What is MEMS? How do you build tiny moving machines on a chip?

MEMS stands for micro-electro-mechanical systems (also written microelectromechanical systems). It means using the methods that make chips to build tiny moving machines on silicon: small beams, membranes, springs and comb-shaped structures. When they get pushed, bent or shaken, a circuit next to them turns that motion into an electrical signal. The circuit can sit on the same chip, and it's also common to make it as a separate chip and package the two together. The accelerometer and gyroscope in your phone, and many microphones, are MEMS.

For what an ordinary chip is, start with "What is a chip?".

How is MEMS different from an ordinary chip?

Inside an ordinary chip, only electricity moves. A huge number of tiny switches keep turning on and off (What is a transistor?), but no part of it physically moves.

In MEMS, something does move. The moving structures are often only tens to hundreds of micrometers across; a human hair is a few tens of micrometers thick. What they do falls into two kinds:

• Sensing: turning motion, sound or pressure into an electrical signal. These are sensors.
• Acting: moving when they get an electrical signal, for example pushing out a drop of ink or tilting a tiny mirror. These are actuators.

Where do you find MEMS in everyday life?

• Your phone knows you turned it sideways: that's the accelerometer. Inside is a small piece of silicon hung on thin springs, with rows of fine teeth like a comb next to it. When the phone tilts, gravity pulls that piece a tiny bit to one side, and the gaps between the teeth change. Two conductors very close together can hold a little electric charge, and that's called capacitance; when the gap changes, the capacitance changes, and the circuit can measure it.
• Your phone knows you're turning it: that's the gyroscope. Inside is a structure that keeps vibrating in tiny movements; when the phone turns, it gets pushed a little to the side.
• Microphones: a very thin membrane faces a stiff plate full of small holes, with a very narrow gap between them. Sound pushes the membrane, the gap changes, the capacitance changes with it, and you get a voltage that rises and falls with the sound (for how it then becomes numbers, see "Analog vs digital chips: what's the difference?").
• Tire pressure and air pressure: a very thin silicon membrane with a sealed cavity behind it. When the outside pressure changes, the membrane bends, and resistors built on the membrane change too. Air pressure changes with height, so it can also be used to estimate altitude.
• Inkjet printer heads: a row of very small nozzles. In one kind, each nozzle has a tiny heater that heats the ink in an instant until a small bubble forms, and the bubble pushes a drop out; another kind pushes with a material that changes shape when a voltage is applied.
• Tiny mirrors in projectors: one kind of chip is covered with very small mirrors, and each one tilts to one side or the other on its own, deciding whether the light for that point reaches the picture.

How is MEMS made?

Most of the steps are the same as for an ordinary chip: films are grown layer by layer on a wafer (What is thin film deposition?), patterns are printed on with litho (What is photolithography?), and the unwanted parts are removed by etch (What is etching?).

The difference: in an ordinary chip, every layer sits tight against the one below. In MEMS, some parts have to hang in the air so they can move. That takes an extra step: the release.

How do you get a moving part to hang free? Step by step

Step 1: Lay down a "sacrificial layer"

First, grow a film on the wafer that will later be removed completely, often silicon dioxide, the main ingredient of glass; where the structure will be anchored, open a few holes in it. The structure on top has to end up hanging free, but while it's being built, something has to hold it up; those holes become its feet. Without this layer, the structure would be stuck flat on the surface below and couldn't move.

Step 2: Add the structural layer and etch its shape

Grow another layer on top of the sacrificial one, often polysilicon (silicon made of many tiny crystal grains), and use litho and etch to cut out the beams, springs and comb teeth. Large areas also get lots of small holes. The holes are paths for the chemical in the next step. Without them, the chemical could only eat inward slowly from the edges, and it would take a very long time to clear out the layer under a large structure.

Step 3: Etch away the sacrificial layer

Put the wafer in a chemical that eats the sacrificial layer and barely touches the structure. For example, hydrofluoric acid dissolves silicon dioxide but hardly touches silicon. The chemical has to work its way in sideways, under the structure: the "eating sideways" that's a problem when etching fine lines is exactly what you want here. If it doesn't clear everything, part of the structure is still attached underneath and can't move; with the wrong chemical, the structure gets eaten too.

Step 4: Dry it without letting it stick

After the chemical, the wafer is rinsed with water and dried. The gap under the structure is often just a few micrometers, so as the water slowly dries, its surface tension pulls the structure down; once it touches, it sticks. This is called stiction. A structure that sticks can never move again, and that device is lost. So fabs often use a drying method that doesn't let the water slowly dry away, or skip the liquid and do the release with a gas instead.

Why does MEMS often need very deep etching?

Ordinary chips usually only etch a thin film. MEMS often has to etch deep: in some accelerometers and gyroscopes, the moving silicon has to be heavy enough and the comb teeth tall enough, so they're cut straight through silicon tens of micrometers thick. In some pressure sensors, the membrane is made by etching from the back of the wafer to the front until only a thin sheet is left.

To etch deep and straight, the common method is to "etch a little, coat the sidewalls with a protective film, etch a little more," over and over. The protected sidewalls keep it going straight down, and the walls end up with fine ripples, ring after ring.

Why is MEMS packaging different?

An ordinary chip is often fully encased in plastic molding, chip and wires together (What is chip packaging?). MEMS can't be done that way:

• Parts that must reach the outside: microphones and air-pressure sensors need a small hole in the package to let sound or air in, while still keeping dust and moisture out.
• Parts that must be shut in: a gyroscope keeps vibrating, and air drags on that vibration like friction, so it's often sealed, sometimes with the inside pumped down close to a vacuum.
• Room for the moving part: it can't be filled with plastic, so a small cap is often put over it on the whole wafer, before the wafer is cut apart. Blade dicing sprays water as it cuts, and a free-hanging structure without a cap can be damaged by the water and debris.

Why is testing more than just powering it up?

Testing an ordinary chip means connecting it to a tester, sending electrical signals in and measuring how it responds (How are chips tested?). For MEMS, that's often not enough: whether a membrane is stuck or a spring is too stiff doesn't always show up from electrical signals alone.

So the test has to give it the outside world for real: an accelerometer goes on a table that shakes, a gyroscope on a table that spins, a microphone hears a sound of known loudness, and a pressure sensor goes into a box where the pressure can be set, to see whether the output is right. Every device comes out a little different, so they often need calibrating as well.

In fab24hr

There's no MEMS in the game. The main story, The life of a chip, follows a fictional hearing-aid chip, "Glowmere": Ms. Shen's small company wants hearing aids that use less power, so the same battery lasts several days longer. In the finale, chapter 13, "Into His Hands", Mr. Shen puts the battery in, and the hearing aid gives a short chime; later, through a wall, he hears his wife in the kitchen calling him to dinner.

The game doesn't say what kind of microphone this hearing aid uses. Public, general sources say hearing-aid microphones have traditionally been mostly another kind (electret condenser microphones), and that some use MEMS microphones.

The closest thing in the game to how MEMS is made is the "Batch wet bench" in the codex's Machines list. Open it and tap "See it work": the wafers soak in dilute HF, which dissolves the very thin silicon oxide on the surface; at the end, in the dryer, IPA (a kind of alcohol) vapor pulls the water film off. It cleans ordinary wafers, not MEMS. But a MEMS release often uses a similar chemical, only it dissolves a whole layer under the structure, and the drying that follows is exactly when MEMS is most at risk of stiction.

The companies, the chip and the people in the game are fictional; the parts about MEMS come from public, general sources. For other jargon, see Chip fab jargon in plain English.