Why Do SSDs Wear Out When RAM Practically Never Does?

Table of Contents (click to expand)
An SSD stores each bit by forcing electrons through a layer of glass-like insulation at 15 to 20 volts, and every trip through the glass leaves a little permanent damage behind, so a flash memory cell of the mid-2010s was rated for only about 1,000 to 3,000 rewrites. RAM never forces charge through its insulation: it fills and drains a tiny electric bucket through a tap at low voltage, which is why it can be topped up about half a billion times a year without wearing, and also why it forgets everything the moment the power stops. RAM does still age (Google’s servers showed rising error rates after ten to eighteen months), but not from being written to.

Sixty-four milliseconds. That is how long a cell of RAM is trusted to hold one bit before the computer must top it up. Do the sum and it turns absurd: about fifteen top-ups a second, roughly half a billion a year, for each of the billions of cells on the module. All of that, just to remember what it already knew.

Now walk over to the SSD in the same machine. A flash cell of the mid-2010s was rated for around 1,000 rewrites. Not per year. In its whole life.

So the chip topped up half a billion times a year ships with no wear rating at all, and the chip that tolerates a thousand comes with a warranty that counts every terabyte you write. Your intuition says that is backwards. It has the wrong picture of what a “write” is. The answer lies in what each chip does to an electron to make it stay put, and in a layer of glass a few millionths of a millimeter thick.

One of these cells is topped up about 493 million times a year. The other is rated for about a thousand rewrites, ever. The log scale is doing a lot of work here.
One of these cells is topped up about 493 million times a year. The other is rated for about a thousand rewrites, ever. The log scale is doing a lot of work here.

How Does A Computer Chip Store A 1 Or A 0?

Press a key and somewhere in your machine a bit flips. A bit is the smallest unit of information, a one or a zero. Inside a chip, it is not a number written down anywhere. It is a small pocket of electric charge, or the lack of one.

Everything else follows from one question: where do you keep the charge, and how do you get it in and out?

The first is a bucket with a tap. You put a tiny capacitor, a part that holds charge, behind a transistor, an electronic switch that works like a tap. Open the tap, let charge pour in or drain out, close the tap. That is RAM.

The second is an island sealed inside glass. You park the electrons on a scrap of conductor wrapped on every side in an insulator, with no tap at all. That is flash, the memory inside every SSD, phone and USB stick.

Both chips hold a bit as charge. The difference is the tap: RAM has one, and flash sealed it over. The forgetting, the wear and the speed gap all come from that single choice.

Two ways to park a bit. RAM keeps its charge in a leaky bucket behind a tap. Flash seals it on an island inside glass, and has to force electrons through the glass to get them there.
Two ways to park a bit. RAM keeps its charge in a leaky bucket behind a tap. Flash seals it on an island inside glass, and has to force electrons through the glass to get them there.

How Does RAM Hold A Bit, And Why Does It Forget?

Each cell of the RAM on your memory sticks is one transistor and one capacitor, and the charge sitting on that capacitor is the stored bit.

To write a one, the chip opens the tap and the wire behind it pours charge into the bucket. To write a zero, it opens the tap and drains the bucket. Either way the charge moves through the transistor at the chip’s normal low voltage, the same path any ordinary current takes. Nothing crosses an insulator, so nothing gets stressed.

The price is that the bucket leaks. Left alone, a RAM cell holds its charge for milliseconds to seconds. To cope, the computer tops up every cell in place, a chore engineers call a refresh. The rule is that no cell goes more than 64 milliseconds without one, and that is where the half a billion top-ups a year come from.

Cut the power and the top-ups stop. The buckets drain in under a second and your unsaved document is gone. This is why RAM forgets when the power goes off, and why a computer needs a permanent store as well. A bucket with a hole in it is a strange way to build a memory. It is also why the memory never wears out.

