Timelines · Sourced data
Silicon Real Estate
Memory dies drawn at the same scale beside a U.S. quarter, so a bigger box is more silicon and the coin shows how big that is. Switch between NAND, from planar chips of the early 2000s to recent 3D dies, and DRAM, where the few areas that teardown labs have measured are worth drawing. Every area links to the paper, release or teardown it comes from.
NAND: how to read it
A flash die is a rectangle of silicon cut from a wafer. Its area decides how many fit on one wafer, which is one of the things that sets a die’s cost, and bit density (gigabits per square millimeter) is how a maker fits more bits into the same silicon. The stage draws each die at one scale, so a box twice the size really is twice the area.
Most sources print a die’s area but not its width and height. Those dies are drawn as squares of the same area, so a shape can mislead even when the size is right. Planar and 3D dies sit on the same field, but they are different structures: a 3D die gets its capacity from stacked layers, not from smaller cells alone.
The full list of these dies, with every other figure, is on the NAND die timeline.
The dies at one scale27 drawn
The ruled squares are 10 mm across, and the circle is a U.S. quarter drawn at the same scale. Both boards start at 4 pixels per millimeter, near real size on a desktop screen, so hold a quarter up to the screen to compare. Drag the scale to zoom, pick a maker, or select a die for its figures.
- Samsung
- SK hynix
- Micron
- Kioxia/WD
- YMTC
- Intel
Every die is drawn at the same scale as the quarter beside it, so a bigger box is a bigger piece of silicon. Each die is drawn as a square of its area, because no source printed a width and height.
Put two dies side by side
Density by year27 dated dies
Bit density is capacity divided by die area, so a smaller die for the same capacity sits higher. It is computed here from the two printed figures, not typed in.
Bit density of the drawn dies by year (Gb/mm²)
- Samsung
- SK hynix
- Micron
- Kioxia/WD
- YMTC
- Intel
Every figure and its source27 rows
Tap a heading to sort. Sort by capacity to rank the dies by how much each holds, or by the quarter column to rank them by how much a quarter-sized patch of each would hold, in gigabytes (gigabits over 8). Each area links to the paper, release or teardown page it was read from.
| Die | Maker | Year | Capacity | Area (mm²) | Gb/mm² | Quarter-sized patch (GB) | Source |
|---|---|---|---|---|---|---|---|
| 321-layer 1Tb TLC (die area from a secondary source) | SK hynix | 2024 | 1 Tb | 43.54 | 23.52computed | 1,359computed | SK hynix H25GTD0 321-Layer (V9) 1 Tb TLC 3D NAND Process Flow Analysis |
| X4-9060 512Gb TLC (Xtacking 4.0) (die area from a secondary source) | YMTC | 2024 | 512 Gb | 40.44 | 12.66computed | 732computed | YMTC Xtacking4.0: Breaking New Ground in 3D NAND Technology |
| 144L FG 1Tb QLC (product interface speed from a review, secondary source) | Intel | 2021 | 1 Tb | 74.03 | 13.83computed | 799computed | 30.2 A 1Tb 4b/Cell 144-Tier Floating-Gate 3D-NAND Flash Memory with 40MB/s Program Throughput and 13.8Gb/mm2 Bit Density |
| 176L TLC 512Gb (4D PUC) (die area from a secondary source) | SK hynix | 2021 | 512 Gb | 46.5 | 11.01computed | 636computed | Comparison: Latest 3D NAND Products from YMTC, Samsung, SK hynix and Micron |
| BiCS4 1.33Tb QLC (96 WL layers) | Kioxia/WD | 2020 | 1365 Gb | 158.4 | 8.62computed | 498computed | A 1.33-Tb 4-Bit/Cell 3-D Flash Memory on a 96-Word-Line-Layer Technology |
| 176-layer TLC (replacement gate) (die area from a secondary source) | Micron | 2020 | 512 Gb | 49.84 | 10.27computed | 594computed | Micron B47R 3D CTF CuA NAND Die, the World's First 176L (195T) |
| ISSCC 2019 512Gb TLC 128 WL layers, CuA, 4-plane | Kioxia/WD | 2019 | 512 Gb | 66 | 7.76computed | 448computed | 13.5 A 512Gb 3-bit/Cell 3D Flash Memory on 128-Wordline-Layer with 132MB/s Write Performance Featuring Circuit-Under-Array Technology |
