A quartz watch, slowed down

A stone that
keeps counting.

A mechanical watch counts swings of a wheel. A quartz watch counts vibrations of a laser-trimmed sliver of silicon dioxide, 32,768 of them every second, then divides that number in half fifteen times until one pulse is left over, and spends that pulse turning a magnet through half a revolution.

Follow the signal ↓

01 / The whole chain

Seven stages,
two domains

Select a stage. Everything left of the motor is electronics; everything right of it is a watch in the ordinary sense.
Signal path through a quartz watchBattery, oscillator with quartz crystal, fifteen-stage divider, motor driver, Lavet stepper motor, reduction gear train, and hands. ELECTRICAL DOMAIN · no moving parts MECHANICAL DOMAIN BATTERY1.55 V DCsilver oxide, 25 mAh OSCILLATOR32 768 Hzquartz fork + CMOS DIVIDER÷ 2¹⁵15 toggle flip-flops DRIVER1 pulse / salternating polarity LAVET MOTOR180° / step30 rpm mean GEAR TRAIN21 600 : 1rotor to hour hand HANDS1 step / s6° of dial per tick AVERAGE CURRENT BUDGET · total ≈ 1.3 µA from a 25 mAh cell ≈ 2.2 years oscillator 0.3 µAdivider + logic 0.5 µAmotor pulses 0.5 µA The motor is the only part that does mechanical work, and it costs about as much as all of the silicon put together.

Educational schematic. Tooth counts, currents, and layouts are typical rather than specific to any one calibre.

02 / The resonator

Why a stone
vibrates on cue

Quartz is piezoelectric because its lattice has no centre of symmetry. Squeeze it and it produces a voltage; apply a voltage and it changes shape.
Quartz tuning fork vibrating in its flexural modeTwo etched tines bend in antiphase; their momenta cancel at the base so almost no energy escapes into the mount. hermetically sealed mount · vacuum can FLEXURAL MODE · TINES IN ANTIPHASE L ≈ 2.4 mm t ≈ 0.1 mm (tine thickness) tip travel shown×2000 exaggerated:real excursion ≈ 1 µm tine A: + tine B: − net momentum at the base ≈ 0
  1. 1
    No centre of symmetry

    α-quartz belongs to trigonal class 32. Strain displaces the Si⁴⁺ and O²⁻ sublattices by different amounts, so a deformed crystal carries a net dipole and a surface charge.

  2. 2
    The converse effect drives it

    Electrodes patterned along each tine apply a field that makes one side of the tine extend and the other contract. A bar that lengthens on one face and shortens on the other bends.

  3. 3
    The direct effect reads it back

    The same electrodes pick up the charge generated by bending. That is what closes the feedback loop: the crystal is simultaneously the actuator and the sensor.

  4. 4
    Antiphase is the whole trick

    The tines carry equal and opposite momentum, so the reaction forces cancel where the fork meets its mount. Almost nothing leaks out, which is why Q reaches 10⁴ to 10⁵. Drive the tines in phase instead and the base shakes, energy escapes, and the resonance collapses.

Nominal frequency32 768 Hz2¹⁵ exactly
Quality factor5×10⁴ – 10⁵in a sealed vacuum can
Motional resistance20 – 50 kΩthe loss the circuit must beat
Ringdown time≈ 0.5 s2Q/ω after drive removed
Geometry sets the pitch

For a flexural tine, f ∝ (t/L²)√(E/ρ). Frequency scales with tine thickness and with the inverse square of tine length, which is why the fork is photolithographically etched to micron tolerances and then finished by laser.

Trimming by evaporation

Final tuning removes or adds mass at the tine tips, where the mode shape has maximum displacement and therefore maximum sensitivity. Ablate gold and the frequency rises; deposit gold and it falls. A few parts per million per pass.

Why 32 768 and not 100 000

It is 2¹⁵, so fifteen halvings land exactly on 1 Hz with nothing but toggle flip-flops. It is also a sweet spot for power: CMOS dynamic power goes as C V² f, so a higher-frequency crystal buys short-term stability at a linear cost in battery life.

The resonator, in one linef ≈ 0.162 · (t / L²) · √(E / ρ)

With E ≈ 78 GPa and ρ = 2650 kg/m³ for quartz, a tine roughly 2.4 mm long and 0.1 mm thick lands near 32 kHz. Everything after this point in the watch is counting; nothing after this point decides the rate.

