Crystal Aging
There have been a number of threads on the Time Nuts Mailing List about crystal aging, what affects it and how to predict/compensate for it. Here are a few posts from October 2007:
I was the person who posted the original comments. My comments were based on discussions with notable crystal guru, Jack Kusters, who was merely an inventor of the SC cut. He didn't have much confidence in the various FCS papers Jeffrey was referring to.
Basically, he was aware of all those effects and if there were any consistency to the proposed treatments, the HP crystal fab would have been doing them. For instance, the idea that, if only you polish the crystal well enough, it won't age has been debunked. Some people currently making high quality crystals learned much of what they are doing from Jack in the first place.
If any of them go beyond what Jack taught them, they might publish at FCS. Or they might not and keep it as a trade secret.
Rick Karlquist
Indeed, the observed crystal aging is a sum of a larger number of physical processes:
- Some of them are stronger than others
- some of them decay faster or slower than other others
- many of them show a logarithmic frequency change over time
- a few of them have the shape of an exponential decay.
The usual approximation used to predict frequency aging - as used in MIL-PRF55310 - is df/f = f0 + a1*log(a2*t+1), where a1 can be understood as a measure of the strenght of aging, and 1/a2 is a kind of time constant, i.e. small a2 values stand for slowly aging processes and large ones for fast agers.
A better approach to describe frequency aging would thus be a mathematical sum of several logarithmic terms as in the above equation.
As nicely shown in John Vig's tutorial, you can easily get a reversal of the aging rate (i.e. a change from positive to negative aging or vice versa), if you simply superimpose a fast and weaker (positive) aging process with a slower but stronger negative aging process - see attached copy.
O.K. you now can say: why not make aging prediction by mathematical fitting of a sum of logarithmic terms through the observed frequency data over time? Basically this would/could improve the degree of fitting, correct. BUT: There is a big leveraging effect, if you make a mathematical prediction over a long period of time from data taken over a relatively small time period. In other words: If you just vary one or a few data points by a very small amount (i.e. to eliminate jumps), the effect on the extrapolated curve is enormous. Just try it using EXCEL's solver …
I had published an example for that 10 years ago - see a copy of my paper “ Correlation of predicted and real aging behaviour” on our website www.axtal.com. In that paper I used data from about 9 months of aging, and compared the predictions made from different time periods with the real aging.
From my standpoint the MIL-PRF55310 method of prediction is just a standardized method to define how the aging numbers in a spec are “verified” in a reasonable time of 4 weeks, It does not say: “This is how the crystal will age over decades”
There is much more to say, but I will stop here, hoping that this contribution gave some time nuts a better understanding - and may kill their firm belief into aging predictions 
Best regards
Bernd Neubig DK1AG
I support what Rick has said about crystal ageing. I also work for an OCXO manufacturer, and will add my 2c worth.. Here are some points -
- Manufacturers do indeed have their own special techniques for minimizing ageing and numerous other quality related parameters. Cleanliness, quality and attention to process are very important. Many techniques will be specific to their own manufacturing process, and of course they are trade secrets.
- Because many different ageing processes are involved, prediction of future ageing from past history is not in general reliable. A very good analogy is prediction of weather - to say that tomorrow's weather will be similar to today's is a fairly safe bet, but is no help in predicting the weather for next weekend! The same applies to good crystal oscillators - you can reasonably expect the rate of ageing next month to be closely similar to last month's, but the slight differences build up in extrapolation, and prevent long term prediction. Make your measurements, predict the trend, then do the same next month and note the differences.
- It is very important to recognise the difference between INITIAL ageing (when a newly made crystal is first used) and long term ageing. Most users will never see initial ageing, as it takes place in the factory. Initial ageing is much more predictable, and the factory will monitor this and from this behaviour, within a week or so can predict when (if ever!) the device will be within specification and so ready for final calibration and delivery.
I recommend reading the various papers on ageing published by the UFFC and others. Start with this fairly comprehensive and authoritative one:
http://www.ieee-uffc.org/freqcontrol/vigaging91/aging.htm#CONTENTS
73, Murray ZL1BPU
I have been working hard on adaptative algorithms to predict OCXO ageing, and I completely agree with Bernd.
Besides the fact that ageing is actually random (althoug it can be linearly modeled for time intervals of days, may be a week), there is an effect that should be added to ageing: temperature variations. Depending on many factors, temperature effect can be orders of magniture more significant than ageing.
Moreover, temperature change (altough small in a good oscillator) produce non linear effects: change temperature and go back to original one: frequency will not be the original. It is beautiful to perform an experiment: measure temperature and oscillator frequency and change temperature smoothly (e.g. normal room day/night variations). Revove ageing from frequency measurements and make a temperature/frequency X-Y plot. Do not expect a line, but something simmilar to a cloud. It slongly depends on oscillator and the amount of temperature variation.
