NAND vs DRAM: what is the difference?
DRAM and NAND flash are the two highest-volume memory products in the semiconductor industry, and almost every phone, PC and server contains both. They sit at different levels of the memory hierarchy, are built on different cell structures and scale in different directions, which is why the fabs, process steps and equipment that make them also differ.
The short answer
DRAM is volatile working memory: it stores each bit as charge on a tiny capacitor, loses its data when power is removed and is read and written in tens of nanoseconds. NAND flash is non-volatile storage: it traps charge in a memory transistor, keeps data without power and is far cheaper per bit, but reads take tens of microseconds and writes and erases are slower still. Computers therefore use DRAM as main memory and NAND for SSDs and phone storage.
Side-by-side comparison
| Aspect | NAND flash | DRAM |
|---|---|---|
| Data retention | Non-volatile: data stays when power is off | Volatile: data is lost when power is off and must be refreshed periodically even while powered |
| Memory cell | One charge-storage transistor per cell (charge trap in 3D NAND, floating gate in older planar NAND), strung in series | One transistor plus one capacitor (1T1C) |
| Typical access time | Page reads in tens of microseconds; programming and block erase take longer | Random access in tens of nanoseconds |
| Unit of operation | Read and program by page, erase by much larger block; no overwrite in place | Byte-addressable reads and writes with no erase step |
| Endurance | Limited number of program/erase cycles, managed by the controller | Effectively unlimited write cycles |
| Bits per cell | One to four (SLC, MLC, TLC, QLC) | One |
| Main scaling route | Stacking more word-line layers vertically (3D NAND) | Shrinking cell pitch laterally while keeping enough capacitance |
| Hardest process steps | Depositing tall alternating film stacks and etching very deep, narrow channel holes | Tall capacitor structures, dense patterning and buried word-line transistors |
| Lithography | Mostly DUV; density comes from layer count rather than finer pitch | Leading-edge nodes use EUV for the densest layers |
| Typical uses | SSDs, phone and tablet storage, memory cards, USB drives | Main memory in PCs, phones and servers; graphics memory; stacked HBM for AI accelerators |
How DRAM stores data
A DRAM cell is an access transistor connected to a capacitor. A charged capacitor represents one value and a discharged capacitor the other. Because the stored charge leaks away through the transistor and junctions, every row must be refreshed within a fixed interval, and when power is removed the data disappears. The JEDEC DDR standards define this refresh behaviour along with the interface timing.
The payoff is speed and simplicity of access: any byte can be read or written directly in tens of nanoseconds, with no erase step and no wear-out. That makes DRAM the natural working memory for processors. HBM, used beside AI accelerators, is not a different memory type but DRAM dies stacked and connected with through-silicon vias to give a very wide interface.
How NAND flash stores data
A NAND cell is a single transistor with a charge-storage layer between the gate and the channel. Programming pushes electrons into that layer, which shifts the transistor's threshold voltage; the shift persists without power. Cells are connected in series strings, which saves area but means data is read and programmed a page at a time and erased in large blocks.
Each cell can hold one to four bits by distinguishing several threshold-voltage levels. More bits per cell lower the cost per bit but reduce speed and endurance, so an SSD relies on a controller for error correction, wear levelling and garbage collection. The result is far lower cost per bit than DRAM, at the price of much slower access.
Why the manufacturing problems differ
Planar NAND ran out of room in the mid-2010s, so the industry turned the string on its side. In 3D NAND, alternating layers of oxide and nitride or polysilicon are deposited into a tall stack, a vertical channel hole is etched through the whole stack and the memory films are deposited inside it. Leading products have passed 200 word-line layers, so channel holes are several micrometres deep yet only on the order of 100 nm wide. Deposition uniformity, high-aspect-ratio etching and stack stress control decide yield, while lithography is less critical.
DRAM still scales mainly in two dimensions. Each shrink leaves less area for the capacitor, yet the capacitance cannot fall much without hurting data retention, so capacitors are built as tall, narrow cylinders or pillars with high-k dielectrics. Dense patterning matters more than in NAND, and leading DRAM makers use EUV lithography for the most critical layers. Vertical and 3D DRAM cells are being researched as the lateral route becomes harder.
Where Japan fits
Kioxia, together with its partner Sandisk, makes 3D NAND at large fabs in Yokkaichi and Kitakami, and Micron produces DRAM in Hiroshima. Japanese equipment and materials suppliers serve both memory types: Tokyo Electron supplies etch and deposition tools, and Japanese companies are major suppliers of silicon wafers, photoresists and process chemicals used in memory fabs.
FAQ
Is the RAM in a phone DRAM and the storage NAND?
Yes. The RAM figure in a phone specification is low-power DRAM used as working memory for running apps, while the storage figure is NAND flash that holds the operating system, apps, photos and video. A phone typically has an order of magnitude more NAND than DRAM.
Why is DRAM more expensive per gigabyte than NAND?
A DRAM cell needs a transistor and a capacitor in a fixed area and stores one bit, and it scales mainly by lateral shrinking. A NAND cell is a single transistor that can store up to four bits, and 3D NAND adds capacity by stacking hundreds of layers on the same wafer area. Both factors give NAND a much lower cost per bit.
Can NAND replace DRAM, or DRAM replace NAND?
Not in general. NAND is too slow and wears out too quickly to serve as main memory, and DRAM loses data without power and costs too much per bit to serve as bulk storage. Systems combine them, often with DRAM caches inside SSDs and with storage-class memories studied for the gap between the two.
How is HBM related to DRAM?
HBM is DRAM. Several DRAM dies are thinned, stacked on a base die and connected by through-silicon vias, then placed next to a GPU or accelerator on an interposer. The memory cells are ordinary 1T1C DRAM cells; what changes is packaging and the very wide interface.
Terms compared
- NAND Flash MemoryNon-volatile memory that stores charge in cells connected in series; the storage inside SSDs, smartphones and memory cards.
- DRAM (Dynamic Random-Access Memory)Volatile memory that stores each bit as charge on a tiny capacitor and must be refreshed; the main memory of computers, phones and servers.
- 3D NAND FlashNAND flash in which memory cells are stacked in hundreds of layers along vertical channels instead of being laid out in a single plane.
- HBM (High Bandwidth Memory)A memory made by stacking DRAM dies with through-silicon vias, giving a very wide interface placed next to a GPU or other processor.
Japanese companies involved
- Kioxia · NAND flash memory and SSDs
- Tokyo Electron (TEL) · Coat/develop, etch and deposition equipment
- Micron Memory Japan (Micron Hiroshima) · DRAM fab in Hiroshima (Micron Technology)
- Kokusai Electric · Batch deposition, diffusion and ALD equipment
- DISCO · Dicing, grinding and polishing equipment
More memory comparisons
Sources
Semiconductor comparisons
- All semiconductor comparisons
- HBM vs GDDR6: what is the difference?
- HBM3E vs HBM4: what is the difference?
- SiC vs GaN: what is the difference?
- SiC vs silicon: what is the difference?
- Gallium oxide vs SiC: what is the difference?
- SiC MOSFET vs IGBT: what is the difference?
- CoWoS vs SoIC: what is the difference?
Last updated: November 2, 2026