03 / The electrical twin

A mechanical
resonator, in ohms

To the circuit the fork looks like an LCR branch of absurd values in parallel with the capacitance of its own electrodes. Those values are the mechanics in disguise.
Butterworth–Van Dyke equivalent circuit of a quartz crystalA motional branch of inductance, capacitance and resistance in parallel with the static electrode capacitance. C₀1.4 pF · electrodes + can L₁7 863 H · tine inertia C₁3.0 fF · stiffness R₁30 kΩ · all losses BUTTERWORTH–VAN DYKE MODEL Q = ωL₁/R₁ ≈ 54 000
Crystal impedance against frequencyA deep series-resonance minimum and a sharp parallel-resonance maximum separated by about a thousand parts per million. frequency · Hz (span ≈ 63 Hz, or 1 900 ppm) |Z| · ohms (log) fₛ fₚ operating point
Series resonance32 768.0 Hz|Z| minimum, = R₁
Parallel resonance32 803.1 Hz|Z| maximum
Operating frequency32 772.7 Hz+144 ppm above fₛ
Quality factor54 000rings for 0.52 s
Eight thousand henries

No real inductor of that value exists at this size. L₁ is not an inductor. It is the mass of the vibrating tines expressed in electrical units, because a series LCR and a mass–spring–damper obey the same second-order equation. C₁ is the tine stiffness; R₁ is every loss mechanism added together.

The gap is the stability

The circuit can only pull the frequency inside the window between fₛ and fₚ, and that window is only about C₁/2C₀ ≈ 1000 ppm wide. The crystal simply refuses to run anywhere else. Compare a balance wheel, whose rate moves with amplitude, position, and mainspring torque.

What the trimmer really does

Sliding CL from 4 pF to 20 pF moves the operating point by roughly 200 ppm, about nine minutes a month of adjustment authority. That is the entire regulation range of a quartz watch, and in a modern movement it is done digitally instead.

04 / Sustaining the vibration

One inverter,
run backwards

A digital gate biased into its own forbidden analogue region becomes an amplifier. Wrap the crystal around it and it has no choice but to oscillate at 32 768 Hz.
CMOS Pierce oscillator schematicAn inverter with a feedback bias resistor, a series drive resistor, the crystal, and two load capacitors to ground. –A Rf ≈ 20 MΩ Rd 32 768 Hz fork Cg Cd gate nodedrain node CMOS PIERCE OSCILLATOR inverter 180° + pi-network 180° = 360° · loop gain > 1 → oscillation
Oscillator waveforms and startup envelopeGate and drain voltages in antiphase, and the exponential growth of amplitude from thermal noise at power-on. STEADY STATE · ONE CYCLE = 30.5 µs gatedrain the two nodes are always 180° apart START-UP FROM THERMAL NOISE 00.5 s1.0 s amplitude envelope · τ = 0.12 s
Negative resistance−164 kΩ−gm/(ω²CgCd)
Gain margin5.5×oscillation sustains
Start-up time constant0.12 s2L₁/(|Rneg| − R₁)
Oscillator draw0.3 µA≈ 0.45 µW
Negative resistance, not gain

The tidiest way to see a Pierce oscillator is that the inverter and its two capacitors present −gm/(ω²CgCd) to the crystal. If that negative resistance is larger in magnitude than R₁, net loss is negative and any disturbance grows. Designers want three to five times margin, no more, because excess margin wastes current and overdrives the fork.

Why it takes a second to start

Amplitude grows as e^(t/τ) with τ = 2L₁/(|Rneg|−R₁). The same enormous L₁ that makes the fork stable makes it slow to wake. A high-Q resonator is a flywheel: hard to disturb, and equally hard to spin up.

Drive level matters

Rd limits how hard the fork is driven. Overdrive causes non-linear bending, accelerated ageing, and in extreme cases fracture at the tine roots. Watch crystals are typically driven at well under a microwatt, and the tines move about a micron.

05 / Counting down

Fifteen halvings
land on one

32 768 is 2¹⁵. A chain of fifteen toggle flip-flops turns the crystal’s buzz into exactly one pulse per second, using nothing but binary arithmetic.
counter 00000000000000 · 0.000 s
Fifteen-stage binary divider chainEach toggle flip-flop halves the frequency of the previous stage, ending at one hertz. RIPPLE DIVIDER · EACH STAGE TOGGLES ON ITS INPUT’S FALLING EDGE 32 768 Hz in 1 Hz The first stages are a blur even in slow motion. That blur is the precision: everything downstream inherits it.
Waveform ladder of the first five divider stagesEach trace has twice the period of the one above it. ONE PERIOD OF STAGE 5 every stage is a perfect 50 % square wave · edges never drift relative to one another
  1. 1
    A toggle flip-flop

    A D flip-flop with its inverted output wired back to its own input. Each active clock edge flips its state, so the output changes once per two input cycles. That is division by two, exactly, with no analogue error term.

  2. 2
    The chain is free

    Fifteen stages is a few hundred transistors. In 1969 this was the expensive part; today the divider is a rounding error on the die and the crystal is the costly component.

