Devices
The moving parts inside a phone are etched, not assembled
Springs, weights and combs a few microns across are carved out of one piece of silicon, and then asked to survive being dropped.

There is a settled way of talking about mEMS sensors. It is worth asking how much of it survives contact with the detail.
The argument in brief
- An accelerometer cannot distinguish gravity from real acceleration.
- A gyroscope outputs rate, so orientation drifts as errors accumulate.
- Sensor fusion pairs a stable slow reference with a fast drifting one.
Machines built by the methods used for chips
Micro-electromechanical sensors are carved out of silicon using much the same masking and etching steps that produce ordinary transistors. Instead of leaving a circuit behind, an etch removes material beneath a structure so that part of the silicon becomes free to move. The result is a beam or plate suspended on silicon springs, with dimensions measured in microns and mass measured in micrograms.
Nothing is assembled afterwards, because the springs, the moving mass and the sensing electrodes all emerge together from one piece of material. Since they are produced in batches across a wafer, thousands of identical mechanisms are made at once for very little money each.
An accelerometer measures force, not movement
The suspended mass lags behind whenever the chip accelerates, and that lag changes the spacing between interleaved comb-shaped electrodes. Changing the spacing changes capacitance, which the surrounding circuit converts into a voltage proportional to acceleration along that axis. Gravity is indistinguishable from acceleration, so a completely stationary sensor still reports a steady pull of one gravity towards the ground.
That is the signal behind screen rotation: the phone is not detecting a turn, it is detecting which direction down has moved to. A device in free fall reports almost nothing, which is how a laptop once knew to park its drive heads before hitting the floor.
A gyroscope needs something already vibrating
There is no spinning wheel anywhere inside; instead a mass is driven back and forth continuously at a fixed known frequency. When the device rotates, the Coriolis effect pushes that oscillating mass sideways, perpendicular to both its own motion and the rotation axis.
In the datasheet, a second set of electrodes measures the sideways displacement, which turns out to be proportional to the rate of turn. The output is therefore a rate rather than an angle, so orientation has to be obtained by adding those rates up over time. Any small bias in the rate accumulates without limit during that addition, which is why gyroscope-only orientation drifts within seconds.
Fusion hides the weaknesses of each sensor
The accelerometer is stable over long periods but nearly useless during motion, because real acceleration and gravity arrive mixed together. The gyroscope is excellent over short intervals and drifts badly over long ones, which is precisely the opposite pattern of failure.
Under load, combining them lets the slow absolute reference continuously correct the fast relative one, and that exchange is the heart of sensor fusion. A magnetometer adds an absolute heading, but it is disturbed by any nearby iron, magnet or wire carrying a current.
The figure-of-eight movement an application asks for exposes the magnetometer to many orientations so software can estimate and subtract local distortion.
The barometer nobody mentions
Most phones include a pressure sensor sensitive enough to detect the difference between one floor of a building and the next. It works as a tiny diaphragm sealed over a cavity, flexing as outside pressure changes and altering a capacitance once again.
Navigation uses it to decide which level of a road interchange or multi-storey building you are on, which satellites cannot resolve. Absolute altitude from pressure is unreliable because weather shifts the baseline, so the sensor is used for changes rather than heights. It also gives step counters a way to distinguish walking upstairs from walking the same distance on level ground.
How microscopic machines fail
A hard drop can drive the mass past its stops, and although stops are designed in, a severe shock may leave a permanent offset. The symptoms are subtle: a screen that rotates late, a compass reading consistently wrong, or step counts that drift steadily high.
Temperature changes the stiffness of silicon springs, so sensors carry calibration tables and still shift slightly when the device runs hot. Loud sound at the right frequency can excite a gyroscope's driven mass, which is why extreme noise has been shown to disturb some sensors. Because the mechanism is sealed inside its package, none of this is repairable; the part is replaced or the fault is simply tolerated.
The takeaway
The most mechanical thing in your phone was made by dissolving silicon away.
Once you know what it is trading away, the design stops looking arbitrary.
Questions readers ask
Why does my compass ask me to wave the phone around?
It is sampling the magnetic field in many orientations so the software can work out and subtract the distortion caused by the device itself.
Can these sensors be recalibrated at home?
Magnetometers usually can, through the movement routine. Accelerometer and gyroscope offsets are estimated automatically while the device is left still.





