中文 English

Behind The Intel-GlobalFoundries Rumor

Analysis: Why this deal suddenly makes a whole lot of sense.

popularity

A Wall Street Journal report that Intel is looking to buy GlobalFoundries has sparked discussions across the industry. But what exactly this would mean, and why now versus a couple years ago, needs some context.

There are layers upon layers of irony behind this would-be deal, and it dates back decades to some rather famous encounters. Consider former AMD CEO Jerry Sanders’ 1991 comment that “real men have fabs.” (This was a reference to a best-selling book about our macho culture entitled, “Real Men Don’t Eat Quiche.”) At the time, TSMC was a minor player in chip manufacturing, and chipmakers that outsourced their manufacturing were at a competitive disadvantage because they had to wait for silicon to come back before making changes to a design. Commercial EDA tools were still in their infancy. DFM/DFY/DFT tools didn’t exist yet. And Sanders apparently had no idea just how fundamentally things were about to change.

Since then, EDA tools have improved, foundries have found their footing, and scores of fabless startups that couldn’t afford their own fab entered the market. Seventeen years after Sanders slammed his competitors during a panel discussion, AMD spun off its fab operations to create GlobalFoundries. That same year, in 2008, a majority share in GF was sold to Advanced Technology Investment Company, a subsidiary of UAE-based Mubadala Investment Co.

GF entered the market with a fat checkbook. In 2009, it acquired Singapore’s Chartered Semiconductor, and in 2014 it acquired IBM Microelectronics. The IBM deal came with lots of upside, as well as some caveats. IBM paid GF $1.5 billion to take its fab in Essex Junction, Vermont, off its hands, with the understanding that GF would continue to make chips for its servers down to 10nm, and possibly beyond.

But competing against TSMC and Samsung proved both difficult and expensive. Last month, IBM filed suit against GlobalFoundries for breach of contract because GF never produced 10nm chips. The foundry first claimed it was going to skip the 10nm node, which it considered a half node, and move directly to 7nm. Then-CTO Gary Patton described GF’s forthcoming 7nm process as “a very strong shrink off of 14nm.” But in 2018, GF dropped plans for 7nm, focusing instead on its FD-SOI process. A year later, GF CTO Gary Patton left the company and moved to Intel, the last major U.S. microprocessor company to still own a fab.

Dig deeper into semiconductor history and you encounter more irony. Intel ‘s emergence as a microprocessor giant stems in part from the U.S. Department of Justice’s efforts to prosecute IBM in the mid-1950s, culminating in a 1956 monopoly-busting consent decree. Under the terms of that deal, IBM agreed to stop bundling services, software, and hardware, a “market-basket” tactic it had used to corner the mainframe market. As soon as a competitor entered the field, IBM would cut prices on whichever of the three baskets were being threatened. So as proof that it had changed its colors and become more competitive, IBM opened up its PC business to outside suppliers on the assumption that the PC would never be a serious competitor to its margin-rich mainframes and minicomputers.

Prior to that, PCs were largely do-it-yourself kits, and “portable” computers typically had to be wheeled around on carts because they weighed as much as 75 pounds. The introduction of the IBM Personal Computer put IBM’s stamp of approval — at the time, it meant a great deal — on what it had previously dismissed as a toy. That, in turn, created a market for everything from processors and memory to software applications. The PC proved so successful that prices dropped from thousands of dollars per computer to just hundreds, ultimately forcing IBM to sell its PC division to China’s Lenovo in 2005.

IBM wasn’t the only company to miss out on the next big thing. Intel was so focused on the PC/laptop explosion that it completely missed the mobile phone/smart phone market, and it misjudged the IoT market for connected things that followed. Even in markets where Intel traditionally had been dominant, it now faces competition from Arm, AMD, and increasingly from RISC-V competitors. Single- and multi-processor chips with multiple cores have given way to heterogeneous, domain-specific designs filled with accelerators and specialized processors. Apple is developing its own Arm-based processors, and cloud-based companies such as Google, Facebook, and Amazon are designing their own server chips. And to make matters worse, Intel recently lost its global process technology leadership position to TSMC and Samsung, at a time when it has been scrambling to leverage its foundry services to help pay for increasingly expensive fabs.

