Memory and storage, tested and explained

How SSDs and Memory Work, from the Silicon to the System

NAND, SSD validation and the memory that could follow it, explained by Sean Webster through guides, sourced charts, free browser labs and open-source tools.

15+Years building systems
500+Reviews, articles and white papers
FreeLabs that run in your browser

Atlas

Turn the drive, see the stack

One interactive 3D reference with three specimens: an NVMe SSD, an HBM stack and a DDR5 memory stick. Rotate a board or a package, click a part, and go down through it to one cell.

The SSD

Turn the drive

Controller, NAND packages and the path between them.

HBM

Walk the stack

Core dies, base die and the vertical connections through the stack.

DRAM

Read the stick

A DDR5 server memory stick: its packages, the die inside and the one-transistor, one-capacitor cell that holds a bit.

Also:How the Atlas works Labs Timelines

Latest

New on the site

Writing · November 6, 2026

An MCP Server for the Eyes on Flash Data

Eyes on Flash now has a public, read-only MCP server: what it serves, how to connect, which sources it leaves out, and how it runs on Cloudflare's Free plan.

Best picks

The hardware I would buy

Each pick comes from a tested review and the maker's data sheet, held to a bar the page states up front. A commission never changes which products are picked or their order.

Best RAM

Desktop DDR5 and DDR4 kits by tier, with the DRAM chips inside.

Best HDDs

Internal hard drives by tier, with recording type, workload rating and warranty.

Best GPUs

Gaming and SI graphics cards by tier, picked on memory, with chip and VRAM.

Best Peripherals

Flash drives, docks, hubs, card readers, M.2 enclosures and Switch 2 cards.

Best NAS

NAS boxes from two bays up, and the multi-gig switches to run them.

Best UPS

Battery backup for a NAS, a desktop, a home lab rack or an AI workstation.

Learn

From the cell to the system

Series read in order and myths checked against data. Start at Part 1 or jump to the part you need; every chart and number links to its source.

All Guides

Flash and storage

Flash Foundations, Part 1: a host and a drive joined by per-core queues, four link generations each twice the last, and an M.2 card beside an E3.S cardA three-panel schematic. Left: a host box above a drive box, joined by five columns of queue slots, one queue per core. Center: four bars rising in doubling steps for PCIe Gen3, Gen4, Gen5 and Gen6, the last in copper. Right: an M.2 card above a larger E3.S card, drawn at one scale, each with gold contact fingers on its left edge.GUIDE / FLASH FOUNDATIONS, PART 1Limits set before the firmware runsPROTOCOL · LINK · PACKAGE01 PROTOCOLHostDrive02 LINKGen3Gen4Gen5Gen603 PACKAGEM.2E3.S

10 parts

Flash Foundations

How an SSD works, from the physics of a flash cell to the controller, the translation layer and error correction.

The memory and storage hierarchy: bandwidth from a hard drive to one GPU’s HBM, the five-minute rule from 1987 to 2017, and one SSD’s write speed through its phasesA three-panel schematic. Left: two bars on a log scale, a hard drive at 270MB/s and one GPU’s HBM at 19.2TB/s, the HBM bar far longer. Center: a line of the DRAM to hard drive break-even interval from 400 seconds in 1987 to 4 hours in 2017 on a log scale, with a dashed line at five minutes. Right: a line of one SSD’s write speed against data written, high while the cache absorbs writes, then a slower phase, then a steady middle speed.GUIDE / HOW STORAGE WORKS, PART 1Every memory wins on one axisBANDWIDTH · FIVE-MINUTE RULE · WRITE SPEED01 BANDWIDTHHDD 270MB/sHBM 19.2TB/slog scale02 BREAK-EVEN5 min400s4h1987201703 WRITE SPEEDcache, fold, then TLC

4 of 8 parts published

How Storage Works

What is inside drives and the links and protections around them, starting with the memory and storage hierarchy.

Two DDR5 modules seated in slots A2 and B2 of a four-slot motherboardTop view of a motherboard's four memory slots, labeled A1 and A2 for channel A and B1 and B2 for channel B. Two DDR5 modules, each with eight memory chips, sit in A2 and B2; A1 and B1 are empty.GUIDE / STORAGE AND MEMORY IN PRACTICE, PART 1Two modules, four slotsTOP VIEW · NOT TO SCALEA1A2B1B2CH ACH B

8 of 9 parts published

Storage and Memory in Practice

Choosing, setting up and keeping drives and memory in real machines, starting with DDR5 for builders.

