TSMC has publicly denied reports that it is preparing to manufacture advanced memory chips, saying it remains focused on its core logic foundry business. The original report, from The Information, said the company had begun drafting plans for 3D DRAM at the request of Nvidia.
Both statements can be true. A foundry can explore a process without committing to a product line, and a denial issued to protect a core business should be read with that in mind.
Why 3D DRAM is attractive
Conventional DRAM scaling has slowed as capacitor structures approach physical limits. 3D DRAM stacks memory cells vertically, trading area for height, in roughly the way NAND flash moved to 3D years ago. If it works at scale, it offers a path to density that planar DRAM cannot match.
For Nvidia, the appeal is obvious. Accelerator performance is increasingly limited by how much memory can be attached and how fast it can be fed, and the company has watched HBM supply become the binding constraint on how many systems it can ship.
Why a foundry would hesitate
Memory is a different business. It is capital intensive, historically cyclical, and dominated by Samsung, SK Hynix and Micron, all of which have decades of process knowledge. Entering it would put TSMC in direct competition with customers and partners, and would require a manufacturing culture oriented around extremely high volume at low margin.
TSMC’s current position is that it cannot make 3D DRAM, a claim echoed by industry figures this week. That may be a statement about present capability rather than future intent.
The signal to watch
The practical indicator is hiring and patent activity, not press statements. Advanced packaging engineers moving into memory-adjacent roles, or new capacity being described as flexible rather than logic-specific, would suggest the exploration is more than exploratory. Until then, the official denial stands, and the memory shortage that prompted the question continues.