On August 14, Intel announced that Synopsys would develop a portfolio of intellectual property (IP) for the Intel 3 and Intel 18A process nodes, which will broaden and accelerate access to Synopsys silicon IP for Intel Foundry Services (IFS), its customers, and its growing ecosystem. The support of Synopsys, one of the “big three” Electronic Design Automation (EDA) vendors and a significant player in semiconductor IP, is essential for any semiconductor foundry that wants to be taken seriously in the world market.
Just two days later, Intel announced that it was terminating its 18-month effort to acquire Tower Semiconductor and that the company will pay Tower a $353 million termination fee “due to the inability to obtain in a timely manner the regulatory approvals required under the merger agreement.” More specifically, the deal fell through thanks to the Chinese government’s stonewalling on its approval.
These two announcements, just days apart, represent the zenith and then the nadir of Intel’s efforts to enter the ranks of the world’s top semiconductor foundries through IFS. However, IFS’s orbital path will likely include many more apogees and perigees as it circles Mother Intel. In fact, a surprise announcement made yesterday by Intel and Tower reinforces that orbital analogy.
Having Synopsys as a strategic IP EDA partner further bolsters IFS’s aspirations to become a major foundry player. Although the announced partnership agreement explicitly covers the Intel 3 and 18A process nodes, both Intel and Synopsys characterize this agreement as a multi-year, multi-generational agreement. According to John Koeter, SVP of IP Product Management and Strategy at Synopsys, the agreement includes a framework for extending the strategic partnership to future process nodes beyond Intel 18A.
No semiconductor foundry can call itself a major player without similarly engaging all of the top IP and EDA vendors. Free-range foundry customers want and demand choice, something that was lacking in Intel’s previous ventures into the foundry arena. EDA partners in the IFS Accelerator Ecosystem Alliance include Cadence and Siemens as well as Synopsys, while IP partners include microprocessor IP heavyweights such as Andes, Arm, and SiFive in addition to IP vendors with broad IP libraries including Cadence and Synopsys.
Make no mistake, this agreement is a good deal for Synopsys as well, because it ensures ready access to Synopsys IP by another bleeding-edge semiconductor vendor and its customers. With Intel’s declared intent to regain semiconductor process leadership by 2025, IFS will surely be a very important foundry partner for Synopsys.
This announcement is yet another milestone in the roadmap for Intel to regain the company’s leadership position in chipmaking. Recently, Intel CEO Pat Gelsinger combined IFS with Intel semiconductor manufacturing and packaging operations and TD, the company’s process technology development group. Gelsinger then gave this organization its own profit and loss (P&L) statement, reporting directly to him. This move requires Intel manufacturing to prove itself in the real world, with real competitors, and the P&L makes visible every cost – chip revisions, expedited orders, special process tweaks, and so on – that were once absorbed as corporate overhead. In addition, Intel business units will now become more financially responsible for their behavior than ever before, all of which should be good for Intel.
OK, that’s the fun stuff.
The loss of the Tower deal is not nearly so much fun. Intel and Tower announced the planned acquisition on February 15, 2022. The technical goals used to justify the $5.4 billion purchase were quite reasonable:
- To create a “globally diverse end-to-end foundry” to help meet growing semiconductor demand.
- To create a wider product and services offering to access “the nearly $100 billion addressable foundry market.”
- To accelerate Intel’s path towards becoming a major global provider of foundry services by creating “the industry’s broadest portfolios of differentiated technology” by merging Intel’s leading-edge nodes and scale manufacturing with Tower Semiconductor’s specialty semiconductor processes.
- To assimilate Tower’s “customer-first approach.”
All these goals would clearly make IFS even more credible as a top-tier semiconductor foundry.
Tower Semiconductor is an Israeli success story founded upon trailing-edge or legacy semiconductor processes. The company was founded in 1993 and started with the acquisition of National Semiconductor’s 150mm wafer fab in Migdal HaEmek, just outside of Haifa, Israel. The industry had already migrated to 200mm wafers at that time and was soon to jump to today’s 300mm wafers, so the facility in Migdal HaEmek was clearly a legacy fab.
Tower acquired Jazz Semiconductor in September 2008 and renamed itself “TowerJazz” a year later. Jazz Semiconductor had its own storied history. Jazz Semiconductor Systems was founded on February 15, 2002 and absorbed Conexant’s fab in Newport Beach, California. Strange coincidence: I attended Jazz Semiconductor’s coming out party at that Newport Beach facility back then. Conexant itself grew out of the dissolution of Rockwell Semiconductor, which had been a division of Rockwell International.
Rockwell created that Newport Beach fab when it started a semiconductor manufacturing operation for its Autonetics Division in 1967. Autonetics developed various military/aerospace avionics systems including inertial navigation and guidance systems for US submarines and ICBMs, which created the need for advanced semiconductors. North American Rockwell Microelectronics Corp (NRMEC) developed an early MOS/LSI process technology for its military and aerospace projects and then became a major commercial semiconductor player in the 1970s, when it developed its own line of microcontrollers. (See “A History of Early Microcontrollers, Part 3: The Rockwell Microelectronics PPS-4/1.”) Rockwell was also a second source for Motorola’s 68000 microprocessor.
In June 2011, TowerJazz acquired a Micron Technology fab in Nishiwaki City, Hyogo, Japan, which nearly doubled the company’s manufacturing capacity. Then, in February 2016, TowerJazz acquired a 200mm fab in San Antonio, Texas from Maxim Integrated Products. TowerJazz reverted its name to Tower Semiconductor on March 1, 2020. By acquiring all these fabs and their proprietary semiconductor process technologies, Tower now offers a rich mix of semiconductor processes that include mixed-signal CMOS, digital CMOS, silicon photonics, BiCMOS, SiGe, MEMS, BCD, SOI, RFCMOS, and embedded Flash memory. These are not leading-edge processes as measured by sheer lithographic prowess, but they nevertheless remain incredibly useful processes, in high demand. Had Intel been able to acquire Tower and add its varied process technologies to IFS’s angstrom-era digital CMOS processes, IFS would immediately have become a broad-spectrum semiconductor foundry player. However, that acquisition was not to be.
Intel has had plenty of warning that its Tower acquisition was in trouble thanks to Chinese regulators. It’s just another skirmish in the current economic warfare taking place between the US and China. I’ve got to believe there’s a Plan B that will now be executed. It’ll be extremely interesting to see what plans Intel and IFS devise in the wake of the aborted Tower Semiconductor acquisition.
Note: Well, we didn’t need to wait long. Intel and Tower announced Plan B the day before this article was scheduled to appear and two weeks after the above was written. Yesterday, September 5, Intel and Tower announced a foundry agreement where Tower will be purchasing 300mm wafer manufacturing services from IFS, which initially will manufacture wafers based on Tower’s 65nm BCD (a combined bipolar, CMOS, and DMOS process) in Intel’s existing Fab 11X located in Rio Rancho, New Mexico. Tower will invest as much as $300 million to acquire and install fab equipment and other fixed assets in Intel’s Fab 11X to support this process flow. For Intel and IFS, this agreement is a step towards creating an “open system foundry,” but perhaps using agreements with other foundries rather than by acquisition. Tower acquires access to another, far more efficient, 300mm fab faster and for less than the cost of standing up another new fab or converting an existing 200mm fab to 300mm. Tower’s BCD process is projected to be running in Fab 11X sometime next year. This latest agreement neatly sidesteps the regulatory interference that nixed Intel’s acquisition of Tower.