| 64-layer 256Gb TLC (Xtacking) (die area from a secondary source) | YMTC | 2019 | 256 Gb | 57.96 | 4.42computed | 255computed | Now, YMTC is a Leading Pioneer in 3D NAND |
| BiCS3 512Gb TLC (Toshiba/WD, 64 WL layers) | Kioxia/WD | 2017 | 512 Gb | 132 | 3.88computed | 224computed | 11.1 A 512Gb 3b/cell flash memory on 64-word-line-layer BiCS technology |
| 768Gb 3b/cell 3D floating-gate NAND with CMOS under array (Micron and Intel, ISSCC 2016 paper 7.7) | Micron | 2016 | 768 Gb | 179.2 | 4.29computed | 248computed | 7.7 A 768Gb 3b/cell 3D-Floating-Gate NAND Flash Memory |
| 14nm 128Gb MLC planar NAND (ISSCC 2016 paper 7.5) | Samsung | 2016 | 128 Gb | 130 | 0.985computed | 57computed | 7.5 A 128Gb 2b/cell NAND Flash Memory in 14nm Technology with tPROG=640µs and 800MB/s I/O Rate |
| 4th-gen V-NAND 512Gb TLC (64 WL) | Samsung | 2016 | 512 Gb | 128.5 | 3.98computed | 230computed | A 512-Gb 3-b/Cell 64-Stacked WL 3-D-NAND Flash Memory |
| 32-layer 384Gb TLC (Micron and Intel) (die area from a secondary source) | Micron | 2015 | 384 Gb | 168.2 | 2.28computed | 132computed | Intel/Micron 64L 3D NAND Analysis |
| BiCS 48-layer 256Gb TLC (Toshiba) (die area from a secondary source) | Kioxia/WD | 2015 | 256 Gb | 105.4 | 2.43computed | 140computed | Toshiba 64L NAND, BiCS FLASH |
| 16nm 64Gb MLC planar NAND (ISSCC 2014) | SK hynix | 2013 | 64 Gb | 93.4 | 0.685computed | 40computed | 19.2 A 93.4mm2 64Gb MLC NAND-flash memory with 16nm CMOS technology |
| 19nm 64Gb MLC planar NAND (ISSCC 2012) | Kioxia/WD | 2012 | 64 Gb | 112.8 | 0.567computed | 33computed | A 19nm 112.8mm2 64Gb multi-level flash memory with 400Mb/s/pin 1.8V Toggle Mode interface |
| 24nm 64Gb MLC planar NAND (ISSCC 2011) | Kioxia/WD | 2011 | 64 Gb | 151 | 0.424computed | 24computed | A 151mm2 64Gb MLC NAND flash memory in 24nm CMOS technology |
| 20nm 64Gb MLC (Intel and Micron) | Micron | 2011 | 64 Gb | 118 | 0.542computed | 31computed | Intel, Micron Extend NAND Flash Technology Leadership, Introduce Industry's Smallest, Most Advanced 20-Nanometer Process |
| 25nm 64Gb MLC (Intel and Micron) | Micron | 2010 | 64 Gb | 167 | 0.383computed | 22computed | Intel, Micron Introduce 25-Nanometer NAND, The Smallest, Most Advanced Process Technology in the Semiconductor Industry |
| 32nm 32Gb MLC planar NAND (ISSCC 2010) | Samsung | 2010 | 32 Gb | 159 | 0.201computed | 12computed | A 159mm2 32nm 32Gb MLC NAND-flash memory with 200MB/s asynchronous DDR interface |
| 32nm 32Gb MLC planar NAND (Lee et al., ISSCC 2010) | SK hynix | 2010 | 32 Gb | 146 | 0.219computed | 13computed | A 32Gb MLC NAND-flash memory with Vth-endurance-enhancing schemes in 32nm CMOS |
| 34nm 32Gb MLC (Intel and Micron) | Micron | 2008 | 32 Gb | 172 | 0.186computed | 11computed | Intel and Micron First to Deliver Sub-40 Nanometer NAND Flash Memory Device |
| 43nm 16Gb MLC planar NAND (ISSCC 2008) | Kioxia/WD | 2008 | 16 Gb | 120 | 0.133computed | 7.7computed | A 120mm2 16Gb 4-MLC NAND Flash Memory with 43nm CMOS Technology |
| 56nm 8Gb MLC planar NAND (ISSCC 2006) | Kioxia/WD | 2006 | 8 Gb | 98.8 | 0.081computed | 4.7computed | A 56nm CMOS 99mm2 8Gb Multi-level NAND Flash Memory with 10MB/s Program Throughput |
| 70nm 8Gb MLC planar NAND (JSSC 2006) | Kioxia/WD | 2005 | 8 Gb | 146 | 0.055computed | 3.2computed | A 146-mm2 8-Gb Multi-Level NAND Flash Memory With 70-nm CMOS Technology |
| 120nm 1Gb planar NAND (Lee et al., JSSC 2002) | Samsung | 2002 | 1 Gb | 129.6 | 0.008computed | 0.4computed | High-performance 1-Gb-NAND flash memory with 0.12-um technology |
| 130nm 1Gb planar NAND (JSSC 2002) | Kioxia/WD | 2002 | 1 Gb | 125 | 0.008computed | 0.5computed | A 125-mm2 1-Gb NAND flash memory with 10-MByte/s program speed |
Left out, and why
The NAND die timeline holds 30 more dies that carry a die area. They stay out of the drawing until a saved copy of their source shows the same figure, because a die drawn to scale is a claim about its size. Their numbers are still on the timeline, each linked to where it came from.
