15 March 2013, Clink Street, London
If you look at the history of watchmaking, many complications were mastered in medieval times - look at de Dondi's Astrarium, for example, which he completed in 1364 (after 16 years of work). Only later did the focus swing timekeeping accuracy. Most of Breguet's patents, for example, relate to timekeeping; Harrison's marine chronometer designs introduced temperature compensation and permitted Britain to build a sea-faring empire; in the last century, the quartz oscillator turned the watchmaking industry inside out.
Even today with quartz timepieces, temperature compensation is still necessary in the most accurate of watches, since they drop quadratically in frequency as the temperature deviates from 25°C. In our watches, we compensate by taking the temperature and then adding or subtracting a number of cycles every minute, followed by a fine-tuning adjustment every fifteen minutes. Each crystal needs to be individually trimmed to compensate for component differences. The photo shows our calibration station. Top left is the frequency counter, which you can see connected to the GPS disciplined oscillator (GPSDO) on the right. The GPSDO is a very accurate oscillator, with an accuracy of at least 1 part per billion, far in excess of what we need to tune our crystals.
Below is the power supply that feeds the Peltier heat pump (the white square block in the second photo, gunked to the circuit board with heat sinking silicone). The heat pump was an experiment in changing the temperature and measuring the resulting frequency change - it didn't work out too well, so we've since replaced it with a fridge and an oven.
Achieving accuracy is difficult. There are so many factors affecting the setup. What, actually is the temperature? (The variation even in different parts of our windowless lab is several degrees.) How much does the frequency actually vary with temperature? How accurate is the temperature reported by the thermistor circuit? What is the temperature gradient between the thermistor and the crystal? How accurate is the frequency counter? How accurate is the GPSDO?
I remember once, having calibrated a crystal, I wore the watch for a few days and compared it with the time on my iPhone. I was disappointed to note a deviation of a couple of seconds - far more than it should have been. I eventually worked out that the deviation was in the iPhone's clock, not the watch.
These days, we think we're achieving an accuracy in the order of a few seconds per year, but we haven't been able to independently verify this yet. The Contrôle Officiel Suisse des Chronomètres won't certify us because we're not Swiss, and the UK's National Physical Laboratory no longer provides the service in this country. This is something we're trying to fix in collaboration with a number of other UK watchmakers.