Three views of NAS operating systems: software inside a maker's box, hardware and software picked apart, and a support bar that endsThree numbered panels. First, the box: a NAS case with four drive bays and a software layer drawn inside it. Second, the build: a hardware block with a separate software block above it and a gap between them. Third, the updates: a bar that is solid while updates last and dashed where support ends.GUIDE / RUNNING A NAS, PART 1The software that runs a NASBOX · BUILD · UPDATES01 THE BOXSoftware inside02 THE BUILDYou pick both03 THE UPDATESSupport can end

9 parts

Running a NAS

The software and setup side of a NAS, from the operating system and pools to the network, backup and the faults owners find too late.

10 parts

Benchmarking

How to test an SSD so the number holds up: a repeatable client method in 101, then enterprise drives at steady state in 201.

Memory and chipmaking

Working Memory, Part 1: a six-transistor SRAM cell, three cache levels, and a die stacked on a dieA three-panel schematic. Left: a six-transistor SRAM cell between two bitlines, with a wordline above it. Center: three horizontal bars of growing width for the cache levels, the shortest in copper. Right: a logic die with a smaller cache die stacked on top of it and a row of bonds between them.GUIDE / WORKING MEMORY, PART 1The memory inside the processorCELL · LEVELS · STACKED DIE01 CELLSix transistors02 LEVELSLarger and slower03 STACKA second die on topL1L2L3SRAM AND THE PROCESSOR CACHE

7 parts

Working Memory

The memory a processor works from: SRAM and the cache, DRAM from the inside, what could follow DRAM, and the HBM stack.

Beyond NAND, Part 1: a NOR cell with a bitline contact, a clocked flash access with wait states, and grade bars under a reflow peakA three-panel schematic. Left: a transistor with a floating gate in copper and a contact rising from the drain to a bitline. Center: a square clock wave above a row of boxes, the first four dashed as wait states and the last one solid in copper. Right: three bars of growing width for the temperature grades and a copper curve that ramps up to a peak above them.GUIDE / BEYOND NAND, PART 1Code memory that runs in placeNOR · EXECUTE IN PLACE · THE DIE01 CELLA contact per cell02 ACCESSWaits, then data03 LADDERReflow tops gradesNOR AND MEMORY ON THE PROCESSOR DIE

5 parts

Beyond NAND

The non-volatile memories beside flash: NOR and on-die memory, phase-change memory and Optane, MRAM, ferroelectric and resistive memory.

Built for SI

Tokens going through trained weights to the next tokenThree word tokens on the left feed into a large grid of shaded cells, the model's trained weights, with one row lit up as the token selects it. An arrow on the right leads to the next token the model writes.GUIDE / UNDERSTANDING SI, PART 1Tokens in, the next token outTOKENS · TRAINED WEIGHTS · NEXT TOKENTOKENSTRAINED WEIGHTSNEXT

3 parts

Understanding SI

SI itself: what a language model knows, running one on Windows, and the agents built on one.

Memory and Storage for SI, Part 1: a weight as two conductances, a staircase of update steps and a spread around a targetA three-panel schematic. Left: two vertical bars of different heights with a copper bracket between their tops, a weight stored as the difference of two conductances. Center: a conductance staircase that rises in equal steps and falls in equal steps, with the pulses drawn as ticks beneath. Right: a bell-shaped spread of conductances around a dashed target line.GUIDE / MEMORY AND STORAGE FOR SI, PART 1A weight, a step, a spreadWEIGHT · UPDATE · SPREAD01 WEIGHTG+G−Two cells,one weight02 UPDATESmall steps,up and down03 SPREADtargetMisses land arounda target

8 parts

Memory and Storage for SI

Memory and storage built for SI, from a memory cell used as a weight to high bandwidth flash and the drives made for inference.

NVMe over Fabrics: a local SSD on a short PCIe link, the same SSD reached across a switch with capsules on the wire, and many senders converging on one receiverA three-panel schematic. Left: a host chip sits directly above an SSD, joined by one short copper link. Center: a host on the left and a storage box with SSDs on the right sit on either side of a network switch, and small teal capsules travel along the copper link between them. Right: four sender boxes on the left send copper lines into one switch, which feeds a single receiver box on the right.GUIDE / NETWORKING FOR SI, PART 1NVMe across the networkLOCAL · FABRIC · INCAST01 LOCAL02 FABRIC03 INCASTHOSTSSDHOSTcapsulesSSDsswitchmany to one

5 parts

Networking for SI

The networks that carry storage traffic for SI clusters, starting with NVMe over Fabrics.

Newest guides

Explore

Run it, turn it, trace it

Simulators and drive checks that run on your device, sourced timelines, and software to install. No uploads, no accounts.