- BiCS4 512Gb TLC (Toshiba Memory, 96 WL layers; product interface speed from a review, secondary source)
- XL-FLASH G1 128Gb SLC (96 WL layers)
- ISSCC 2021 1Tb TLC 170+ WL layers, CuA, 4-plane
- BiCS6 1Tb QLC (162 WL layers, WD/Kioxia)
- 96L QLC 1Tb (PUC)
- 192L FG 1.67Tb 5b/cell (PLC)
- 96L FG 1Tb QLC (CuA; product interface speed from a review, secondary source)
- 2nd-gen V-NAND 128Gb TLC (32 WL)
- 3rd-gen V-NAND 256Gb TLC (48 WL)
- 6th-gen V-NAND 512Gb TLC (product interface speed from a review, secondary source)
- X2-9060 128-layer TLC (Xtacking 2.0; die area, density and planes from a teardown, secondary source)
- X3-9070 1Tb TLC (Xtacking 3.0; layer count, die area and density from a teardown, secondary source)
- 64-stacked-WL 1Tb QLC (ISSCC 2018)
- 92-stacked-WL 1Tb QLC (ISSCC 2020 paper 13.1)
- 128-layer 512Gb TLC (die B37R; teardown, secondary source)
- 128-layer 512Gb TLC (single deck, die K9AHGD8U0B; teardown, secondary source)
- 1st-gen V-NAND 128Gb MLC (24 WL) (die size and density from a Kioxia table, secondary source)
- 32-layer 256Gb MLC (Micron and Intel) (die area from a teardown, secondary source)
- 90nm 2Gb planar NAND (JSSC 2003)
- 70nm 16Gb 4b/cell planar NAND (VLSI 2007)
- 56nm 16Gb TLC planar NAND (Li et al., ISSCC 2008)
- 56nm 16Gb MLC all-bitline NAND (ISSCC 2008)
- Sub-35nm 32Gb TLC planar NAND (ISSCC 2009)
- 48nm 32Gb TLC planar NAND (ISSCC 2009)
- 43nm 64Gb 4b/cell planar NAND (ISSCC 2009)
- 34nm 32Gb TLC planar NAND (Intel and Micron, ISSCC 2010)
- 19nm 128Gb TLC planar NAND (ISSCC 2012)
- Sub-20nm 64Gb MLC planar NAND (ISSCC 2012)
- 20nm 128Gb TLC planar NAND (ISSCC 2013) (die size and density from a Toshiba table, secondary source)
- 15nm 64Gb MLC planar NAND (ISSCC 2015)
What it will not tell you
Area is not cost. Wafer price, yield and the controller, packaging and test around the die are all missing here, and a large die is not a bad one: early planar parts were built for the process of their day. Different sources also measure a die differently, for example with or without the seal ring or the pad area, so a small gap between two dies is inside what the sources can tell apart.
Capacity is the headline figure each source prints, not usable space after spare blocks and error correction. A trend line drawn through these dies is a trend among them, not every die ever made. For a ranking, see the density ranking on the NAND die timeline.
DRAM: how to read it
A DRAM die stores one bit per cell as charge in a tiny capacitor, so its area is mostly the cell array plus the sense amplifiers and wiring around it. Teardown labs measure the finished die and publish its width, height or area, and from that and the capacity the page works out how many gigabits fit in each square millimeter.
