GlobalFoundries Finally Hits Its Stride On The Road Not Taken
Dr. Tom Caufield became CEO of semiconductor supplier GlobalFoundries (GF) early in 2018, announced that the company was dropping its plans to develop a leading-edge 7nm process technology five months later, and then took the company public late in 2021. Saying that it’s been a busy few years for Caufield seems like an understatement. During his keynote last week at the GlobalFoundries Technology Summit (GTS), Caufield was crystal clear about what GF’s goal is: to be the chip foundry for the rest of us.
Along the journey, Caufield and GF have developed a bit of magic for transforming what others view as commodity semiconductor process nodes into seven unique process platforms that serve the diverse chipmaking needs for a wide variety of customers. Of the 126 end markets for semiconductors with a $170 billion foundry Total Addressable Market (TAM) that GF has identified, the company decided to focus on and serve only 30 of these end markets and ignore the rest, which reduces the TAM by 30% and results in a Serviceable Addressable Market (SAM) of about $120 billion. Given that GF’s annual revenue is currently around $6 billion, there’s still plenty of headroom available for growth with that smaller SAM.
The seven semiconductor process platforms allow GF to stand apart from other semiconductor foundries through clearly differentiated offerings in these selected end markets, which include smart mobile devices, home and industrial IoT, automotive and transportation, communications infrastructure, and data centers. GF’s seven current semiconductor platforms are:
- 28nm, 40nm, 55nm, and 130nm Planar CMOS (Complementary Metal-Oxide Semiconductor), the workhorse process technology for digital applications
- 12nm and 14nm FinFET (Fin-shaped Field Effect Transistor) for high-speed digital applications
- 22nm FDX (an FD-SOI (Fully Depleted Silicon On Insulator) process) for low-power applications
- 45nm, 90nm, 130nm, and 180nm RF SOI (Radio Frequency SOI)) for applications such as 5G and automotive radar
- 45nm and 90nm SiPh (Silicon Photonics) for high-speed optical interconnect in data centers
- 45nm, 90nm, and 130nm SiGE (Silicon Germanium) for high-frequency, RF, and power applications
- Wide Bandgap (200nm gallium nitride and, in the future, silicon carbide) for high-power applications such as motor control and automotive
From a purely lithographic perspective, even the most advanced of these platforms is years old. However, it’s the add-on semiconductor process modules that GF has developed to enhance these semiconductor platforms that really make these offerings stand out. For example, the company offers a BCDLite module for the 55nm and 130nm nodes in its planar CMOS platform. BCD (bipolar-CMOS-DMOS) combines bipolar, CMOS, and DMOS (Double-Diffused MOS) transistors on one chip for diverse power applications, including audio amplifiers, power management, cellular and WiFi RF power amplifiers, interface circuitry, and battery charging. According to a press release dated November 4, 2020, “five of the seven leading top-tier smartphones currently on the market” incorporate GF chips manufactured with the company’s 55nm BCDLite process technology.
Nonvolatile MRAM (magnetic RAM) is another unusual add-on module that GF offers for its planar CMOS and FDX semiconductor process platforms. MRAM harkens back to the early days of computing when computers stored programs and data in magnetic core memory. Core memory came onto the scene in 1953, but it essentially vanished overnight when Intel introduced the first practical semiconductor DRAM, the 1103, in late 1970. Today, MRAM applied as a thin film layer atop planar FET circuitry promises to help fast, magnetic, nonvolatile storage make a comeback, at least in certain markets. MRAM is an important embedded non-volatile memory alternative to embedded flash, especially for 28nm process nodes and below where it becomes difficult to scale Flash memory.
GF’s SiPH semiconductor process platform for developing and manufacturing silicon photonics devices is another differentiated GF offering and the company formally announced its second-generation GF Fotonix platform in March of this year. Devices manufactured with the GF Fotonix processes combine photonic emitters and detectors, silicon optical waveguides, RF components, and high-performance CMOS logic on one silicon die. In addition, GF has developed advanced optical packaging using anisotropic etching to create precise V grooves that simplify passive alignment and attachment of optical fibers directly to the silicon die.
