AMD’s decision to spin off its manufacturing business into a separate entity has created a fairly unique event in semiconductor manufacturing: an ostensible newcomer has opened some of the world’s most advanced chip fabrication capabilities to paying customers of all stripes. Last week, we traveled to the new offices of GlobalFoundries in Saratoga County, New York to meet with the firm’s executives and to understand their plans for the newly minted chip foundry.
Although much is new about GlobalFoundries, including the company’s name and mission, key parts will be familiar to industry observers, because they include many assets and personnel formerly from AMD. Those assets include two chip fabrication plants in Dresden, Germany. Now called Fab 1 module 1, the former Fab 36 produces AMD’s processors, including Phenom II and Opteron CPUs, using an advanced 45nm silicon-on-insulator process on 300 mm wafers. The former Fab 30 is being retooled for 300 mm wafers, as well, and has been renamed Fab 1 module 2 in GlobalFoundries’ lexicon.
The transaction that created GlobalFoundries gave majority ownership in the firm to the Advanced Technology Investment Company, which is wholly owned by the government of Abu Dhabi, in exchange for a multi-billion-dollar investment. AMD retained 34% ownership in GlobalFoundries and voting rights equal to ATIC’s. AMD will be GlobalFoundries’ single most important customer for the foreseeable future, as well.
The big change introduced by the spin-off is GlobalFoundries’ intention to enter the foundry business and, as we’ve noted, offer its chip fabrication capabilities to a range of clients. To that end, GlobalFoundries has been assembling a team of veteran executives, both from within the former ranks of AMD and from outside, with experience in key phases of the semiconductor business. Hearing them talk about GlobalFoundries as “a big startup” may fall strangely on the ear of anyone familiar with AMD, but there’s more than a kernel of truth in the sentiment.
Grose point blank
GlobalFoundries CEO Doug Grose makes the case for a new entrant in this business by arguing that chip design is most fundamental to innovation. The foundry’s role, as a partner, is to provide leading-edge manufacturing capabilities to its customers, so they can remain focused on design. Staying at the forefront of chipmaking technology is no trivial undertaking, he points out, with R&D budgets and fab construction costs ballooning in recent years. The barriers to reaching new process nodes, or “red brick walls,” are getting higher and thicker. By consolidating the knowledge gained from its technology alliance with IBM and other partners, and by serving a diverse portfolio of customers, GlobalFoundries aims to provide a stable source of manufacturing capacity while remaining at the forefront of process advances.
As proof of the opportunities open to GlobalFoundries, Grose points to a host of companies who have traditionally owned fabs but are now looking outside for manufacturing help at advanced process nodes. Among them are some household names like Sony and Toshiba, alongside industry stalwarts like LSI and Freescale. Other big names may fall soon, too.
Grose’s pitch for his firm also emphasizes security of supply, a potential worry for fabless customers of firms like TSMC whose fabs are centralized in Asia. GlobalFoundries can provide an additional measure of security with its fabs located in Germany and, soon, in upstate New York, as well.
The CEO identifies several challenges for his company, and at the top of the heap is establishing that GlobalFoundries knows how to work independently of AMD. Doing so will mean building out the company’s design enablement services, so that it has engineers capable of assisting customers in creating viable, workable chip designs. The firm has one design team in Dresden, formerly of AMD, and another in California to help with enablement. Sales and market efforts are ramping up, as well.
Grose says the current plan is to focus on 45nm and newer process technologies, but he recognizes that the 45nm node will eventually be a generation or two behind the state-of-the-art. As that happens, GlobalFoundries may expand its focus to serve a broader set of customers with its older fabs, presumably at lower costs.
Few customers other than AMD are likely to take advantage of the full gamut of advanced materials and techniques used in GlobalFoundries’ 45nm SOI DSL process. Adapting a design to such a process is not a trivial undertaking. Many products don’t require and may not benefit from the combination of additional switching speed and associated cost that comes with an advanced process, either. As a result, GlobalFoundries must offer a more conventional bulk silicon option in order to attract customers.
According to Jim Doran, the company’s Senior VP and General Manger, Dresden, Fab 1 module 2 will provide that bulk silicon capacity at 32nm when its retrofit is complete. At present, production is scheduled to ramp late this year, and the module should be capable of 25,000 wafer starts per month when it reaches peak capacity. Meanwhile, Fab 1 module 1 will remain focused on SOI production for AMD, with the crossover point from a majority of 65nm wafers to a majority of 45nm wafers expected to come very soon, near mid-year. Module 1 is also capable of up to 25,000 wafer starts per month.
