Start two identical stopwatches at the same instant. Leave them running for a day. Come back.
They will not agree. Not because either is broken — because there is no such thing as two crystals cut exactly alike, and because the room they sat in has a bigger vote than you would guess.
The tuning fork inside
Almost every quartz timepiece contains a tiny fork-shaped sliver of quartz, cut and trimmed so that it vibrates at 32,768 times a second when a voltage is applied.
That number is not arbitrary. It does sit just above the top of typical adult hearing, and that is often quoted as the reason — but it is a side effect. The reason is that 32,768 is 2¹⁵. Halve it fifteen times with simple digital logic and you land exactly on one pulse per second. In an era when every transistor cost money and power, a frequency you can divide down with fifteen identical flip-flops was worth more than a rounder-looking number.
The crystal is a resonator, not a metronome handed down from an authority. Its frequency depends on its physical dimensions, and no manufacturing process cuts two of them precisely alike.
Tolerance: the disagreement you buy
Watch-grade crystals are typically specified at around ±20 parts per million at 25 °C.
Parts per million is an unhelpful unit until you convert it. Twenty ppm is about 1.7 seconds a day, or 52 seconds a month. That is the specification — a crystal inside that band is working correctly.
Now take two of them. If one sits at the top of its tolerance and the other at the bottom, they are 40 ppm apart from each other. That pair diverges by a full second roughly every seven hours — about three and a half seconds a day, and getting on for two minutes a month.
Neither watch is faulty. They are both performing to specification. Specification simply never promised they would agree with each other.
Temperature does most of the damage
The dominant term is not manufacturing spread, though. It is heat.
A tuning-fork crystal’s frequency traces a parabola against temperature. It peaks at a “turnover” point — typically somewhere between 25 °C and 28 °C, with 27 °C a common figure — and falls away on both sides. Crucially, it falls away quadratically: twice as far from the turnover means four times the error, and it is always in the same direction. Cold and hot both make the clock run slow.
Typical datasheet coefficients put this at roughly −0.03 to −0.04 ppm per degree squared, which turns into numbers you can feel:
- In a fridge at 0 °C — losing about 2 seconds a day
- On a sunlit car dashboard at 55 °C — about 2.7 seconds a day
- On your wrist at around 31 °C — about 0.1 seconds a day
Body temperature happens to sit near the turnover point, which is not a coincidence: wristwatch crystals are trimmed for it. It is also why a watch left in a drawer keeps slightly worse time than one that is worn, and why the same stopwatch will disagree with itself between a winter morning and a summer afternoon.
There is a third effect, quieter and permanent: aging. Crystals drift as they settle, commonly specified at ±3 to ±5 ppm per year, depending on how hermetically the crystal is packaged. A ten-year-old timer is not the timer it was.
What money buys
The expensive solution is to measure the temperature and correct for it.
Grand Seiko’s thermocompensated 9F movement takes the temperature inside the watch 540 times a day and adjusts, holding around ±10 seconds a year. Citizen’s Calibre 0100 goes further and specifies ±1 second a year — about 0.03 ppm, roughly a thousand times tighter than a standard crystal. It gets there by leaving the tuning fork behind entirely: a high-frequency AT-cut crystal, vacuum-sealed, with its own thermal compensation on top.
Worth clearing up a common misconception: these are not tiny ovens. A thermocompensated oscillator measures and corrects electronically. The oven approach is a different technology — the older one, as Marrison’s 1927 clock below shows, though far from obsolete: the best oven-controlled oscillators are still the stability leaders.
1927, and a clock in a box
The first quartz clock was built in 1927 at Bell Telephone Laboratories in New York by Warren Marrison and J. W. Horton — and it was already fighting this exact battle.
Its crystal ran at 50,000 cycles per second, not 32,768, and shifted by about four parts per million per degree Celsius. Marrison’s solution was brute force: put the crystal in an oven and hold it near 40 °C. Ninety-odd years later your watch solves the same problem by being trimmed to the temperature of your arm.
Quartz reached the wrist on Christmas Day 1969, when Seiko released the Astron at 450,000 yen — oscillating at 8,192 Hz, not the 32,768 that later became standard. The New York Times reported in January 1970 that around a hundred had sold.
Atomic clocks are not just better crystals
It is tempting to picture an atomic clock as the same idea, executed superbly. It is not the same idea at all.
A caesium standard does not approximate the second — since 1967 it has defined it. The second is 9,192,631,770 periods of a particular caesium transition. That is not a very good measurement of a second; it is what a second is.
The difference matters. A quartz crystal is an artefact whose frequency must be checked against something outside itself. An atomic standard is the reference everything else is checked against. Caesium fountains reach uncertainties of order one to two parts in 10¹⁶ — some eleven to thirteen orders of magnitude beyond a watch crystal — but the real distinction is one of kind, not degree.
So why is your phone right?
Because it is being told.
Your phone’s own oscillator is ordinary quartz with ordinary drift. What it has that a stopwatch does not is a network. Time protocols pull an atomic-referenced time from servers, typically landing within tens of milliseconds over the internet, and the phone nudges its clock into line.
But it does not do this continuously. Poll intervals can stretch to many hours, and between syncs the local crystal drifts at exactly the rates above. “My phone has atomic time” is half-true: it receives atomic-referenced time periodically, and then keeps ordinary time until the next correction.
What it means for timing something
Here is the reassuring part, and it is the reason none of this usually matters.
Drift is a rate, and rates need duration to accumulate. At 20 ppm, timing a four-minute run costs you about five thousandths of a second — far below the roughly two hundred milliseconds your thumb contributes at each end. For anything a person starts and stops by hand, the crystal is nowhere near the limiting factor.
It starts to matter when intervals get long. Timing an eight-hour process at 20 ppm is worth about half a second; a week is worth about twelve. If you are timing something that runs for days, the oscillator is finally in the conversation.
And it matters when you compare two devices. Your training partner’s phone and yours will not agree to the second, and after a month they may be a minute or two apart — even if both are perfectly healthy. Which is why, if two measurements have to be compared, they should come from the same instrument.
TiCaNo Stopwatch reads elapsed time from your device’s clock, so it inherits that device’s crystal and its corrections — and, like every stopwatch, it measures a difference rather than an absolute. A difference is the forgiving kind of measurement: whatever your oscillator’s personality, both ends of the interval are subject to the same one.
That is a nicer property than it sounds. It is also, in a small way, the same trick the whole history of timekeeping keeps returning to — agreeing on a shared reference and measuring against it, rather than trusting any single clock to be right on its own.
Sources
- Michael Lombardi, “The Accuracy and Stability of Quartz Watches” — NIST, Horological Journal, on turnover temperature, ppm conversion and measured watch spread
- Warren Marrison, “The Evolution of the Quartz Crystal Clock” — Bell System Technical Journal, 1948, on the 1927 clock, 50 kHz crystal and oven
- A walk through time — quartz clocks — NIST
- SI second — BIPM, on the caesium definition
- Network Time Protocol Version 4 (RFC 5905) — on poll intervals and typical internet accuracy
- Grand Seiko 9F quartz — Seiko, on thermocompensation and annual accuracy
- Abracon AEC-Q200 32.768 kHz crystals — for the ±20 ppm tolerance figure
- ECS Inc. — tuning fork crystal parabolic temperature curve — for the parabolic coefficient
- Electronics Notes — quartz crystal ageing — for aging rates by package type