Pick up a phone, tilt it, and the screen rotates. The interaction feels like magic in the way most well-engineered things do: instant, silent, boring in its reliability. Underneath, though, something genuinely mechanical is happening. A tiny structure carved out of silicon is bending under gravity, and a circuit is measuring how far it bent.
The component is called a MEMS accelerometer — micro-electro-mechanical system — and it is the closest thing modern electronics has to a moving part in the classical sense. It has beams. It has a mass. The mass sags when you tilt the phone, exactly the way a weight on a spring would sag, only the whole apparatus is etched into silicon and the sag is measured across a fraction of a micrometre of travel.
The proof mass on a silicon spring
Inside the sensor die is a small block of silicon called the proof mass, suspended by thin silicon beams that act as springs. When the phone is still and held flat, gravity pulls the proof mass straight down and the beams take a specific resting deflection. Tilt the phone on its side and gravity now pulls that same mass sideways relative to the chip. The beams bend a different way. The circuit reads the new position and infers the new orientation.
The beams are not separate components clamped to anything. They are continuous with the frame at one end and with the mass at the other, cut from the same slab of silicon, so the “spring” is really just a deliberately weak neck of material left behind when everything around it was etched away. Nothing slides. Nothing rubs. The structure only flexes.
The travel involved is minute — a fraction of a micrometre, less than the width of a red blood cell, which is about seven to eight micrometres across. That is enough to change the electrical geometry of the chip, but only barely, which is the whole engineering problem the rest of the die exists to solve.
That is enough.
How the movement is turned into a number
The dominant sensing technique in phone accelerometers is capacitive. The proof mass carries a set of silicon fingers that interleave with fixed fingers anchored to the chip frame, forming rows of tiny parallel-plate capacitors. When the mass shifts, the gap between each pair changes, and because capacitance depends on the distance between plates, the shift changes the capacitance predictably. Analog Devices describes the arrangement in its own reference material: the change from any single pair is so small that many movable and fixed electrodes are wired in parallel to build a signal worth reading, which then passes through charge amplification, demodulation and low-pass filtering before a sigma-delta converter turns it into a digital value.
Three axes means three sensing directions on the same die, oriented at right angles. One measures acceleration along the phone’s long axis, one across it, one through the screen. The vector sum of those three readings, when the phone is sitting still, points at the centre of the Earth. That is how “which way is up” is computed. It is not a compass reading and it is not a gyroscope reading. It is gravity, measured as a force, on a suspended mass.
Sample rates on consumer parts commonly run from a few hundred to a couple of thousand samples per second per axis, with the exact rate configurable in software. Bosch Sensortec, a supplier used widely in phones and wearables, specifies a configurable accelerometer output data rate from 12.5 Hz up to 1600 Hz in the datasheet for its BMI160 combined accelerometer and gyroscope.
The sample rate is not the same as the mechanical bandwidth, and the difference matters. Analog Devices notes in the same reference that a capacitive accelerometer’s usable bandwidth is only a few hundred hertz, because the spring geometry and the gas sealed inside the package damp the mass. The circuit can read the structure thousands of times a second. The structure itself cannot respond that fast.
How the beams are actually made
The beams and the proof mass are not assembled from parts. They are etched, in place, from silicon using photolithography and reactive-ion etching. A review of monolithic multi-axis capacitive accelerometers in Micromachines describes deep reactive-ion etching as the standard route to the high-aspect-ratio structures these devices need, and notes the trade-off: thicker bulk-micromachined structures give lower noise and better stability, at higher cost and process complexity than surface micromachining.
They are also small. STMicroelectronics, one of the largest suppliers of these sensors, advertises 2 × 2 × 1 mm packages across its accelerometer portfolio. The moving silicon inside is a fraction of that. A grain of long-grain rice is roughly 6 to 7 millimetres long, so calling the sensing structure smaller than a grain of rice is, if anything, generous to the rice.