The crown jewel of GlobalFoundries’ roadmap is Fab 2, a new 300 mm fab to be located on the Luther Forest Technology Campus in Saratoga County, New York. Fab 2 will be geared toward chip production on 32nm and later 22nm process tech, and the facility represents a $4.5 billion investment. Although the firm hasn’t yet broken ground at the site, it hopes to start initial silicon at Fab 2 module 1 in late 2011, with shipping products rolling off the line in 2012. Capacity should reach at least 35,000 wafer starts per month once the production has fully ramped. In addition, the Fab 2 site has been intentionally planned with further expansion in mind, so second and third modules of the same sizeor of varying sizesshould be feasible, although the company hasn’t committed yet to any specific course of action or time frame, according to Terry Caudell, Director of Wafer Manufacturing.
Just as AMD has done in Germany, GlobalFoundries has worked closely with local governments, especially the state of New York, to secure a rich set of incentives and tax breaks as inducements to build its fab in this location. In return, Fab 2 should directly bring upwards of 1,400 jobs, many of them highly skilled, to the area. In addition, GlobalFoundries projects 5,000 “spin-off” jobs to be created by Fab 2’s presence as a high-tech anchor in the area.
One of the advantages of the Saratoga County location is its relative proximity to key research partners, including IBM’s 300 mm fab in East Fishkill, New York and the College of Nanoscale Science and Engineering of the University at Albany. AMD has long worked with IBM and other members of IBM’s collaborative technology alliance, and GlobalFoundries continues to participate in that R&D network. The alliance conducts early-stage research and develops new process technologies, taking them to production readiness. Most of that work takes place in IBM’s East Fishkill location, and GlobalFoundries has a team of over 70 people working there now. John Pellerin, GlobalFoundries’ Director of R&D, says the current agreement runs with IBM though 2015 and includes development of bulk silicon technologies, as well as SOI.
Already, 32nm technology is deep in development, in both SOI and bulk variants. Like the AMD/GlobalFoundries 45nm process, the 32nm process will use immersion lithography to enable smaller geometries. In addition, 32nm will see the introduction of the alliance’s “gate first” version of high-K/metal gate technology, along with an ultra-low-k insulator made possible by a porous material. Both changes should improve performance and reduce power consumption, if they work as intended, although the company cited no specific targets for overall performance or power improvements at 32nm.
Pellerin said the first 32nm SRAM test chips were run in Dresden in the fourth quarter of 2008. GlobalFoundries expects to accept tape-outs in late 2009, and production is on schedule for early 2010.
The next step beyond that, at least for bulk silicon, is a 28nm “half-node” process. This slightly smaller process should be an easy conversion from 32nm for any existing GlobalFoundries customers, and it’s projected to achieve 40% higher performance and 20% lower power consumption than a comparable 45nm process.
Pellerin names a handful of technologies as possible foundations for subsequent generations of fab tech, including 3D ICs that allow the stacking of components, a new transistor structure called multi-gate FinFET, EUV lithography, and computational lithography. Further out, possibilities like carbon nanotubes and molecular self-assembly enter the conversation, although Pellerin admits such talk is a bit tongue-in-cheek at present.
But perhaps only a bit. One of the first EUV lithography tools in the world is located not far from the future site of Fab 2, at the College of Nanoscale Science and Engineering of the University at Albany. The CNSE is a highly distinctive institution that brings together the academy, employees from a range of companies in the semiconductor industry (including the major tool makers and many members of the IBM alliance), and government. With 80,000 square feet of clean rooms, the CNSE is engaged in everything from very early-stage, nano-scale research to chip prototyping, all while granting degrees. Many elements of the next-generation process technologies used by GlobalFoundries and the other members of the IBM alliance are likely to be developed there.
The spin-off of GlobalFoundries is certainly an intriguing way for AMD to address what it perceives as the problems it faces in advanced semiconductor manufacturing, a way that pushes its collaborative approach to new extremes. What remains to be seen is whether other firms in the industry see their challenges in the same way and are willing to trust GlobalFoundries with their manufacturing needs in a large-scale fashion. Then there’s the fact that some of GlobalFoundries’ best potential customers are AMD competitors, which adds another wrinkle to the mix. Would, say, an Nvidia hand over the plans to its most advanced GPU to a subsidiary of its biggest competitor? GlobalFoundries execs express optimism on these fronts, hinting that they expect to earn the business of some very good customers, even perhaps some surprising ones, without naming any names.
We will simply have to wait and see about that. These are early days still for this venture.
For now, Grose is keeping his message simple and clear: GlobalFoundries is now open for business, engaging with customers on many fronts, and ready to go in 2009.