[time-nuts] questions on uncompensated crystal
drfrank.stellmach at freenet.de
drfrank.stellmach at freenet.de
Thu Jul 6 05:07:51 EDT 2006
Hi Stan,
Your question is really quite interesting, as high precision by high
cost is easy to achieve, but the real art is to do it by low cost. In
this sense, I would like to give you some really practical hints.
The company's division I'm working for, produces automotive
electronics, i.e. instrument clusters with a "precise" clock, i.e. < 1
min /month deviation over a temperature range of -40 to +85°C, giving
+/-50ppm overall. We have production volumes of 100k-300k/yr. per
application, or several 10Million applications for all of our
customers. High and longtermed reliability / quality in production and
field and ridiculously low costs are mandatory. We cannot use external
oscillator circuitry, hardware trimming, TCXO or temperature
compensating methods for those reasons.
We use embedded controllers; the oscillator consists of an internal
CMOS inverter, 2 ceramic caps 5-10% and an AT-cut, fundamental crystal
with clock trimming by software correction. SMD components are
mandatory, also for cost reason, so I recommend an AT51-GW (gullwing)
type crystal, which is an AT51 / CS10 / HC49 (supplier dependant) case
with bended leads and an additional metal clamp for mounting. Other
cases with plastic plate underneath are more expensive.
Price for the crystal should be less than 20 Cents, depending on
further requirements.
I highly recommend NDK ( [1]www.ndk.com), they produce the quartz
blanks on their own, and also design nice OCXO's , so they have a lot
of technological experience.
Vectron (formerly Corning, formerly Telequarz) or Kinseki as 2nd
sources, also have automotive experience.
Ask for tighter automotive specs, but lower price, as ours depend also
on our automotive quality requirements (e.g. AEC Q200 and more plus
tough warranty regulations).
27MHz can be realized by fundamental crystal, but it's on edge of
technology because the crystal is pretty thin already. Therefore, a
tighter specification is critical and costly, so use PLL design with
9MHz or less instead!
We often have discussions between the (digitally minded) designers of
the inverter and the crystal supplier: For us, an additional critical
aspect for series production are oscillation start margin (which is
very crudely checked by our hardware designers) and ESR dips over
temperature. Crystal quality (e.g. residual silver particules from
electrodes) vs. input/output impedance etc. of inverter are to be
observed.
Due to cost and practical reasons, we also do not deal around with
simulations or excessive measurements. I also confirm
, that SPICE simulations etc. are not applicable here.
Instead, we use cost free service/knowledge of suppliers. We send a
sample board to NDK, they measure the circuitry and propose values of
C1, C2 and the nominal center frequency of the crystal they will
deliver. Perhaps they know your type of inverter already, as they have
a huge database. Takes about 4-6 weeks.
It's not possible to choose other parameters freely anyhow, in
contradiction what you assumed, as nominal frequency, case / crystal
blanket size and oscillator determine the overall parameters!
Typical specification for a 16MHz crystal over automotive temp. range
of -40..+85...+125°C are:
+/- 20...50ppm freq. deviation at room temperature, 50µW
additionally 20..50ppm over temperature (20 ppm will double the price
compared to 50ppm)
50µW-1mW max. power
5ppm/yr. aging
ESR and Dips < 80 Ohm over temp.
I think that you have to do some qualifications of your product on
your own to safeguard your requirements.
Your initial precision and also the long term aging requirements are
hardly to be achieved by design or specification only.
On the other hand, it depends on your or your purchasings' negotiation
skills, what your supplier is willing to specify, as you for sure need
tighter specs.
Best regards
Dr. Frank Stellmach, Germany
References
1. http://www.ndk.com/
More information about the time-nuts
mailing list