AW: [Re: thermal tips re expo converters:]

[Ian Fritz]

Terry --

Thanks for the thoughtful comments. I'm afraid I don't understand your
examples well enough to comment in detail, but here are a couple of
thoughts. 

I assume you are saying that the HP xtal osc is in an oven?  It would be
interesting to see exactly how they do this.  Remember, though, that
this is not a critical temperature control application(!) These units
use a specially oriented quartz crystal whose oscillation frequency has
a zero temperature coefficient. (Developed by a genius working down the
hall from me -- I saw it happen. He now has his own company making xtals
for frequency control.) So the oven just needs to hold the xtal near the
compensation temperature. Also, how long does it take the unit to
stabilize? And, what is the relative thermal coupling between the heater
and the xtal vs. the coupling between the heater and the sensor?  For a
typical oven, of course, the heater and sensor are tightly coupled, and
any small oscillations in the control loop tend to be smoothed by the
lag to the object being heated.  This is not the case for the chip
heater.  You can actually hear the servo-loop noise feeding to the audio
output in the SSM 2033 VCO if you tune the output frequency to above the
audible range.

I agree -- and Juergens simulations seem to confirm this -- that you can
indeed obtain stable control with a well-insulated system using a low
power input.  But I still think that to avoid oscillation it has to be
very slow. Again I would ask you: if rapid cooling is not required, why
does everybody run their chips at 60-80 deg C? This high temperature has
nothing but disadvantages, as far as I can see, except for ensuring
rapid cooling.

The point about thermal gradients across the chip is certainly something
to watch out for. I tend to ignore this potential problem because the
thermal conductivity of Si is so high. 

  Ian


terry michaels wrote:
> 
> I'm not sure I fully agree with you on this.  I don't believe you need a
> high cooling rate to achieve loop stability, it can be done with the proper
> choice of loop compensation.  I haven't run the math for this, but I can
> think of two real world examples of how this can work.
> 
> Hewlett Packard makes high stability crystal oscillators for test equipment
> where the oscillator is packed in a metal container, and is VERY well
> insulated.  They use long, thin wires to connect the oscillator to the pins
> on the container, to cut heat loss.  As far as I know, it's stable.
> 
> An electrical analogy to this is a high current regulated power supply, say
> using a 723 regulator chip driving some 2N3055 pass transistors.  It will
> typically have a small cap on the output to improve transient response.
> This capacitor holds electrical charge, same as an insulated expo convertor
> holds heat.  If you turn on the power supply with no load attached the
> output cap comes up to the output voltage extremely fast, because such a
> power supply can deliver amps of current.  The curent drain is almost
> nothing, maybe milliamps or less.  That's equivilant to high thermal
> insulation.  The output voltage is stable, I've never seen a power supply
> break into oscillation when no load is attached.
> 
> I might be completely wrong on this, but I think a heated expo convertor
> should be designed with a high current supply so it heats quickly, and high
> thermal isolation so it is less affected by ambient temperature changes,
> uses less current to maintain temperature, there is less error between the
> set point temperature and actual chip temperature, and the entire chip is
> at the same temperature, with no gradients across the chip.
> 
> BTW, a possible drawback to using fat heavy traces to create a thermal
> drain is you might end up with a thermal gradient across the chip.  This
> gradient will vary with the difference between ambient temp and chip temp.
> If the chip geometry is such that this gradient causes a temperature
> difference between the expo transistor and the transistor that generates
> the Vbe offset, you will have poor VCO stability.
> 
> Terry Michaels