Only one die here has a printed width and height, so it is drawn as a true rectangle; the rest are squares of their measured area. A mobile LPDDR die and a server DDR5 die are built for different jobs, so compare density across them with care.
The full list of these dies, with every other figure, is on the DRAM die timeline.
The dies at one scale6 drawn
The ruled squares are 10 mm across, and the circle is a U.S. quarter drawn at the same scale. Both boards start at 4 pixels per millimeter, near real size on a desktop screen, so hold a quarter up to the screen to compare. Drag the scale to zoom, pick a maker, or select a die for its figures.
- Samsung
- SK hynix
- Micron
Every die is drawn at the same scale as the quarter beside it, so a bigger box is a bigger piece of silicon. A die with a printed width and height is drawn as that rectangle; every other die is a square of the same area, because its aspect ratio was not captured.
Put two dies side by side
Density by year5 dated dies
Bit density is capacity divided by die area, so a smaller die for the same capacity sits higher. It is computed here from the two printed figures, not typed in.
Bit density of the drawn dies by year (Gb/mm²)
- Samsung
- SK hynix
- Micron
Every figure and its source6 rows
Tap a heading to sort. Sort by capacity to rank the dies by how much each holds, or by the quarter column to rank them by how much a quarter-sized patch of each would hold, in gigabytes (gigabits over 8). Each area links to the paper, release or teardown page it was read from.
| Die | Maker | Year | Capacity | Area (mm²) | Gb/mm² | Quarter-sized patch (GB) | Source |
|---|---|---|---|---|---|---|---|
| Samsung D1b, LPDDR5X (secondary source) | Samsung | 2024 | 16 Gb | 36.68 | 0.436computed | 25computed | Samsung D1b DRAM Confirmed, Cell Size Much Smaller than Micron D1β (Nov 2024) |
| SK hynix D1y, DDR5 (secondary source) | SK hynix | 2022 | 16 Gb | 75.21 | 0.213computed | 12computed | Industry-leading DDR5 Technology: Micron vs. Samsung vs. SK Hynix (TechInsights, 2022) |
| Samsung D1y, DDR5 (secondary source) | Samsung | 2022 | 16 Gb | 73.58 | 0.218computed | 13computed | Industry-leading DDR5 Technology: Micron vs. Samsung vs. SK Hynix (TechInsights, 2022) |
| Micron D1z, DDR5 (secondary source) | Micron | 2022 | 16 Gb | 66.26 | 0.242computed | 14computed | Industry-leading DDR5 Technology: Micron vs. Samsung vs. SK Hynix (TechInsights, 2022) |
| Micron D1α, DDR4 (secondary source) | Micron | 2021 | 8 Gb | 25.41 | 0.315computed | 18computed | Micron D1α, '14 nm'! The Most Advanced Node Ever on DRAM! (TechInsights, 2021) |
| Micron Y52K D1β, DDR5 (secondary source) | Micron | not published | 16 Gb | 35.83 (6.33 × 5.66 mm) | 0.447computed | 26computed | Micron Y52K D1β 16 Gb DDR5 DRAM transistor characterization (TechInsights) |
Left out, and why
These dies appear in industry slides or research notes, but this page could not read their area from a source saved in the library. A die stays off the stage until its figure is confirmed from a saved copy, so a missing box means unconfirmed, not small.
- SK hynix D1b DDR5
- Micron D1β LPDDR5
- CXMT G1, G3 and G4 DRAM dies
- Micron 1γ, SK hynix 1c and Samsung 1c, for which no area has been found
What it will not tell you
These are a handful of dies, not the market. Makers rarely print die areas, so every figure here is a teardown measurement of one part, and the same part can measure slightly differently depending on whether the seal ring is counted. The Micron Y52K entry carries both the die-seal measurement and the full-die figure in its source, and the page draws the sealed one.
Area also does not show speed, power or cost.
The guides behind this data
- Flash Foundations, Part 3: From Cell to Die
Follow cells into pages, blocks, and dies to understand how NAND is organized.
- Flash Foundations, Part 4: 3D NAND Stacking
See how stacking layers raised density and created new problems for fabs and firmware.
- Working Memory, Part 2: DRAM From the Inside
How a DRAM cell is read and refreshed, how banks and ranks organize it, what DDR timing numbers measure, and how one cell serves three buses.