A DDR4 memory module. Under the heatsink, every cell is a tap and a bucket that gets topped up more than fifteen times a second. (Photo Credit: ElooKoN, Wikimedia Commons, CC BY-SA 4.0)
A DDR4 memory module. Under the heatsink, every cell is a tap and a bucket that gets topped up more than fifteen times a second. (Photo Credit: ElooKoN, Wikimedia Commons, CC BY-SA 4.0)

How Does Flash Memory Keep Data With The Power Off?

A flash cell is a transistor with one extra piece: the island, which engineers call a floating gate. It is sealed above and below by layers of silicon dioxide, the same compound as window glass. The layer underneath is about 7 nanometers thick, and a nanometer is a millionth of a millimeter.

Sealed like that, the electrons stay put with the power off. Where RAM leaks in milliseconds, flash holds its charge for several days to years. The consumer standard asks an SSD to keep its data for one year unpowered at 30 °C (86 °F).

But how do you get electrons onto an island with no tap? You force them through the glass. It takes 15 to 20 volts on the gate above the island. At that voltage, electrons cross the glass by quantum tunneling, a trick in which a particle turns up on the far side of a barrier it lacks the energy to climb.

Erasing pulls the electrons back out the same way, and it only works on a whole block of cells at once. Flash cannot change one byte in place; it erases a block, then writes it again. One erase plus one write is a rewrite, which spec sheets call a program/erase cycle, and it is the unit flash wear is counted in. The same glass is why a flash write is so much slower than a RAM write: RAM opens a tap at low voltage, while flash forces electrons through a glass wall on every single write.

Cross-section of a flash cell. The island (the floating gate) sits between the control gate and the silicon, with insulating glass (not drawn) sealing it above and below. (Diagram: Cyferz, Wikimedia Commons, CC BY 2.5)
Cross-section of a flash cell. The island (the floating gate) sits between the control gate and the silicon, with insulating glass (not drawn) sealing it above and below. (Diagram: Cyferz, Wikimedia Commons, CC BY 2.5)

What Wears Out Inside An SSD?

Every trip through the glass costs something. With repeated rewrites, electrons become stuck in the glass, and the high voltage leaves tiny flaws in it. Stuck charge changes how much charge the cell appears to hold, so the chip’s reader mistakes one level for another, and the flaws give the stored electrons a path to seep out. The damage adds up over the life of a block.

The SSD fights back with error correction: extra check bits stored beside the data, used on every read to repair a limited number of wrong bits. So a worn-out cell is not a dead cell. It is a cell that can no longer be trusted to hold your file safely for the year the maker promises.

Most of today’s flash stores its electrons inside an insulator rather than on a conducting island, which suffers less from rewriting. It still writes by forcing electrons through glass, so less is not the same as none.

Left: a fresh cell. Right: the same cell after thousands of rewrites, with electrons stuck in the glass and leak paths opening through it.
Left: a fresh cell. Right: the same cell after thousands of rewrites, with electrons stuck in the glass and leak paths opening through it.

Why Do Cheaper, Bigger SSDs Wear Out Faster?

The short answer is bits per cell. The first flash stored one bit per cell by splitting the cell’s charge range into two windows. Later cells split the same range into four windows for two bits, eight for three bits and sixteen for four bits. Each step packs more data into the same silicon, which is what makes a drive cheap, and each step makes the bits harder to store and read without error. Spec sheets label the four designs SLC, MLC, TLC and QLC.

The margin for damage shrinks with the windows. In a one-bit cell, a lot of stuck charge must build up before a one starts to look like a zero. A four-bit cell has to tell sixteen shades of “some electrons” apart.

The ratings tell the story. Older one-bit cells were rated for around 150,000 rewrites. By the mid-2010s, with cells also shrinking, two-bit cells were down to about 3,000 and three-bit cells to about 1,000. Today’s drives are built differently, but the trade-off has not gone away.

Bigger cuts the other way. More capacity means more cells to spread the writes across. Take one current Samsung drive: the 1 TB model is rated at 600 terabytes written and the 4 TB model at 2,400, under the same five-year warranty. The bigger drive gets four times the write budget with no tougher cells. It simply has four times as many of them.

