[time-nuts] Is quartz crystal aging really a logarithmic curve?

WarrenS warrensjmail-one at yahoo.com
Sat Jul 10 20:59:16 EDT 2010


As you likely know, most good xtral oscillators tend to have a lot of drift, 
when first turned on after being powered off for a while.
Attached is what one well run in HP 10811 double oven osc looked like when 
first turned back on after being off for about a month.
Drift was about 5e-9 (0.05 Hz) during the first day, with ageing looking 
pretty logarithmic out to about 3 or 4 days.
What I have seen is that some tend to settle down to more like parts in e-11 
/day given enough time and a good environment.
This one is still about ten times that after 6 days.

Also attached is a one week plot from a well ran in single oven 10811, which 
has about a 3e-11 / Deg C temperature coefficient.
On this osc it is the changing room temperature, not time what is the more 
significant source of freq drift.

Three things I have found with oscillators is:
a) One of the better ways to improve short term performance is clean up the 
Power supplies and ground loops.
b) One of the better ways to improve medium term performance is DO not let 
the temperature change.
c) One of the better ways to improve longer term performance is let it age.
If you log the drift as it is aging, (and take into consideration other 
effects such as temperature change), you can see from the data plot when it 
has been long enough.



If you have a *really bad* crystal, it will follow a very nice log aging 
cure. A good crystal is much less predictable.
The reason is fairly simple, the bad crystal has a single dominant cause for 
it's aging.
It missed a step somewhere along the line and it's got a problem.
A precision part is going to be run through a process that results in no 
single effect being dominant.


On Jun 24, 2010, at 2:40 PM, Don Cross wrote:

Hi, my name is Don.  I have been lurking on this list for a while, so here 
is my first post.
I am a hobbyist who has just built my second home-made quartz digital clock. 
Both are based on a microcontroller that counts timer interrupts and uses 
software tricks to allow me to tune the clock rate based on comparison with 
atomic time via NTP.  The second clock uses a temperature sensor and a 
heater (a grid of resistors) enclosed with the microcontroller board in a 
glass jar to regulate its own temperature.
Of course, I am noticing a drift in the clock rate over time due to crystal 
aging.  The frequency of the crystal is gradually increasing over time.  For 
example, on June 2, the frequency was 15.99927052 MHz.  As of yesterday it 
is hovering around 15.99927796 MHz.
I have read through several online resources, including the very interesting 
one that was posted here recently:
On page 4-7 of that document there is a slide titled "Typical Aging 
Behaviors" where it shows that long-term crystal aging can be represented as 
a sum of logarithmic functions.  I was wondering if this is just an 
approximation, or if there is a theoretical reason why logarithms would 
describe such phenomena?  Looking at the causes of aging, they seem to do 
with changes in the bonding with the electrodes, deposition or oxidation of 
the components inside the crystal enclosure, etc.
I am thinking about trying to measure the aging process over the coming 
months, and then try to model and even predict future aging.  If I can get 
that to work, perhaps I can even incorporate the formula for predicted aging 
right into my software.  Any insights on this would be much appreciated.
FYI, here is a link to what I did on my first clock.  I have not yet 
finished the web page for the second clock, but I will get around to it, 
I do have the firmware for my second clock online, along with some crude 
schematics, if anyone is interested:
- Don 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Osc Startup Drift ws1.gif
Type: image/gif
Size: 27928 bytes
Desc: not available
URL: <http://lists.febo.com/pipermail/time-nuts_lists.febo.com/attachments/20100710/fb268683/attachment.gif>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 7day_Ref 6-19-10c.gif
Type: image/gif
Size: 45045 bytes
Desc: not available
URL: <http://lists.febo.com/pipermail/time-nuts_lists.febo.com/attachments/20100710/fb268683/attachment-0001.gif>

More information about the time-nuts mailing list