Amidst soaring demand for AI hardware and a persistent global DRAM shortage driving up prices, a recent report from The Information suggested that TSMC was exploring the development of 3D DRAM memory at the request of its major client, NVIDIA. Such a move could have positioned TSMC to enter a new business sector, albeit one fraught with the cyclical risks inherent in the memory market, while simultaneously bolstering its influence within the AI supply chain.
However, TSMC has officially refuted these claims, reiterating its steadfast commitment to its core business: the professional integrated circuit foundry sector. A spokesperson for TSMC stated to Tom’s Hardware, “Please be advised that TSMC has always focused on its core business, which is the professional integrated circuit foundry. Our mission is to be the world’s trusted technology and capacity provider for the logic integrated circuit industry for many years.”
The Information’s Report and Internal Efforts
Despite TSMC’s denial, The Information, citing two individuals familiar with the matter, reported that TSMC had indeed begun formulating plans for 3D DRAM manufacturing following a request from NVIDIA. The report indicated that this initiative was reportedly overseen by L.C. Lu, a TSMC vice president responsible for advanced packaging, who previously led operations at Micron Technology in Taiwan.
The Global Memory Landscape
This alleged plan surfaces at a time of widespread memory supply shortages. Memory has transitioned from a commoditized product to a strategic resource, potentially dictating a company’s ability to fulfill customer orders on time. While TSMC collaborates closely with DRAM manufacturers and integrates memory into system-in-package (SiP) solutions for its clients, the company itself lacks the specialized technology and manufacturing capabilities required for DRAM production, let alone the more complex 3D DRAM.
3D DRAM: An Emerging Technology
The Information’s report highlights that traditional DRAM arranges storage cells on a single plane. In contrast, 3D DRAM vertically stacks multiple layers of storage cells, significantly increasing storage density per wafer. Global DRAM leaders – Micron, Samsung, and SK Hynix – are all pursuing their own 3D DRAM projects. However, industry roadmaps suggest that a true mass production of 3D DRAM is still several years away, with intermediate planar DRAM process technologies needing to be released first.
Developing 3D DRAM presents substantial manufacturing hurdles. It necessitates entirely different fabrication techniques to construct multi-layered memory cells while maintaining acceptable yield rates and defect control. Without a competitive planar DRAM process technology, directly developing a mature 3D DRAM manufacturing process is nearly impossible. TSMC, currently focused on leading-edge logic chip fabrication, would need to develop these entirely new technological foundations.
Acquisition as a Potential Entry Strategy
Should TSMC decide to enter the DRAM market, acquiring a smaller, specialized memory manufacturer with existing fabrication facilities could be a viable route. This would provide a tangible entry point into advanced DRAM manufacturing without the extensive effort of developing foundational memory technology and intellectual property from scratch. However, such smaller players often lack the most advanced mainstream DRAM process technologies. Consequently, even post-acquisition, TSMC would likely face billions of dollars in R&D investment and several years of development to establish a competitive advanced DRAM process.
Market Risks and Competitive Dynamics
Entering the DRAM manufacturing sector would expose TSMC to significant market risks. Memory shortages often spur capacity expansion, which can eventually lead to oversupply, driving down prices and profitability. The unpredictable cyclical nature of the memory market would undoubtedly make TSMC proceed with extreme caution.
Furthermore, a move into DRAM production could complicate existing partnerships. For instance, SK Hynix has a collaboration with TSMC for manufacturing logic substrate chips for its next-generation High Bandwidth Memory (HBM) products in 2024. Becoming a direct DRAM competitor could add complexity to such relationships.
Antitrust Scrutiny Concerns
The expansion into DRAM manufacturing might also attract antitrust scrutiny. TSMC already holds a dominant position in advanced logic chip manufacturing and leads in advanced packaging. Further vertical integration encompassing logic, packaging, and memory could prompt regulatory review. A key concern would be whether such integration would lead TSMC to favor its own memory products or create an unfair competitive advantage against rivals lacking similar vertical integration capabilities.
Conversely, TSMC developing its own DRAM could potentially enhance market competition, introducing a new supplier into a market currently dominated by three major players.
Leveraging Existing Strengths for Memory Market Gains
TSMC can still capitalize on the growth in advanced memory demand without directly manufacturing DRAM. The company can secure a larger share of the value created by the DRAM industry through other avenues. This includes manufacturing complex logic substrate chips for HBM, utilizing its CoWoS (Chip-on-Wafer-on-Substrate) or SoIC (System-on-Integrated-Chip) technologies for integrating memory with processors, and collaborating with companies like Global Unichip Corporation (GUC) and established DRAM manufacturers to develop customized memory solutions. This strategic approach allows TSMC to expand its influence in the advanced memory market while avoiding the substantial investments required to develop competitive DRAM process technologies from the ground up and the capital expenditure for new fabs.