  3. 3
    Power lives in the first stage

    Dynamic power is C V² f per stage, and f halves each time. The whole chain therefore costs about twice what stage one costs. Designers shrink the first flip-flop and let the rest be lazy.

  4. 4
    The pulse is shaped, not just tapped

    The motor does not want a 50 % duty cycle at 1 Hz. It wants a 4–8 ms kick. The driver gates a fast stage (say 256 Hz) with the 1 Hz stage to cut a short pulse, then alternates its polarity every second.

06 / Back into the world

The most-built
motor on Earth

Every second, one pulse turns a magnet the size of a grain of rice through exactly half a revolution, and it must never turn the wrong way.
Lavet-type single-phase stepper motorA bipolar magnet rotor in a notched soft-iron stator bore, driven by one coil with alternating pulse polarity. NS stator field axis detent axis, 45° off index notch index notch COIL i = 0 RESTING · held by detent torque step 0 · rotor angle 0° · pulse polarity + coil ≈ 12 000 turns of 20 µm wire · 2 kΩ · rotor ≈ 1.4 mm SmCo, magnetised across a diameter soft-iron stator · saturable isthmus at each notch sets the rest orientation Polarity alternates every second, so the rotor sees a reversing field but always turns the same way.
Torque against rotor angle, and the simulated stepDetent torque, coil torque, their sum, and the resulting angle-versus-time trajectory of one step. TORQUE vs ROTOR ANGLE 90°180°270°360° 0µN·m grey = detent · gold = coil · blue = totalthe coil pushes the rotor the whole way past 90°, then the detent captures it at 180° SIMULATED STEP · ANGLE vs TIME 180°360° 020 ms40 ms STEP COMPLETED
Coil current, pulse mean116 µA1.55 V / 2 kΩ, chopped
Energy per step0.88 µJ86 400 steps a day
Mean current1.37 µAmotor + logic
Predicted cell life2.1 yearsfrom 25 mAh
Why the notches exist

With a symmetric bore the rotor would rest exactly along the stator field axis. A pulse would then produce zero starting torque and no preferred direction: a dead centre. The notches saturate locally and rotate the rest orientation about 45° away, guaranteeing both a starting torque and a sign for it.

Why polarity alternates

After one step the rotor has turned 180°, so the same field polarity would now push it backwards. Reversing the pulse each second restores the geometry. The bonus is zero net DC through the coil: no electrolytic corrosion, no progressive demagnetisation.

Adaptive drive

Modern ICs cut the pulse short, then sense the rotor’s own back-EMF to confirm it moved. If the step landed, the next pulse is trimmed shorter still; if it failed, a full-power correction pulse fires immediately. The motor therefore runs permanently a few percent above its own failure threshold, which is where the battery life comes from.

07 / Reduction

The gear train,
running backwards

A mechanical watch gears up from a slow barrel to a fast escapement. A quartz watch gears down from a fast rotor to slow hands. Same wheels, opposite errand.
Reduction train from stepper rotor to hour handSuccessive reductions take thirty revolutions per minute down to one revolution per twelve hours. ROTOR30 rpm mean INTERMEDIATE6 rpm FOURTH1 rpm · seconds hand THIRDtransmission CENTRE / CANNON1 rph · minute hand HOUR WHEEL1 turn / 12 h 30 × 60 × 12 = 21 600 : 1 from rotor to hour hand · no escapement anywhere in this chain Wheel sizes and ratios are schematic; real calibres split the reduction differently and often add a second motor for the date.
Stepping seconds hand versus sweeping seconds handThe quartz dial jumps six degrees once a second; the mechanical dial advances in small beats. QUARTZ · 1 step / s MECHANICAL · 8 beats / s
  1. 1
    The tick is a signature

    A quartz seconds hand advances 6° once per second because the rotor makes exactly half a turn per second and the train divides by sixty. The “sweep” of a mechanical watch is really 28 800 tiny steps an hour, also discrete, just below the threshold where the eye separates them.

  2. 2
    Loads are trivial

    The train carries roughly a microjoule per second. Pivots need no jewelling for wear, wheels are often moulded polymer, and lubrication requirements are mild. The engineering difficulty moved upstream into the silicon.

  3. 3
    Nothing here regulates anything

    This is the deepest structural difference. In a mechanical watch the train is inside the timing loop, so friction and torque variation reach the balance and change the rate. In a quartz watch the train is strictly downstream: gum it up and the hands stop, but until they do they are never late.

  4. 4
    More hands, more motors

    Quartz chronographs use two to four independent Lavet motors, one per subdial, because adding a motor is cheaper than adding a differential. Perpetual-calendar quartz drives the date ring from its own motor with an independent counter.

