There is a machine with moving parts inside your phone, and the parts move by about the width of an atom

Rotate your phone and the screen turns. The usual assumption is that this is software noticing something, or some solid-state trick with no moving parts. It is neither. Inside the phone there is an actual machine — a block of silicon suspended on silicon springs, free to swing — and when you tilt the handset, gravity pulls that block a little to one side. The distance it moves is on the order of a nanometre, roughly the width of a few atoms. The chip measures that, and the screen rotates.

These devices are called MEMS, micro-electro-mechanical systems, and they are one of the strangest products of the semiconductor industry: mechanical engineering practised at a scale where you cannot assemble anything, only grow and dissolve it.

Weighing a movement too small to see

An accelerometer contains a proof mass — a deliberately made lump of silicon — held by thin silicon beams that act as springs. Accelerate the chip and the mass, having inertia, lags behind; the springs stretch; the mass sits displaced by an amount proportional to the acceleration. Tilt it instead, and gravity does the same job, which is why the same sensor reports both motion and orientation and cannot by itself tell you which it is seeing.

The question is how you measure a displacement of a nanometre on a device costing a few rupees. You do not measure it optically or mechanically. You measure it electrically.

The moving mass carries a comb of fingers that interleave with a second, fixed comb attached to the chip. Two interleaved conductors separated by a gap form a capacitor, and a capacitor's value depends on that gap. When the mass shifts, one set of gaps narrows and the other widens, so one capacitance rises as the other falls. The circuit measures the difference between them, which cancels most temperature and manufacturing drift, and which converts a nanometre of motion into a change in capacitance far smaller than a picofarad. Detecting that reliably, next to a radio transmitter, is the part of the design that is genuinely hard — and it is why the mechanical structure and its amplifier are built on the same die, so the signal never has to travel.

Carving a moving part out of a solid wafer

Nothing here is assembled. There is no way to make a spring and attach it to a mass at this scale. Instead the moving structure is created in place, and then whatever was holding it up is dissolved away.

In surface micromachining, layers are deposited on a wafer in a chosen order: first a sacrificial layer, usually silicon dioxide, then a structural layer of polysilicon that is patterned by ordinary photolithography into the shape of the mass, the springs and the combs. At that point the structure is fully formed and completely stuck down, sitting on solid oxide. The final step is the release etch — a chemistry chosen to dissolve the oxide rapidly while barely touching the silicon, typically hydrofluoric acid. The oxide under the structure disappears, and what was a flat stack of films becomes a suspended machine with clearance beneath it.

For thicker, heavier structures the route is deep reactive ion etching, which cuts near-vertical trenches many tens of micrometres deep by alternating between an etching step and a step that deposits a protective film on the sidewalls, so the etch advances downward and not sideways. It leaves a characteristic scalloped wall, and it is what allows a proof mass with real mass, which is what a sensitive device needs.

Everything in a MEMS device is made by adding and removing films across a whole wafer at once. The design freedom is enormous in two dimensions and severely constrained in the third — and almost every clever MEMS structure is a way of getting three-dimensional behaviour out of a stack of flat layers.

The failure that defines the field

Release a micrometre-scale structure and you meet the problem that has shaped this entire industry: stiction.

At this scale, surface forces dominate over weight. If the release etch is done in liquid and the part is then dried, the receding liquid pulls the flexible structure down onto the substrate by capillary action, and once two smooth silicon surfaces touch with that little separating them, van der Waals forces and surface chemistry can hold them there permanently. The device is mechanically perfect and completely dead.

The countermeasures are a good illustration of design at the small scale. Release in vapour rather than liquid, so no meniscus ever forms. Add small bumps — dimples — beneath the moving structure so that contact, if it happens, is on tiny points rather than flat areas. Roughen surfaces deliberately to reduce true contact area. Apply a molecularly thin anti-stick coating that lowers the surface energy. This is precisely the regime where a designer's intuitions from ordinary mechanical engineering stop being useful: friction, gravity and inertia become minor characters, and adhesion becomes the plot.

