changing things. I was in packaging early in my career, and packaging is always the future…but then someone would integrate everything. But now that is finally changing. Processors for AI might be a game changer since memory and processing need to be closer together, and that is unlikely to be integration on a big single die. AMD and GF have been joined at the hip and are now following diverging paths, but there will still be ways for them to use GF's technology. In 14nm Lisa [Lisa Su, AMD's CEO] made a big bet going with GF as a single source. But she had no choice, and AMD has never released so many great products, and rebuilt their brand. In 7nm, our plan was never to have more than 50% of their volume just due to customer diversity. But we wouldn't be here today if we hadn't made AMD a success. Lisa took her stock price from $1.50 to $30. Packaging The team was asked about packaging and whether GF plans more investments beyond 2.5D and perhaps doing their own fanout-level packaging. The GF approach is to innovate in the packaging technology, but have the OSATs make the investments. They see other foundries entering the space, but customers can get leery of being locked in. For GF's scale, it works to help the OSAT industry become more of a developer of technology. Gary pointed out that they brought the advanced packaging team over from IBM so they can develop any packaging technology they need, but then they partner with the OSATs for high-volume manufacturing. Talking of interposers, Gary had said earlier in the day that they could do an interposer twice as large as the maximum reticle size, so one question was whether they could go larger. Gary said that they are using a stitched approach. They could go more than 2X, but that seems about the right size since it gives you a processor and HBM around it. The registration to do the stitching is not hard at the resolution of an interposer. Bami pointed out that the technology enablement for complex interposer-based designs is important, to do a complete simulation including the package, thermal, TSVs, and silicon all in one. A specific very complex example is to put phased antenna arrays on one side of a chip, and the silicon design itself on the other. A question on how they see RISC-V. Bami said that they are part of the GF ecosystem, and that SiFive is working with them. "There is a lot of buzz in China" and a lot of interest in NB-IoT (narrowband, which is low power cellular but with huge numbers of devices per cell). But Tom emphasized that GF are IP-agnostic. "Our job is to create the richest environment for users." The discussion moved on to "weird" technology. Bami said that silicon photonics is strategic for them, especially in the datacenter. They see it coming across the PCB ultimately. The growth rate is phenomenal. "it is a great example of where our customers could not have done it on their own, and we couldn't have done it without them." Production is in East Fishkill. They do MEMS in Singapore, but aren't looking to grow that business much. Capacity and Technologies Tom told us to "look at what we've managed to do with little more than a skunkworks. Now imagine if we put in another $20M". To put that in context, he pointed out that $20M is about one month's 7nm development. "We've been running with weights on our ankles and what these guys have accomplished in areas like silicon photonics with minimal budget is amazing." The next question was for an update on China, where GF have a joint-venture building a fab. Reading between the lines, it is not going that well. Tom said: China was to access demand we could only reach with manufacturing in China. We weren’t doing it for lower cost manufacturing. China wants more of manufacturing to be done in China. But the fabless guys didn't get the memo. So demand in China has been slow to materialize. We have agreed we need to double down on getting more of the domestic supply and getting companies to commit. The guys who should be doing it aren’t creating the demand. Tom also addressed capacity outside China, saying that GF is adding capacity in Burlington, Fishkill and Singapore for differentiated technology "now that we are not spending 95% of our capex on 7nm." Across the board GF has added capacity and it is in the process of being installed. GF is also adding more 12nm capacity (presumably in Malta). But Tom did emphasize that GF is not taking everything that they were spending on 7nm and now spending it somewhere else. "We are spending aggressively but being prudent. We are making sure we can multisource in more fabs for RF." Bami pointed out that already GF is shipping RF-SOI at 300mm from both Fishkill and Singapore "But FinFET is too expensive so we can't multisource that, although we could bring other mature technologies into Malta." When GF was started, there had been talk by ATIC (the investment arm of the government of Abu Dhabi, that owns GF) to build technology infrastructure and eventually a fab in Abu Dhabi. "Has that talk faded?" Tom said that he didn't think it was high on the priority list. They used to have $60B under management, and now they have $240B under management. It is truly a sovereign wealth fund focused mainly on growth of the fund. Also, they have a more pragmatic view of what it takes to build a greenfield fab. Overall, the investment in spinning GF out from AMD and taking a big position in AMD too has produced a huge return on the AMD side. The team was asked how they seem themselves against the competition. "You were a fast follower before?" Tom said that they are "the #1 differentiated foundry. We are leading at 22nm and RF. Today 35-40% of the portfolio is like that and we want that to be 85-90%." Gary complained that the press is working against the message. "I can spend 95% of my time talking about RF and FD, and still 9 times out of 10 the article is about 7nm...but that's not where the bulk of the industry makes its money." Wrapup A final question as to whether the portfolio is too broad. Tom said he is less worried about the breadth of the portfolio since it is differentiated. But he is concerned with the footprint and GF needs to "make sure it can fill all these assets." Tom gave his final comments. First, they have a simpler management structure to go with the simpler roadmap. Customers tell him "you are so much easier to deal with now." Finally, he looked forward: It is an exciting time for the industry. It is a golden age for the semiconductor industry: PC, internet, mobile. Now IoT. But the real next App is "everything connected." It will drive a sensational amount of silicon and it mostly won’t be 7nm but in more in mature technologies. Stay tuned. The Last Word The very last word goes to Wally Rhines, CEO of Mentor. I happened to talk to him at the end of the day. "I've always worked in underdog businesses," he told me. "But this strategy really has a chance to work. Whereas competing at 7nm with inadequate capacity certainly would not have done." Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