08 / What is left to go wrong

One parabola,
and a trick

A tuning-fork crystal’s rate falls off as the square of temperature away from its turnover point. Almost the entire error budget of an ordinary quartz watch is that one curve.
Frequency deviation against temperatureA downward parabola with its turnover near 25 degrees Celsius, and the flat residual after digital thermocompensation. temperature · °C rate deviation · ppm Δf/f = −β(T − 25 °C)² · β ≈ 0.035 ppm/°C²
Rate deviation0.00 ppmat the turnover point
Per day0.00 s±0.07 s/day = ISO quartz chronometer
Per month0.0 s±15 s/month = ordinary quartz
Per year0 s±10 s/year = thermocompensated
The error only goes one way

The parabola opens downward, so every departure from about 25 °C makes the watch run slow. A watch on a 33 °C wrist by day and a 20 °C nightstand by night is slow in both states. This is why ordinary quartz watches, as a class, lose rather than gain.

Trimming by deletion

The crystal is deliberately left running slightly fast. Once a minute the IC deletes a few counted pulses. One pulse per minute out of 32 768×60 is 0.51 ppm, about 1.3 seconds a month of resolution, adjusted in the digital domain where the adjustment itself cannot drift.

Thermocompensation

Add a temperature sensor, sample it every ten to sixty seconds, and look up how many extra pulses to inhibit. The parabola is a known, stable, per-crystal curve, so it can be cancelled almost entirely. What remains is ageing, roughly 1–2 ppm in the first year, less thereafter, from stress relaxation in the mount and mass transfer on the tines.

The whole error budgetΔf/f = −β(T − T₀)² + ageing + trim residual

No amplitude term. No positional term. No torque term. A mechanical watch has all three, and they interact. That absence, not the raw frequency, is what makes quartz timekeeping a different category rather than a better version of the same thing.

09 / The comparison

Two ways to
divide a second

Both watches do the same three jobs: hold energy, generate a repeating interval, count it. They differ in where the interval comes from, and in one number.
Ring-down comparison of a balance wheel and a quartz forkAmplitude decay against number of cycles on a logarithmic axis; the quartz fork survives hundreds of times more cycles. FREE RING-DOWN · AMPLITUDE vs CYCLES 11010²10³10⁴10⁵ cycles of free oscillation balance wheel · Q ≈ 250 quartz fork · Q ≈ 54 000 Q counts the radians of oscillation a resonator keeps before losing 1/e of its energy,which is the same thing as how hard it is to talk out of its own frequency.
The one number that decides itδf/f ≈ φ / 2Q

For a resonator perturbed by a phase error φ per cycle (friction, an off-centre impulse, a change in position) the fractional rate error scales inversely with Q. A quartz fork has roughly two hundred times the Q of a balance wheel, so it converts the same disturbance into two hundred times less error. Everything else (the divider, the motor, the trim) is bookkeeping around that fact.

Two hybrids sit between the categories. Seiko’s Kinetic replaces the battery with a rotor-driven generator but keeps the quartz timebase. Spring Drive keeps the mainspring, barrel and gear train of a mechanical watch and deletes only the escapement, replacing it with an electromagnetically braked glide wheel that a quartz oscillator holds to exactly eight turns per second, which is why its seconds hand genuinely sweeps.

Mechanical, chronometer gradeQuartz, ordinaryQuartz, thermocompensated
Timebasebalance wheel + hairspringetched quartz tuning forkquartz fork + temperature sensor
Frequency4 Hz · 28 800 vph32 768 Hz32 768 Hz
Quality factor200 – 3005×10⁴ – 10⁵5×10⁴ – 10⁵
Rate spec−4 / +6 s per day±15 s per month±10 s per year
Fractional≈ 5×10⁻⁵≈ 6×10⁻⁶≈ 3×10⁻⁷
Dominant errorposition, amplitude, mainspring torque, temperaturethe temperature parabolacrystal ageing
Regulationalter effective hairspring length or balance inertiatrimmer capacitor or fixed digital inhibitiontemperature-indexed digital inhibition
Energy sourcemainspring, ~40 h reserve, rewound by the wrist25 mAh cell, 2–3 yearscell, 5–10 years with a low-drain IC
Train’s roleinside the timing loop; friction changes the ratedownstream only; friction can stop the hands but cannot make them late
Failure modedrifts, then stopskeeps perfect time, then stops dead

The governing idea

Mechanical watches
measure. Quartz
watches count.

An escapement is an analogue negotiation: energy goes in, the balance answers with an interval, and every imperfection in the negotiation shows up as rate. A quartz watch removes the negotiation. The crystal decides the interval alone, in a domain where a divider can be exact and a trim can be an integer. What is left for the mechanism to do is to spend one pulse a second turning a magnet, the only place in the whole watch where anything is still allowed to be difficult.