Detecting rotation with a vibrating mass

A gyroscope has to sense rotation, and there is nothing to push against. The trick is the Coriolis effect.

The device keeps a mass vibrating steadily back and forth along one axis, driven electrostatically. If the chip is then rotated, the vibrating mass experiences a force perpendicular to both its motion and the rotation axis — the same effect that deflects winds on a spinning planet — and it begins to move slightly in a second, perpendicular direction. That secondary motion is proportional to the rate of rotation, and it is picked up by another set of capacitive combs. A gyroscope, in other words, is an accelerometer with a deliberately shaking mass inside it.

This explains a packaging detail that seems arbitrary until you know the mechanics. An accelerometer generally wants damping: gas trapped in the narrow gaps resists rapid motion and stops the structure ringing after a shock, so it is sealed at close to atmospheric pressure. A gyroscope wants the opposite — its drive vibration must be as lossy-free as possible to keep amplitude high with little power — so it is sealed under vacuum. Two nearly identical-looking silicon machines, hermetically bonded at the wafer level, with opposite atmospheres inside.

Why your phone still gets lost

MEMS sensors are cheap, tiny and remarkably good, and they all drift. A gyroscope's zero point wanders with temperature and with mechanical stress from the package — soldering a chip to a circuit board physically strains the die, and that strain shows up in the reading. Integrate a slightly wrong rotation rate over time and the estimated heading walks away from reality; integrate acceleration twice to get position and the error grows faster still. This is why purely inertial indoor navigation fails within a minute, and why cheap MEMS units cannot replace the expensive inertial systems in aircraft.

The answer is sensor fusion: combine sensors whose errors are unrelated. The gyroscope is excellent over short intervals and drifts slowly; the accelerometer is noisy but always knows where down is on average; the magnetometer knows roughly where north is but is thrown off by any nearby iron or magnet. An estimator blends them so that each one corrects the others' weaknesses. And that is what the figure-of-eight wave your phone occasionally asks for is doing — it is not calibrating the accelerometer but mapping how the phone's own metal distorts the magnetic field, so the compass can subtract it.

Why it matters for students and researchers

MEMS is where microfabrication stops being about electronics and becomes general-purpose manufacturing. The same lithography, deposition and etching that make transistors also make pressure sensors, microphones, inkjet nozzles, optical switches, microfluidic chips for diagnostics, and the resonators that keep time in modern devices. The processes were paid for by the chip industry; MEMS inherits them and applies them to anything that can be built out of patterned films.

It is also unusually good training, because a working device requires mechanics, electronics, materials and packaging to be right simultaneously, and the failure modes are dominated by surface effects rather than by the bulk properties taught in undergraduate courses. That combination — a structure that is only as good as its release step, a signal that is only as good as its amplifier, and a package that changes the answer — is what real device engineering looks like. For anyone studying nanotechnology, materials or electronics, it is the clearest available demonstration that at small scales, the surface is the device.

Frequently asked questions

How does a phone know which way it is being held?

A MEMS accelerometer inside contains a small silicon mass on flexible springs. Gravity displaces that mass by about a nanometre depending on orientation, and the chip detects the displacement as a tiny change in capacitance between interleaved comb electrodes.

How are MEMS devices manufactured?

They are built with the same processes as integrated circuits. Films are deposited and patterned by photolithography, with one layer intended to be sacrificial; a final etch dissolves that layer and leaves the remaining structure suspended and free to move.

What is stiction in MEMS?

It is the permanent sticking of a released microstructure to a nearby surface. At this scale capillary and van der Waals forces overwhelm the structure's weight, so devices are released in vapour and given dimples or anti-stick coatings to prevent it.

How does a MEMS gyroscope measure rotation?

It vibrates a mass along one axis. When the device rotates, the Coriolis effect pushes that vibrating mass sideways in proportion to the rotation rate, and capacitive electrodes detect the secondary motion.

Why do phone sensors need calibration and fusion?

Because each sensor drifts or is disturbed in a different way — gyroscope bias shifts with temperature and package stress, accelerometers are noisy, and magnetometers are distorted by nearby metal. Combining them lets each compensate for the others' errors.