The same charge range split into two, four and eight windows. The more windows, the less stuck charge it takes to tip a cell across a line.
The same charge range split into two, four and eight windows. The more windows, the less stuck charge it takes to tip a cell across a line.

How Long Does An SSD Actually Last?

An SSD survives its own chemistry with two tricks. The first is wear leveling: the controller steers each new write toward the least-used blocks, so the whole drive wears at an even rate. The second is spare room. Makers give the controller more physical blocks than the advertised capacity, and the spares rotate in as blocks tire.

The write budget lets you do a rough sum. Take that 1 TB drive rated for 600 terabytes written and a laptop writing an illustrative 40 gigabytes a day. Divide one by the other and the budget lasts about 15,000 days, or 41 years. For that user, the five-year warranty runs out decades before the write budget does.

Nobody has watched drives for twenty years. Google watched its own for six, across many millions of drive-days. The calming part: raw error rates grew much more slowly with wear than expected. The catch: even after allowing for rewrites, error rates rose with the months a drive had spent in service. Drives age on the calendar, not just on the write counter. And since the retention rule is one year unpowered, an SSD in a drawer is not an archive.

Inside a 512 GB Samsung 860 Pro: the large black chip on the left is the flash, and the small chips on the right include the controller that spreads writes across it. (Photo Credit: Michael Bemmerl, Wikimedia Commons, CC BY 3.0 DE)
Inside a 512 GB Samsung 860 Pro: the large black chip on the left is the flash, and the small chips on the right include the controller that spreads writes across it. (Photo Credit: Michael Bemmerl, Wikimedia Commons, CC BY 3.0 DE)

So, Why Do SSDs Wear Out When RAM Doesn’t?

The “practically never” in the title is doing some work: RAM does not wear from being written, but it does age. Google watched its memory modules too, across millions of module-days, and found that more than 8 percent of them saw at least one correctable error a year. Most of those were not one-off glitches but permanent physical faults, and the aging set in after only ten to eighteen months in service.

None of that comes from the half a billion top-ups. A top-up is a low-voltage pour through an open tap, and it crosses no glass. The aging is the slow, ordinary decay every chip suffers.

So the answer is the tap. RAM keeps its charge behind one, so a write is a low-voltage pour through an open channel, and a cell can be topped up half a billion times a year for as long as its other parts hold out.

Flash sealed the tap over. That seal is why an SSD remembers for a year unpowered. It is also why every write forces electrons through 7 nanometers of glass at 15 to 20 volts, and each one leaves a few electrons stuck in the glass and a few new flaws. Pack three or four bits into a cell and the margins get so tight that the damage shows sooner.

Neither chip is immortal. Flash drives age on the calendar as well as on the write counter, and RAM develops faults of its own after a year or two. But only one of them is damaged by the act of remembering. For most people the arithmetic is kind anyway. Keep a backup, for the failures that have nothing to do with wear.

References (click to expand)
  1. Errors in Flash-Memory-Based Solid-State Drives: Analysis, Mitigation, and Recovery — Cai, Ghose, Haratsch, Luo, Mutlu (arXiv:1711.11427)
  2. DRAM Errors in the Wild: A Large-Scale Field Study — Schroeder, Pinheiro, Weber, SIGMETRICS 2009 (ACM, DOI 10.1145/2492101.1555372)
  3. Flash Reliability in Production: The Expected and the Unexpected — Schroeder, Lagisetty, Merchant, USENIX FAST 2016
  4. Tunnel Barrier Engineering for Flash Memory Technology — S. Verma, PhD dissertation, Stanford University
  5. JEDEC SSD Specifications Explained (JESD218 endurance classes) — Alvin Cox, JEDEC JC-64.8
  6. Samsung V-NAND SSD 990 PRO Data Sheet, Rev. 2.0 — Samsung Semiconductor
  7. The Memory Hierarchy, 6.004 Computation Structures — MIT OpenCourseWare