The Physics of NAND Flash — Remembering by Trapping Electrons

Why does a memory stick keep your photos with the power off? Starting from one idea — trap electrons on an island of insulator and shift a threshold voltage — this article builds up tunnelling writes, multi-level cells, 3D stacking and wear-out, and ends where you can read an SSD datasheet and a SMART log for yourself.

Why doesn't it forget when the power goes off

Pull a USB stick out of your laptop, leave it in a drawer for six months, and the photos are still there. That feels ordinary, but inside a computer, parts that hold their state without power are the exception. DRAM, your main memory, loses its contents unless every cell is read back and rewritten every few tens of milliseconds. The registers and SRAM inside a CPU go blank the instant the supply drops. All of them hold state by having voltage applied to them.

So what is flash memory doing differently? The answer is almost disappointingly plain: it traps electrons in a room with no exit. Trapped electrons have nowhere to go when the power stops, so they stay. Memory here is not an electrically maintained state but a physical charge that simply sits there — and that difference is where everything else in this article comes from.

The analogy: an attic with no door, right above the path

Think of a transistor as a sluice gate. Put voltage on the gate and a channel opens so current flows; take it away and the channel closes. That is a MOSFET, and a chip is paved with them. A flash cell adds exactly one part to it: between the gate and the channel, you insert a small island that is wired to nothing at all. It is surrounded by insulator on every side and connected, electrically, to nowhere. That is the floating gate — an island suspended in mid-air, as the name says.

Push electrons onto that island and their negative charge acts as a weight sitting on top of the channel below. You can still open the sluice gate, but you have to push harder to do it. So what a flash cell remembers is not the presence or absence of electrons as such, but the voltage at which this particular switch now begins to open. Reading is nothing more than applying a fixed voltage and asking: did it open, or didn't it?

The mechanism: rewriting the threshold voltage

The voltage at which a switch starts to open is called the threshold voltage, $V_{th}$. Call the charge stored on the floating gate $Q_{FG}$, and the capacitance between the floating gate and the control gate above it $C$. The shift in threshold is then:

Put in words: the more electrons you store, the more voltage it takes to open the switch. $Q_{FG}$ is negative because electrons are negative, and it cancels the minus sign, so $\Delta V_{th}$ comes out positive. $C$ sits in the denominator because the same charge produces a smaller voltage change on a larger capacitance — this is just $V = Q/C$ for a capacitor.

The part worth holding on to is that the expression is continuous. This is not a binary choice between electrons and no electrons; by choosing how many you inject, you can park the threshold anywhere you like. Everything in the multi-level-cell section below falls out of that single line.

Writing means tunnelling

To an electron, the insulating wall is an energy barrier of roughly 3 electron-volts — more than a hundred times the thermal energy available at room temperature (about 0.026 eV). It is not a wall heat can carry you over. That is exactly why the data survives for years, but it also means there is no ordinary way to write in the first place.

The way in is a quantum-mechanical one: tunnelling. Apply a strong enough electric field and the electron does not climb the barrier, it passes straight through it. NAND uses Fowler–Nordheim tunnelling, where the current density $J$ depends on the field $E$ roughly like this:

Or, in words: raise the field a little and the amount getting through jumps by orders of magnitude ($A$ and $B$ are constants set by the material and the barrier height). While $E$ is small the current is effectively zero; past a certain point it switches on hard. That violent nonlinearity is flash memory. It refuses to leak for decades, yet conducts for the instant you put nearly 20 volts across it — non-volatility and rewritability, two properties that ought to be in conflict, living together on one line of algebra.

Why erase happens a whole block at a time

Programming puts a high voltage on the control gate and pulls electrons onto the island. But cells vary, so a single pulse never lands on the target. Real devices climb a staircase instead: apply a pulse, read back to check, and if it is not there yet, step the voltage up slightly and pulse again. The flip side is that the tighter you want the final distribution, the more steps you need, and the slower the write becomes.

Erasing runs the other way: a high voltage on the substrate pulls the electrons back out. The catch is that this substrate — the well — is shared by an enormous number of cells, which means erase can only be done a block at a time, thousands of pages at once. Read by the page, program by the page, erase by the block. Most of what makes flash awkward to use traces back to that asymmetry.