What Is MEMS? How Accelerometers and Gyroscopes Work

Where you see it

Turn your phone sideways, and the screen turns sideways too. When Apple announced the first iPhone in 2007, it said an accelerometer, a sensor that feels tilt, makes this switch. In 2010, the iPhone 4 added a gyroscope, which measures how fast the phone rotates. Both are MEMS with tiny moving parts inside.

MEMS are also widely used in the microphones for calls and voice input. In cars, they have been used in sensors that feel the sudden slowdown of a crash. The sensor then sends the signal to open the airbag. You can also find MEMS in inkjet printer heads and in movie theater projectors. Later sections look at each of these in turn.

What MEMS are: tiny machines made the way chips are made

MEMS stands for Micro Electro Mechanical Systems. Put simply, it means "very small electrical and mechanical systems." The name covers parts that combine electric circuits with mechanical pieces, such as bending springs or vibrating films (MEMS).

What sets MEMS apart is that they are made with the same tools as chips. Films are stacked on a round, thin slice of silicon called a wafer. A pattern is copied onto it with light, and the unwanted parts are removed ("How Are Chips Made?"). An ordinary chip uses these steps to build wiring and transistors. In MEMS, what is left after the removal is a shape that can move.

Silicon is a great material for electronic parts. It is also an excellent material for springs. A tiny silicon spring can be bent and will spring back exactly to its original shape, as long as it does not break. In 1982, Kurt Petersen in the US made this idea widely known in his paper "Silicon as a Mechanical Material."

Here is an example of the size. The ADXL105 is a motion sensor from the US company Analog Devices. Its sensing part is only about 0.5 mm × 0.4 mm. It is just 2 µm thick, or two thousandths of a millimeter (from the company's technical journal).

How it works 1: an accelerometer is a weight on springs

How quickly speed changes is called acceleration. You feel it when a stopped train starts to move, or when a car brakes to a stop. A part that measures acceleration is an accelerometer.

Inside is a tiny weight, called the mass, held by springs. When the outside of the part suddenly moves, the mass tends to stay where it is, so it lags a little. As a result, the mass shifts slightly relative to the outer frame. The larger the acceleration, the larger the shift.

An accelerometer also feels gravity. Even in a phone lying on a desk, gravity pulls the mass down and shifts it slightly. The direction of the shift tells which way the phone is tilted. This is how the screen knows when to switch between portrait and landscape.

This kind of sensor has long been used for car airbags too. In 1991, Analog Devices announced the ADXL50, an accelerometer for airbags. It was in volume production by 1993 (according to an article by the parts distributor Digi-Key). It put the sensor and its readout circuit on one chip. It could also check itself for faults. Before that, mechanical sensors with a ball moving inside a tube were the norm.

How it works 2: reading the shift with electricity

The mass shifts so little that you cannot see it. So a fixed plate is placed right next to the mass. This plate is called an electrode. It conducts electricity and faces the mass across a small gap.

Two plates facing each other across a gap have the same shape as a capacitor, a part that stores electric charge. The narrower the gap, the more charge the plates can store ("What Is a Capacitor?"). When the mass shifts and the gap changes, the amount of charge they can store changes too. A circuit on the chip measures this change.

In practice, electrodes sit on both sides of the mass. When the gap on one side narrows, the gap on the other side widens. Taking the difference between the two cancels effects that change both sides the same way, such as temperature. Reading motion from changes in how easily charge is stored is called capacitive sensing.

The change is very small. An acceleration as strong as Earth's gravity is written as 1 g. In the ADXL105, even 1 g changes the stored charge by only about 1/1500. To read such a small change, the measuring circuit is built right next to the sensor.

Another method reads the strain at the base of a spring instead of the gap. It measures this as a change in how hard it is for current to flow, called resistance (piezoresistive effect). The table in "Capacitive vs Piezoresistive MEMS Sensors" sums up how the two methods differ.

How it works 3: a gyroscope feels a sideways push from rotation

A gyroscope is a sensor that measures how fast something rotates. An accelerometer uses gravity to sense tilt, but some motions are hard for it. Lay a phone flat on a desk and spin it. The direction of gravity does not change. So the accelerometer shows almost no change.

Inside a MEMS gyroscope, a mass is kept moving back and forth in one direction, very fast and over a tiny distance. If the whole part then rotates, the mass is pushed at right angles to its back-and-forth motion. This apparent force acts on things moving on a rotating object. It is called the Coriolis force.

The faster the rotation, the stronger the push. The sideways shift of the mass is read as a change in a gap, just as in an accelerometer (from a paper on MEMS gyroscopes). The key is to use separate electrodes: one set keeps the mass moving back and forth, and another set reads the sideways shift.

Making it 1: dissolve the layer underneath

A moving part needs a gap underneath it. But chipmaking builds things by stacking films from the bottom up. You cannot simply build a floating shape on top of nothing.

So a layer that will be removed later is laid down first. This layer is called the sacrificial layer. A structural film is stacked on top. At the end, only the sacrificial layer is dissolved with a chemical. The film above is then left floating as a moving strip (a beam) or a thin film. This method is called sacrificial layer etching.

The sacrificial layer is often silicon dioxide, the main ingredient of glass. It is dissolved with a chemical called hydrofluoric acid. The structural film is often silicon made of many tiny crystals, called polysilicon (from lecture slides at the University of California, Berkeley). Hydrofluoric acid dissolves silicon dioxide well but barely touches silicon. This difference leaves only the shape you need.

