TSMC denies exploring 3D DRAM at Nvidia's request

A recent report from The Information suggested that TSMC is exploring 3D DRAM production at Nvidia's request as surging demand for AI hardware strains global DRAM supplies and drives up prices. If the proposal were to come to fruition, it could mean a new, albeit prone to cyclical risks, business for TSMC and could dramatically expand its role in the AI supply chain. For its part, TSMC states that it is a pure-play logic foundry and plans to remain a logic foundry, which essentially means a denial of the report.

"Please note that TSMC focuses on our core business at all times, which is dedicated IC foundry," a TSMC spokesperson told Tom's Hardware. "Our mission is to be the trusted technology and capacity provider of the global logic IC industry for years to come."

Despite the denial, according to two people familiar with the initiative cited by The Information, TSMC has begun preparing plans for 3D DRAM manufacturing following a request from Nvidia, its largest customer. The initiative is reportedly overseen by Hsu Kuo-chin, TSMC’s vice president responsible for advanced packaging and former head of Micron’s Taiwanese operations. The effort comes amid widespread memory shortages that turned memory from a commodity into a strategic asset that could determine whether a company can ship its products to customers or not.

Although TSMC works closely with DRAM producers as it packages memory into system-in-packages (SiPs) for its customers, the company lacks the expertise and production capacity to make DRAM, let alone 3D DRAM. Meanwhile, given that 3D DRAM is years away even for major memory makers, pursuing a 3D DRAM strategy cannot solve memory shortages for at least five years.

Unlike conventional DRAM, which arranges memory cells in a single plane, 3D DRAM places multiple levels of memory cells vertically to increase capacity per wafer. All three leading DRAM makers — Micron, Samsung, and SK hynix — are working on their 3D DRAM projects. However, Micron's DRAM strategy includes at least four more planar DRAM process technologies before 3D DRAM becomes a reality.

While TSMC has arguably the industry's best logic process technologies, DRAM production requires entirely different manufacturing processes, which TSMC would have to develop, should it pursue making memory for its customers. Meanwhile, producing 3D DRAM requires completely different fabrication techniques to construct multiple memory-cell layers while controlling defects and maintaining acceptable yields. It is hardly possible to develop a 3D DRAM manufacturing process without having a competitive planar DRAM process technology, and TSMC does not have one.

One of the options for TSMC to enter the DRAM business would be to acquire a small specialty memory producer that has its own fabrication capability. This would give TSMC a credible route into advanced DRAM manufacturing without having to develop the underlying memory technology and IPs from scratch. However, small makers do not have leading-edge commodity DRAM nodes, so TSMC would still need to invest billions of dollars and years of R&D into development of a leading-edge process technology for DRAM.

However, entering DRAM manufacturing would expose TSMC to substantial market risks. Memory shortages frequently encourage capacity expansions that eventually create oversupply, depressing prices and profitability. Such uncertainty could surely make TSMC think twice before going down the DRAM route. Furthermore, entering the DRAM business could also complicate TSMC's relationship with large DRAM makers in general, and SK Hynix, which partnered with the foundry in 2024 to manufacture logic base dies for next-generation HBM products, in particular.

TSMC’s potential entry into DRAM manufacturing could also raise antitrust questions, particularly if the company combines its dominant position in leading-edge logic manufacturing, leading position in advanced chip packaging, and ramping memory production. Regulators could examine whether such vertical integration would enable TSMC to favor its own memory products or disadvantage competitors that lack such a vertical integration. Then again, in-house DRAM development by TSMC could also strengthen competition by introducing another supplier into a market dominated by just three manufacturers.

Yet, TSMC does not necessarily need to manufacture DRAM to capitalize on growing demand for advanced memory. Instead, it could capture a larger share of the value generated by DRAM by producing sophisticated logic base dies for HBM, integrating memory with processors using its SoIC hybrid bonding technology, and developing customized memory solutions with partners such as GUC and established DRAM manufacturers. This approach would allow TSMC to expand its presence in the advanced memory market without investing billions of dollars in developing competitive DRAM process technologies from scratch and/or building additional fabs for tens of billions of dollars.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.