My experience is that nice prediction ('nice' definition: time drift lower than, say 1 us) is difficult to maintain for more than few days, but this strongly depends on temperature variations and oscillator characteristics.
Please notice that 1 us in one day is a really good figure: these all discussions depend on how good matching you want to obtain.
I have run all my work with free running OCXOs. I have no direct experience on disciplined oscillators, but I suspect that non linear effects may be even worst.
Best regards
Luis Miguel Brugarolas
A Xtal is actually a mechanical oscillator, with the quartz slab vibrating (in either its fundamental mode, or on an odd overtone); quartz is a piezo-electric material so the voltage across the pins of the xtal has a direct connection to the mechanical vibration.
When a xtal oscillator is powered up, the associated amplifier generates noise, which then starts the xtal vibrating, which filters out the noise that is not at the right frequency and a feedback loop is set up. When you crank up the power to the mechanical resonator, the signal increases with respect to the background noise (i.e. S/N gets better) which improves the short-term stability. Going in the other direction, the mechanical resonant frequency changes with time because, as the xtal vibrates, microscopic cracks in the structure of the quartz get bigger. Running at high power makes the crystal generate these microscopic faults at a faster rate; this then causes the oscillator to have poorer long-term stability. When a xtal is left vibrating (oscillating) in an undisturbed environment, the rate of cracking of the quartz decreases, and the oscillator is said to “age” to its final frequency.
But if you subject that same crystal to a mechanical jolt, some new cracks are created which re-start the aging “diffusion” process. Ditto turning the oscillator on & off or a thermal shock can aggravate the aging.
If the metal can or glass envelope around the xtal outgasses, some of the resulting crud (a very scientific term!) from the envelope and seal will deposit onto the quartz and also cause aging. For this reason, only the cheapest crystals are housed in a metal can with a solder seal; cold welding of the can is a much better procedure; and a glass envelope is the best (not everyone agrees on this point, see note from Jack Kusters below). Cheaper than cheap are the WW2 “FT243” xtals where the seal is just a rubber gasket or the epoxy seals used in some consumer-grade surface mount oscillators.
The main reason that the 32768 Hz oscillators operate at low power is so that watches can run for years on small batteries. But even at that, the mechanical xtal resonator (which is built as a tuning fork for these low frequencies) is much better than any watch escapement ever was!
73, Tom Clark
Hi,
I posed this question to Jack Kusters, now retired from HP/Agilent. He and Charles Adams commercialized the SC-cut crystal for HP in the 10811A oscillator. He gave me permission to post his response on the reflector.
Jim Johnson
Hi Jim,
In addition to everything Tom Clark said (I agree in general with his explanation), there is another aging mechanism associated with stress. When the crystal blank is manufactured, it is sawn, lapped, ground, etched, and otherwise abused. All of this produces stress in the blank. In addition, there are mounting stresses that arise because of the way the blank is mounted on its header, surface stresses that develop because the electrode material when evaporated and then condensed on the surface shrinks as it cools.
All of these result in long-term aging as these stresses need to equilibrate out. In addition, there are other stress related mechanisms that may result in either long- or short- term aging. The quartz material is anisotropic, the mounts, electrode material are isotropic.
So, lets assume that we've had the crystal at an elevated, constant temperature. Over a period of time, all stresses, material, mount, electrode, cracks, etc. equilibrate to their lowest energy level and it appears that aging has stopped.
Now, take it down in temperature. The anisotropic quartz and the isotropic mount and electrode, have different contraction rates, so the overall system now has a new set of stresses.
Let the unit come to full equilibrium at the new, lower temperature. Now take it up in temperature to where it was before. Now we see a whole new set of aging and stress relief. The only virtue is that aging due to cracks and material stress from manufacturing processes should be mostly gone, so the unit should come to equilibrium much faster.
One further comment, glass sealed crystals are not necessarily the best way to seal a crystal. It takes heat from a source sufficiently elevated in temperature to melt the glass. This tends to cause contaminents to migrate from the area being sealed to a cooler spot in the package, usually the crystal. Contaminents come from gasses from the torch or from junk trapped in the glass.
The cleanest mount one can do is a cold-weld seal under proper conditions. For example, the HP crystals were put into a vacuum furnace, heated to 300+ deg-C overnite at 10E-7 torr with the can stored next to the crystal. After reducing the temperature to about 80-84 deg-C, the crystal was frequency plated to within several parts in 10E7. After that, the mount was placed in the can, the temperature raised up to about 150 deg-C, stabilised in temperature and vacuum, then cold-welded.
Done properly, there is essentially no contamination inside the crystal assembly, most of the other stresses are gone, and the typical HP SC-cut would reach an aging rate of better than 1E-7 per day, within the first 5 days.
Best regards,
Jack Kusters
