Clink Street, London, 15th August 2014
In these hot summer months, I start thinking about wetware. Wetware for tightly fitting curves. Just for once, though, I'm not thinking about bikinis.
Let me explain. First there was hardware - actual stuff you can drop on your toe and it will hurt. Then there was software - instructions that could be loaded onto a computer to perform a function. And then there was firmware - the instructions we put inside our chips on the Hoptroff movements that tell them how to turn. And finally wetware came along - artificial neural networks, or simulations of the human brain.
I have a long history in wetware. Back in the nineties I started a company called Right Information Systems, which produced forecasting software. The forecasts were based on artificial neural networks, because they happen to be very good at taking a lot of data points and drawing an accurate curve through them.
And I find myself doing it again today. Each Hoptroff watch is slowly heated to 50°C and cooled to 0°C, and all the time we take thermistor values and measure the crystal frequency using an atomic clock. That gives us a lot of data on the exact variation of each individual circuit's frequency with temperature reading, and believe me, it's not exactly what the data sheets say, and each one is different. The only way to get a really accurate timepiece is to calibrate each one individually.
So where does the wetware come in? We need it to fit a neural network curve to the data points. This curve is then written into the watch's memory and is used to compensate for variations in temperature. Here in the workshop, we're seeing 0.3 seconds/year accuracy - that's ten parts per billion. There are certain factors, such as crystal aging, that will mean we're unlikely to achieve quite that level of accuracy day-to-day in our customers watches. But it is world-beating, nevertheless.
So there you have it. A brain inside every Hoptroff watch.