It is a little like pouring a concrete floor. Builders put up temporary supports underneath and take them out once the concrete has set. A sacrificial layer, though, is not taken out but dissolved and washed away. The gap it leaves is usually a few µm (a few thousandths of a millimeter) or less.

Making it 2: digging narrow, deep trenches with the Bosch process

In MEMS, there is often a need to dig deep, straight down into silicon. This makes the mass heavier, or gives the electrodes more facing area. But when chemicals or gases remove material, they tend to eat sideways as well as down.

Here, the silicon is etched with plasma, a gas given energy by electricity ("Plasma in Semiconductor Manufacturing"). In the early 1990s, the German company Bosch developed a way to dig deep with plasma (from a document by a German research institute). It is named after the company: the Bosch process.

There are three steps. ① Dig a little. ② Coat the whole inside of the hole with a thin protective film. ③ Fire charged particles (ions) straight down to break through the film only at the bottom. Once the bottom is open, go back to step 1 and dig a little more. The walls stay protected by the film, so the hole barely widens.

Repeating these steps digs holes and trenches far deeper than they are wide. But since the digging happens a little at a time, the walls are left with fine ripples. The first commercial tool was shipped in 1995 by a British equipment maker, today's SPTS (from the company's documents). Today, the process is also used to drill holes that pass straight through a chip.

Why making it small helps

First, it is cheap to make. Thousands or more of the same part (depending on its size) can be laid out on one wafer and made all at once. Compared with mechanical parts assembled one by one, the price of each part drops sharply.

Second, it uses little power. The moving part is very light, so a little electricity is enough to move it or measure it. This low power use is one reason a phone can count your steps all day long.

Third, it can be mass-produced with even quality. According to the market research firm Yole Group, about 31 billion MEMS were shipped worldwide in 2024. Revenue was about 15.4 billion dollars. Making so many parts of the same quality is possible because they are made the way chips are made.

The hard parts: sticking and packaging

At very small sizes, the forces that pull surfaces together matter more than weight. The riskiest moment comes after the sacrificial layer is dissolved, when the rinse water is dried. The surface of water left in the gap tries to shrink. That pull draws the freed beam down to the base, where it sticks. This is called stiction.

One fix is to swap the water for alcohol, then for liquid carbon dioxide, before drying. Raising the pressure and temperature creates a state with no boundary between liquid and gas. So there is no surface to shrink and pull (from the Berkeley lecture slides). Another trick is to coat the surfaces with a water-repelling film to prevent sticking during use.

Another challenge is the protective case, or package. An ordinary chip is protected by sealing it in plastic resin ("What Is Semiconductor Packaging?"). But sealing MEMS in solid resin would stop them from moving. Instead, a tiny cap is placed over the moving part, leaving a space inside.

Gyroscopes are often sealed with the inside close to a vacuum (from a paper on MEMS sealing). With air inside, the back-and-forth vibration would quickly die out. Microphones are the opposite: they need a hole to let sound in. Each part needs its own way of being protected and connected to the outside.

Other everyday MEMS

MEMS microphones

A thin film and a stiff plate full of holes face each other across a gap. Sound changes the air pressure, which makes the film vibrate and changes the gap. The most common method reads this as a change in a capacitor, just like an accelerometer (from a 2024 review paper on MEMS microphones). The US company Knowles brought MEMS microphones to market in the early 2000s. In a 2006 paper, its engineers called theirs "the first commercialized MEMS microphone."

Inkjet printer heads

Here we look at the type that uses heat. A tiny heater deep inside the nozzle heats the ink for an instant. A bubble forms and pushes out a drop of ink. HP released ThinkJet, a printer using this method, in 1984 (from HP's records). A 1997 report by the US National Research Council called these heads the largest commercial success of MEMS, both in number and in value.

Tiny mirrors in projectors

In 1987, Larry Hornbeck of the US company Texas Instruments (TI) made a part with rows of tiny mirrors, called a DMD. The American Society of Mechanical Engineers (ASME) describes it as a part with up to about 2 million mirrors. In the basic design, each mirror handles one point of the image. By changing its tilt, it either sends light into the lens or turns it away.

In an automotive product, for example, the mirror tilts to one of two angles: +12° or −12° (from TI's data sheet). Brightness is set by how long each mirror sends light into the lens. This way of projecting is called DLP, and it is also used in movie theater projectors (from the Lemelson-MIT Program in the US).

Common myths

By the numbers

The "1 g" in the table is a unit of acceleration. It has nothing to do with 1 g (gram) of weight. The mass weighs about 0.5 µg, or one two-millionth of a gram. Yet the sensor reads the shift of such a light mass with electricity.

Learn more

• Related articles: "What Chips Are Inside a Smartphone?" (the other chips besides MEMS), "How Do Touchscreens Work?" (which also reads changes in stored charge), "What Is a Capacitor?", "Plasma in Semiconductor Manufacturing"
• See the differences in a table: Capacitive vs piezoresistive MEMS sensors / Isotropic vs anisotropic etching / KOH vs TMAH (ways to etch silicon with liquid chemicals; intermediate level)
• Terms for the next step: anisotropic etching, anodic bonding, SOI wafer, Hall-effect sensor