All Labs

Runs in your browserWalkthrough

Follow a Read

Follow one 4KiB read through a Samsung 990 Pro 2TB, from the host command to the NAND cell and back, with measured and published times for each stop.

Try this: Step through one 4KiB read in ten frames, then switch DRAM or DRAM-less and fresh or worn pages, and read the time on the bench.

Open Follow a Read

Runs in your browserDecoder

NVMe SMART Decoder

A browser tool that opens on a made-up example drive already decoded, then reads NVMe SMART / Health logs and Identify Controller data from nvme-cli, smartctl or raw hex, with endurance numbers.

Try this: Read the decoded example drive, then change the TBW rating and capacity to see remaining life.

Open NVMe SMART Decoder

Sourced timelines

Every point is dated by its earliest public source and links to it.

All Timelines
Word-line layers, log scale11010020052010201520202025

Sourced chart

NAND Die Timeline

Try this: Switch between layers, bit density, interface speed and cell type, then click a point to see the paper behind it.

Open
dies
97
sources
150
first source
2002
Fastest NAND speed, MT/s02k4k201520202025

Sourced chart

Controller Timeline

Try this: Filter by host bus, sort the table, then click a point to jump to the controller brief behind it.

Open
controllers
72
sources
87
first source
2008
Drive capacity, TB, log scale0.111020052010201520202025

Sourced chart

HDD Timeline

Try this: Follow hard drive capacity by year, then filter by recording method to see where shingled, microwave and heat-assisted drives took over.

Open
drives
29
sources
25
first source
2004
Die density, Gb, log scale0.1110200020102020

Sourced chart

DRAM Die Timeline

Try this: Switch to data rate, watch the JEDEC line step up, then click a die to see the release or datasheet behind it.

Open
dies
88
sources
87
first source
1998
Stack bandwidth, GB/s02k4k201520202025

Sourced chart

HBM Timeline

Try this: Compare stack bandwidth by vendor, then see which accelerators used each generation.

Open
records
44
sources
53
first source
2013

Bit Stream

What the memory makers earn and build

43 memory, flash, controller and drive companies, every quarter: revenue, profit, forecasts and the fabs behind them, read for the drive you buy.

Open Bit Market

Bit Market · Latest quarter

Q3 2026 so far, against the same quarter a year earlier

Quarters are calendar quarters. Each company’s fiscal quarter goes under the calendar quarter in which it ends, except that a quarter ending in the first 7 days of a month counts as the month before. Its own label is under each figure. Won, yen, New Taiwan dollar and yuan figures are converted to US dollars at the Federal Reserve H.10 average for the quarter; the filed figure is under each. Samsung’s tile here is its chip division (Device Solutions); the tables below show the whole company.

  • Samsung chip division revenue$103.8Bup 344.4% against a year earlier, in the filed currencyfiled ₩147.1Ta year earlier $23.9B (filed ₩33.1T)3Q 2026
  • SK hynix operating margin80%up 33.4 percentage points against a year earliera year earlier 46.6%3Q 2026
  • Micron revenue$54.229Bup 379.3% against a year earliera year earlier $11.315BFQ4 2026
  • Sandisk operating margin79.6%up 72.0 percentage points against a year earliera year earlier 7.6%FQ1 2027
  • Nvidia revenue$96.221Bup 105.9% against a year earliera year earlier $46.743BFQ2 FY2027

Q3 2026 so far: 42 of 43 companies have reported. Still to report: YMTC.

News · Week ending November 13, 2026 · 15 items

Flash as Memory, Dense Object Stores and CXL Demos

  • Sandisk introduces embedded flash for devices that still use e.MMC
  • SK hynix reports safety and cybersecurity certifications for physical AI memory
  • Intel describes hardware memory tiering for AI agent sandboxes

Writing

Notes from the bench

Reviews, validation methods, engineering notes and reflections. My published reviews elsewhere are listed under Technical Analysis. A few notes cover Cell, the SSD guide chatbot in the corner of the page.

All Notes

Newest

Blog

Recent notes

Reflections

Why Eyes on Flash Now Writes SI

Eyes on Flash now writes SI, not AI, after Executive Order 14434: what the order says, what it leaves alone, and how the site applies it.

Engineering

How I Made SSD Validation Repeatable

How I turned SSD validation into a repeatable process: fixed phases, write saturation with idle recovery, power capture, and a report each run had to fill in.

Research desk

Webster's Corner

Specs, standards, papers and maker news on memory and storage. Newest first. Every entry links the source.

Open the desk
Sean Webster looking across the mountains from a gondola.

The person behind the work

Curiosity goes
beyond the lab

Memory and storage technologist, writer, photographer, and builder. The same instinct runs through it all: look closer, ask why, and understand how things work.