Recognizing it had to make some changes, Intel overhauled its management team, bringing back former CTO Pat Gelsinger as CEO. The company also began leveraging and touting its numerous developments in advanced packaging technology, including a low-cost bridge to connect multiple dies and chiplets, several different interconnect technologies, and various memory technologies. It also began integrating some third-party IP.

This may prove to be a winning strategy, although not for the reasons Intel initially promoted. The company appears to be warming to the idea of becoming a trusted foundry for the U.S. Department of Defense. That could add billions of dollars in government contracts, and it could make Intel’s foundry efforts highly profitable, especially with the company’s chiplet approach. Acquiring GF would provide specialty processes, such as SiGe and FD-SOI, along with some general 300mm leading-edge capacity that already is up and running, fueled by a variety of foundry customers.

As of today, there are no leading-edge foundries owned by U.S. companies. Intel did not accept military contracts in the past because there wasn’t enough volume in custom chip designs. But the focus on chiplets and advanced packaging is a common cause for both Intel and the DoD, and a potentially lucrative one for Intel. And with the U.S. government’s new-found willingness to invest in domestic semiconductor technology in order to remain competitive with China, Intel could well become the beneficiary of steady government business and possibly a direct cash infusion.

Whether the Intel-GF deal actually happens, and whether it ever gets finalized, is far from certain. Not all talks result in deals, and not all deals go through. But from a business perspective, what would have seemed unthinkable a year or two ago is suddenly within the realm of possibility. And from the U.S. government’s perspective, the company that benefited handsomely from the 1956 consent decree with IBM now may be in a very good position to repay the favor.

Related
Intel/GF Deal: Pros, Cons, Unknowns
Will the deal happen in the first place?
Regaining The Edge In U.S. Chip Manufacturing
Taiwan and Korea are in the lead, and China could follow.
Geopolitical And Economic Outlook For Chips And Equipment
Supply-demand imbalances, new markets for advanced chips, and how a dual supply chain could disrupt the chip industry.
GF Puts 7nm On Hold (2018)
Foundry forms ASIC subsidiary as it focuses on 14nm/12nm and above.
Chipmakers Look Beyond Scaling (2018)
GlobalFoundries CTO Gary Patton digs into how customers’ priorities are shifting with new market opportunities.
China Speeds Up Advanced Chip Development
Efforts underway to develop 7nm, DRAM, 3D NAND, and EUV domestically as trade war escalates.



5 comments

Fred says:

One additional irony is that IBM’s choice of Intel in the 80’s was partially motivated by a concern over Japanese dominance in chip manufacturing. Intel was struggling with Japanese competition in memory and IBM wanted to make sure there was viable manufacturing capability in the US.

Hank Walker says:

Intel was awarded $156M in a DoD contract last year. And they have accepted in the past. They used to product MIL-STD-883 parts.

I believe it is the 1969 to 1982 government antitrust lawsuit that is relevant, since it was during this time that the IBM PC was developed, and IBM’s behavior was governed by the ongoing suit.

Ed Sperling says:

Hank, that’s correct. The IBM consent decree was modified before finally being phased out sometime around the Millennium under Lou Gerstner’s tenure as CEO. Intel has done some work with the government, but it has been relatively minimal for a variety of reasons, ranging from from customized fab processes and materials to limited volume.

HLector Favabeen says:

The Fab Industry needs less consolidation not more! So the Regulators need to step up the Antitrust enforcement. So If Intel wants to spin off it’s foundry operations and combine that with GF then Fine as that would be a separate entity not under Intel’s direct control.

The US Government needs to foster more leading edge fab capacity but not at the cost of further consolidation in the industry. The Antitrust regulators need to monitor the pure play fabs for any one client attempting to buy up all the wafer starts contracts from the pure play foundries around the world and not let any monopoly interest comer the processor market by improperly reserving all the needed pure play foundry wafer starts capacity.

TJN Texas says:

Your comments seem to miss the rather valid barrier to entry that leading edge fabs have. There simply won’t be more than a few – the 3 now is likely the number for the foreseeable future. I think that the current actions by Congress are in fact one of the few methods to get this leading edge capacity within the US – even if some is made by TSMC or Samsung. Intel starting foundry capacity may be in exchange for more access to funding for leading edge capability. It is all needed at some level.

Leave a Reply


(Note: This name will be displayed publicly)