美光放弃4F²转向3D DRAM的战略博弈
美光放弃4F²转向3D DRAM的战略博弈
Conclusion
Micron’s decision to stop pursuing 4F² DRAM as a commercial product and prepare for 3D DRAM instead is a defensible long-term bet, but it is not yet a proven technological win. The logic is that 4F²’s smaller cell layout may not translate into enough whole-die density or cost improvement to justify its added process complexity. But 3D DRAM is itself a difficult, longer-horizon manufacturing challenge. For now, the strategy works only if Micron can keep improving conventional DRAM economics while turning its 3D research into a manufacturable product before rivals gain a durable cost or performance lead.
Data as of October 24, 2026.
What Micron has said—and what it has not yet demonstrated
At its technology webinar on , Micron’s president and chief technology and products officer, Scott DeBoer, said the company had studied 4F² for but would not commercialize it. His stated case is that DRAM density depends heavily on capacitor spacing, so the cell-layout advantage can be less valuable in practice than it appears on paper; he also cited cost, complexity and scaling problems. Micron instead places 3D DRAM on a parallel, economics-gated roadmap, between its 1-zeta and 1-eta generations. These are management’s technical judgments and roadmap intentions, not independent proof that 4F² cannot work or that Micron’s 3D approach will be cheaper.
The distinction matters: 4F² describes the footprint of a DRAM cell; it does not mean the memory array is stacked into many layers. A technical overview estimates an idealized 4F² layout could offer about more density than 6F² without a feature-size shrink, while noting that real gains may be lower because cell layout is not the only limit. That helps frame Micron’s argument: the relevant comparison is not the cell diagram, but realized die density, yield and cost per bit.
Nor should Micron’s separately announced vertical HBM be mistaken for proof of 3D DRAM. Vertical HBM stacks memory over a processor using advanced packaging; Micron’s 3D DRAM roadmap concerns a future DRAM-cell architecture. One can advance without validating the other.
Potential advantages of the shift
- Avoid spending heavily on a limited cell-layout win. If shrinking the nominal cell footprint does not deliver a commensurate increase in saleable bits per wafer, Micron could avoid paying for a complex 4F² process without receiving a lasting cost-per-bit advantage.
- Aim beyond planar scaling limits. A successful 3D cell architecture could add density vertically rather than relying primarily on further planar shrink. That could offer a longer runway for capacity growth if it can be integrated economically.
- Build capabilities useful across the roadmap. Micron says its next DRAM generations include wafer bonding before the proposed 3D insertion point. If the company develops bonding and process-integration expertise along the way, some of that work may support future 3D products. The value depends on whether those techniques improve manufacturing results, not simply whether they appear on the roadmap.
- Preserve timing flexibility. Micron describes 3D DRAM as a parallel path to be introduced when the economics make sense. That gives it room to avoid committing to volume production before costs and yields are acceptable.
Main risks and counterarguments
- The alternative is not easy. 3D DRAM is still a research and development challenge, not a demonstrated high-volume replacement. Stacking or vertically integrating memory elements brings difficult process-integration, bonding, defect-control and reliability questions. The industry’s published discussion has emphasized that cost, durability and reliability remain unresolved issues for commercialization.
- A longer wait could hand rivals an interim advantage. Samsung has reportedly produced a working die using a sub-10nm process with 4F² and vertical-channel-transistor technology, and is targeting mass production in . That is a meaningful development milestone, but it is not evidence of production yield, customer qualification or competitive cost. Still, if Samsung or SK hynix gets a reliable 4F² product to market first, Micron could forgo a useful bridge technology while it pursues a more ambitious architecture.
- The economic hurdle may be higher than the technical one. A denser lab device is not enough. A commercial design must produce good dies at scale, meet retention and performance requirements, and lower cost per usable bit after accounting for extra process steps, equipment and bonding.
- Roadmap timing remains open-ended. Micron has not given a firm volume-production date for 3D DRAM; it says insertion timing depends on economics. That is prudent, but it also leaves investors and customers without a clear schedule against which to judge progress.
- The technology debate is not settled by Micron’s characterization. Micron has an incentive to explain why it is not following rivals’ 4F² path. Its claim that the density advantage is small should be tested against measured die-level density, yield and cost from competitors—not treated as an established industry conclusion.
What evidence would show whether the strategy is working?
-
Conventional-node execution. Micron says its 1-delta DRAM is expected to ramp in the . Watch for an on-time ramp, achieved bit density and yield, and evidence that cost per bit improves. Strong execution would give Micron time to develop 3D DRAM without surrendering its existing economics.
-
Wafer-bonding results before 3D insertion. Micron says wafer bonding is part of the subsequent 1-epsilon roadmap. Evidence should include production-ready products, measured yield and reliability, and an improvement in usable bits per wafer or cost per bit—not merely confirmation that bonding is in development.
-
A working 3D DRAM array, not just a stack or a roadmap slide. Look for technical disclosures showing functioning memory cells across the proposed vertical structure, including write/read operation, retention and refresh behavior, reliability, and the performance and power trade-offs. A packaging demonstration alone would not establish that the DRAM cell architecture works.
-
Manufacturing economics versus the alternatives. The decisive comparison is cost per good bit at production scale, including wafer cost, process complexity, yield and throughput. Compare Micron’s results with conventional 6F² DRAM and any rival 4F² products that progress beyond development samples.
-
Signs of product adoption. Sampling, qualification, named product integration and volume commitments would show that customers value the resulting density, performance and power enough to adopt it. A technical advantage without customer demand or acceptable pricing would not validate the strategy commercially.
-
A roadmap that gets more concrete over time. Watch for Micron to specify the 3D architecture, process milestones, targeted insertion window and eventual production site. Repeatedly deferring those details—or describing the technology only as economically gated—would weaken the case that the research is converting into an executable manufacturing plan.
Bottom line: Micron’s rejection of 4F² may be good capital allocation if the real die-level benefit cannot justify its added cost and complexity. But the choice raises the bar for Micron’s 3D program: the strategy will be validated by competitive cost per bit, manufacturing yield and customer adoption—not by Micron’s critique of 4F² or by the promise of vertical stacking alone.