Walk through the equipment list of a DRAM fab, a NAND flash fab, and a leading-edge logic fab, and you will find a great deal of overlap: lithography scanners, etch chambers, deposition tools, ion implanters, and CMP polishers appear in all three. Yet the chips that emerge from these fabs are built for radically different purposes and bear surprisingly little resemblance to one another at the device level. Understanding why memory and logic manufacturing diverge, despite relying on a shared toolkit of fabrication techniques, reveals a lot about how chip design requirements shape process technology.
Different Chips, Different Priorities
Logic chips, such as CPUs and GPUs, are built from transistors wired together to perform computation, with an emphasis on switching speed, power efficiency, and the ability to implement enormously varied and complex circuit designs. DRAM chips store data temporarily, as long as power is supplied, using a simple repeating cell structure that must be replicated billions of times with extreme uniformity. NAND flash stores data persistently, even without power, using an entirely different physical mechanism for retaining a charge. Each of these priorities pushes process technology in a different direction, even when the chips are manufactured in fabs using many of the same categories of equipment.
Logic Manufacturing: Complexity Over Repetition
Logic chips are defined by design heterogeneity. A modern processor includes many different types of circuit blocks, arithmetic units, cache memory, control logic, input-output interfaces, each with its own layout requirements. Logic processes are optimized for transistor performance and for supporting an enormous diversity of interconnected, custom circuit patterns across many metal layers.
This drives logic fabs toward technologies like FinFET and gate-all-around transistors, which maximize switching speed and reduce leakage current, along with a dense stack of a dozen or more metal interconnect layers needed to wire together the huge number of distinct logic blocks in a design. Because logic chips are not built from a single repeating cell, process development must validate an enormous variety of pattern shapes and interactions, making logic process development notoriously complex and iterative.
DRAM Manufacturing: Precision at Massive Repetition
A DRAM chip, by contrast, is built almost entirely from one repeating unit: a single transistor paired with a single capacitor, repeated billions of times across the chip. Each cell stores one bit of data as an electrical charge in the capacitor; the transistor acts as a switch controlling access to that charge for reading or writing. Because DRAM cells must be refreshed constantly, as the capacitor’s charge leaks away over time, DRAM process technology is less focused on raw transistor switching speed and far more focused on building capacitors with enough capacitance to reliably store a detectable charge, despite the tiny area available to each cell.
This has driven DRAM fabs toward exotic, highly three-dimensional capacitor structures, deep trenches etched into the silicon or tall stacked cylinder structures built above it, to maximize the surface area, and therefore the charge storage capacity, of a capacitor crammed into an extremely small footprint. These structures require etch and deposition processes capable of extreme aspect ratios, often far more extreme than anything found in a logic process, along with specialized dielectric materials engineered specifically for high capacitance in a minimal area.
DRAM also places enormous emphasis on cell uniformity, since any variation between the billions of identical cells on a chip can translate into inconsistent charge storage and retention, directly affecting whether the memory reads back correctly.
NAND Flash: Stacking for Density
NAND flash stores data using floating-gate or charge-trap transistors, which retain a charge on an isolated conductive or dielectric layer within the transistor itself, allowing data to persist without power. Because each flash cell can be read, written, and erased many times, and because the charge-trapping mechanism must reliably distinguish between multiple voltage states (especially in modern multi-level cell designs that store more than one bit per cell), flash process technology is centered on the reliability and endurance of this charge storage layer.
The most dramatic divergence in NAND manufacturing has been the shift to 3D NAND, in which memory cells are stacked vertically in towers, sometimes exceeding a couple hundred layers, rather than spread across the wafer in a flat, two-dimensional array. This vertical stacking dramatically increases storage density without requiring ever-smaller individual cell dimensions, which had become increasingly difficult to achieve reliably in planar NAND designs. Building these vertical stacks requires depositing alternating layers of conductive and insulating material dozens or hundreds of times, then etching extremely deep, high-aspect-ratio channels straight down through the entire stack, a process that places extraordinary demands on etch uniformity and depth control.
Shared Equipment, Divergent Recipes
Despite these differences, DRAM, NAND, and logic fabs do rely on overlapping categories of equipment. Lithography scanners, deposition chambers, etch tools, ion implanters, and CMP polishers appear across all three. What differs is the specific recipe, the combination of materials, timings, chemistries, and patterns, applied using that equipment. A deposition tool used in a logic fab might be tuned for a thin, highly conformal dielectric liner, while the same category of tool in a 3D NAND fab might be tuned for rapidly building up dozens of alternating layers at a larger scale.
This is also why fabs that specialize in memory rarely manufacture advanced logic, and vice versa: the accumulated process knowledge, equipment tuning, and defect management techniques are so specific to the chip type that switching between them requires substantial retooling and process redevelopment, even when the underlying machines share a name and a general category.
Why This Matters for the Industry
This specialization explains why the semiconductor industry has distinct leaders in logic foundry services versus memory manufacturing, and why memory pricing, supply, and technology roadmaps often move on a different cycle than logic chip announcements. It also explains why advanced packaging techniques that combine memory and logic, such as stacking high-bandwidth memory next to a processor, are so valuable: they allow each chip type to be manufactured using the process technology best suited to it, then integrated together afterward rather than compromising either design to fit a single shared process.
Conclusion
DRAM, NAND flash, and logic chips may emerge from fabs that look remarkably similar on a floor plan, filled with the same broad categories of lithography, etch, deposition, and implantation tools. But the physical structures these chips rely on to store or process information are fundamentally different, and that difference cascades through every layer of process development, from capacitor geometry to 3D stacking to transistor architecture. Recognizing this helps explain why “chipmaking” is not a single, uniform discipline, but a family of related manufacturing technologies, each shaped by the very different jobs its end product needs to perform.
Leave a Reply