The important consequence is economic. Because these structures are made with standard integrated-circuit process technology rather than machined and assembled, they inherit the cost curve of the semiconductor industry rather than the cost curve of precision mechanics. That is most of the explanation for where the parts ended up.
Why this sensor ended up in everything
The idea of a suspended mass measuring acceleration is very old. What changed was the price. Analog Devices, writing about integrated circuits that only work because they have moving parts, records that electronic accelerometers once cost hundreds or thousands of dollars and that the cheapest now cost a dollar or two — cheap enough for airbag deployment, which was their first major application, and then for games controllers, disk-drive shock protection and projector keystone correction. The same note explains why the mechanics survive: silicon has very low mechanical hysteresis and does not degrade even when flexed trillions of times.
In a phone, the accelerometer feeds more than screen rotation. Step counting, tap and double-tap detection, freefall detection so the storage system can prepare for impact, gesture recognition, and — in combination with the gyroscope and magnetometer — the orientation used by maps and augmented reality all draw on the same stream of numbers coming off the same little sagging mass.
Outside the phone, the applications are stranger than the phone ones. Applications engineer Harvey Weinberg, writing in Analog Devices’ Analog Dialogue on which accelerometer ideas actually work, lists washing machines that jog the drum back and forth to rebalance a spin load, car alarms that trigger on an inclination change of more than half a degree per minute because the vehicle is being towed, and pedometers accurate to within about one per cent of walking distance. He also rules some ideas out: dead-reckoning navigation from accelerometer data alone accumulates error too fast to be usable.
The accelerometer is part of a larger family of MEMS parts that moved into consumer devices behind it: gyroscopes that measure rotation using vibrating rather than spinning structures, microphones built as micromachined diaphragms, pressure sensors. Optical image stabilisation, often credited to the accelerometer, is mostly the gyroscope’s job. Phones are unusual mainly in how many of these tiny mechanical devices they carry at once.
What the reading is not
A common misconception is that the accelerometer knows the phone is tilted. It does not. It knows the direction and magnitude of the net force acting on the proof mass. When the phone is sitting still, that force is gravity, and the direction of gravity is a useful proxy for orientation. When the phone is being shaken, swung or carried in a moving car, the reading is a mix of gravity and whatever else is happening, and the software has to decide what that mix means.
Specifically, the sensor cannot separate a tilted phone at rest from a phone being accelerated in a straight line. Both arrive as a force on the mass, and nothing in the raw reading distinguishes them. What it can separate is rest from freefall. Android’s motion sensor documentation gives the two numbers directly: a device sitting on a table and not accelerating reads a magnitude of 9.81 m/s², and a device in free fall, accelerating toward the ground at 9.81 m/s², reads zero.
That collapse to zero is a usable signal. In freefall the proof mass is not pressed against its springs at all, on any axis, and shock-protection systems in laptops and disk drives use exactly that signature to park heads before impact.
It is also why phones pair the accelerometer with a gyroscope, which reports rotation rate rather than force. The same Android documentation describes the software sensors built on top of both: a gravity sensor, a linear-acceleration sensor that subtracts gravity out, and a rotation-vector sensor, all of which require a gyroscope to be present and none of which the accelerometer could produce on its own.
The scale of the thing
It is worth holding the picture in mind. A structure smaller than a grain of rice, containing beams thinner than a human hair, suspending a mass you could not see without a microscope, is bending by fractions of a micrometre under the weight of gravity. A circuit next to it is turning that bend into a number a thousand times a second or more. Multiply by three axes, multiply by every phone shipped in a year, and the total motion being measured, worldwide, at any given moment is somewhere between staggering and comic.
Phones are full of components like this. The radio picking up Wi-Fi, the microphone hearing a voice, the camera resolving light — each turns out, on inspection, to be a small physical event being converted into a number. The accelerometer is one of the more literal cases. Something in the phone is actually, mechanically, falling.
The screen rotates because a beam bent.