So far, photonics has been an expensive cottage industry, but these GF innovations promise to drive down the cost of using integrated silicon photonics at the system level, which will in turn drive demand and usage for photonics. Silicon photonics is potentially a game-changing technology for data centers, which are already adopting optical interconnections for high-speed links among servers at 200 Gbps and above. The recent GF Fotonix announcement mentions collaborative work on silicon photonics with Ayar Labs, Broadcom, Cisco Systems, Lightmatter, Marvell, NVIDIA, PsiQuantum, Ranovus, and Xanadu, which indicates the broad potential use of silicon photonics in networking, AI, and quantum computing applications.
Similarly, GF announced a radio frequency (RF) meta platform called the GF Connex portfolio during GTS. GF Connex encompasses elements of the company’s RF SOI, FDX (FD-SOI), SiGe, and FinFET platforms to meet the varied RF needs of smart mobile and IoT devices and communications infrastructure equipment. According to the announcement made during GTS, products based on GF Connex are in the market today through GF’s collaborations with Broadcom, Fujikura, MediaTek, Orca Systems, and Skyworks.
The semiconductor industry’s well-publicized and very expensive quest for smaller and smaller device geometries has been a one-dimensional effort in the much larger, largely unexplored, multi-dimensional world of silicon capabilities. The many add-on modules that GF offers across multiple semiconductor process platforms – such as nonvolatile memory (Flash, MRAM, and RRAM or “resistive RAM”) and high-speed and high-voltage bipolar transistors – augment each platform’s capabilities with additional features including RF, high-voltage and high-power drivers, and analog power amplification. The second-generation GF Fotonix platform and GF Connex portfolio are excellent examples of GF’s quest to bring additional value to existing semiconductor nodes by collaborating with its customers to develop and align specific platform features and add-on process modules more closely with customer requirements.
“We meet our customers where they are,” said Caufield during his GTS keynote speech. In contrast to the “smaller is better” approach taken by other semiconductor foundries, GF’s approach creates myriad new and unique foundry process technologies with much lower development and CAPEX (capital expenditure) costs. Coincidentally, GF’s seven-platform approach dovetails nicely with the current trend towards using chiplets to build packaged devices with more capabilities than can be achieved with monolithic ICs.
However, the new semiconductor processes alone are not usable without tool support. You can’t make what you can’t design. During GTS, GF emphasized its PDKs (Process Design Kits), developed in collaboration with the major semiconductor EDA players including Cadence, Siemens EDA (formerly Mentor Graphics), and Synopsys. Without these PDKs, it’s impossible to predict how the silicon will behave after fabrication and the challenge is complicated by the unique components that GF includes in its platform add-on modules. These complex PDKs are significant differentiators for GF and go well beyond the mere offering of the company’s silicon offerings.
During his keynote, Caufield made a few observations driven by current events. With respect to the current semiconductor shortages, he said “We will be chasing capacity for the better part of the next decade.” Much of the current semiconductor shortage affecting industries such as automotive and industrials have been in the same semiconductor process nodes that GF offers, which presents a significant and immediate growth opportunity for the company. Consequently, GF has started a building program for its existing fabs with the intent of boosting current manufacturing capacity by 60% over the next few years, which seems quite timely, for now at least. It remains to be seen whether the massive fab building programs initiated by the major semiconductor foundries and IDMs will continue to lag demand or whether they’ll trigger another bust cycle.
Caufield also noted today’s geographic uncertainties with respect to the semiconductor supply chain. Citing GF’s global fab footprint – the company has semiconductor fabs in the US, Germany, and Singapore – Caufield said that semiconductor manufacturing needs a global footprint to ensure a secure supply chain. “There are only five foundries left to do the lifting,” said Caufield, and he clearly intends to keep GF in the game and in that league of five through the company’s differentiated semiconductor platforms and increased manufacturing capacity.