DRAM featuring 4F² memory cells has long been considered one of the holy grails of the DRAM industry, as they promise to reduce cell area by 33% compared with conventional 6F² designs and therefore enable building DRAMs with more memory cells in the same area. However, after working on 4F² memory cells for a couple of decades, Micron has concluded that the architecture offers no meaningful density advantages, increases manufacturing complexity and costs, and has fundamental scaling limitations. This has a dramatic impact on Micron's future plans.
"Micron invented 4F² DRAM," said Scott DeBoer, president and chief technology and products officer at Micron, during the company's event on Wednesday. "The path with 4F² DRAM is always been an option through time, and we have evaluated it and not pursued it due to increased complexity and cost through this past two-decade period of time."
Interestingly, both Samsung and SK hynix still have 4F² memory cells in their roadmaps (according to TechInsights), so Micron's rejection of the technology puts the company on a unique trajectory. Several years ago, the company delayed integration of EUV lithography into its manufacturing flow, contrary to what its rivals did, and still managed to retain competitive positions both in terms of costs and in terms of performance. Therefore, skipping 4F² memory cells is not completely surprising.
4F² is not the future of Micron's DRAM
The main attraction of 4F² DRAM is its theoretically smaller memory cell. The 4F² naming refers to a cell area equivalent to four times the square of the minimum lithographic feature size (F), compared with six times F² for conventional DRAM.
At identical feature dimensions, a 4F² cell occupies approximately 33% less area than a 6F² cell, which could theoretically increase the number of memory cells per wafer by 50%. Such an improvement would reduce manufacturing costs per bit and increase per-device memory capacity even without an equivalent reduction in feature dimensions. However, to make these advantages possible, DRAM makers must develop a suitable transistor architecture and a sufficiently compact capacitor arrangement. This is not easy at all.
The CTO of Micron says that the theoretical advantages of 4F² DRAM do not necessarily translate into higher memory density in practice. DeBoer explained that dimensions and pitch of DRAM capacitors, rather than the nominal area of DRAM transistors, ultimately determine how many memory cells can fit on a wafer. Both 4F² and 6F² transistor structures must fit beneath the capacitors, which means that reducing transistor area alone does not guarantee a corresponding reduction in overall cell size because transistors need a capacitor. Furthermore, Micron believes that 4F² introduces additional manufacturing steps and complexity that increase costs without delivering sufficient benefits.
"It is important to understand the capacitor pitch, the capacitor size, and dimensions, is what sets the DRAM bit density, not whether it is 6F², 4F², 8F²," DeBoer said. "From our investigation of this technology over the past couple of decades, it is a bit of a myth that 4F² actually provides anything versus 6F². It is capacitor pitch that defines the DRAM. You can build the cells, the transistors underlying them in different ways, but they still have to fit under the capacitor pitch, which sets the DRAM [cell size]. So, some of that is just smoke and mirrors in terms of talking about 6F² versus 4F²."
4F² memory cells may curb new process technologies
Instead of adopting the 4F² DRAM cell architecture, Micron plans to continue scaling conventional 6F² DRAM. The company believes it can increase bit density by reducing capacitor pitch and shrinking process technologies even without introducing a fundamentally different transistor architecture.
Advanced lithography, new materials, and wafer bonding should allow Micron to extend planar DRAM scaling for several more generations, according to the company. For example, its upcoming 1δ process technology, scheduled to enter volume production in the second half of 2027, promises a mid-20% improvement in bit density compared with its predecessor. Micron also plans to adopt High-NA EUV lithography for future DRAM generations.
Interestingly, DeBoer fully acknowledges that 4F² DRAM is technically feasible and that Micron itself has manufactured experimental devices based on the architecture. However, its manufacturing complexity and costs become increasingly problematic as feature dimensions shrink, so essentially Micron claims that it can either adopt 4F² or use more sophisticated process technologies, a claim that perhaps implies why China-based CXMT recently praised 4F². Although DeBoer did not identify any specific physical limitations involved, he stated that 4F² has no sustainable scaling path beyond a certain point. This suggests that Micron sees little value in introducing an intermediate architecture that would eventually require another major transition.
"When you look at a scaling path, you can go back a few nodes and absolutely, 4F² can be built back a few nodes, no question about it," DeBoer noted. "It is more complicated, it is more complex, it is more costly, but you can do that. When we look at scaling 4F² beyond where we are sitting today, there it is; it is physics. There are fundamental problems with 4F² scaling that ultimately will lead to 3D DRAM coming in instead of 4F²."
3D DRAM is the future
Indeed, Micron's preferred long-term solution is 3D DRAM, which would increase memory density by stacking memory cells vertically rather than relying exclusively on reductions in their lateral dimensions. Alongside development of multiple next-generation planar DRAM process technologies, the company is already developing multiple generations of 3D DRAM production nodes (which is not surprising, as 3D NAND nodes will rely on fundamental innovations developed for planar processes), although it has not disclosed when the new architecture will enter commercial production. DeBoer indicated that Micron will make the transition when 3D DRAM offers the right combination of density, manufacturing complexity, and cost.
Although both Samsung and SK hynix have 4F² DRAM products in their roadmaps, Micron believes that the assumption that reducing the nominal footprint of DRAM cells improves manufacturing economics is wrong and that extending 6F² scaling and eventually adopting 3D DRAM offers a better long-term strategy.