[Server-sky] Server sat thermal structure

[Keith Lofstrom]

(shared with the mailing list)
On Wed, Mar 25, 2009 at 07:04:20PM -0700, Howard Davidson wrote:
> I need some sketches of how you plan to put together on of the 
> satellites. I was imagining the solar cells as the structure.

There is a very crude drawing on the website, see
   http://www.server-sky.com/ServerSatV01
I have not made it more accurate because of too many unknowns
(which I want to fix ASAP, but I've got taxes and business stuff
to do ASAP-er ).  Damn clock has only 24 hours on it.  You are
helping me with the unknowns.

> This can get tricky as the temperature will not be uniform
> across the system.

The cell is the main structure.  If it is 100 microns thick, and
absorbing all 1300W/m2, and radiating half the heat from the
front and half to the back, then 650W/m2 passes through it.  The
thermal conductivity of Si is 149 W/m-K, so the temperature difference
front to back is 650*1e-4/149 K or about 0.4 millikelvins, which will
cause a small amount of curl but not much.  

> A very crude approximation is that the heat spreads about as far 
> sideways as it goes down.

That is true if the second heatsink it is spreading into is good. 
The second heat sink (assuming a high albedo reflector on the sun
side) is black body radiation to a partial hemispherical sky (75%
to 100% depending on position relative to the 300K earth) at 4K. 
So in this case, I expect my "sideways and down" is dominated by
the vacuum/blackbody heatsink, not the silicon.  Assume a 350K
1 watt die that is 2mm X 2mm, backed with a 20um copper plate
heat spreader, and a total radiating area of 50cm2.  My crude
calculations suggest the edge of the spreader will be around 280K.
I need to do better calculations.

> Cooling the electronics may be a very serious challenge. Just to start 
> with you have to dump about 80% of the incident sunlight.

Out in the electronics ring I expect to have reflective aluminization
between electronics and the sunlight, so there will be less heat to
dump.  I expect I the serversat be using many 1W processors rather
than one large multiprocessor.  The processors will be power constrained
rather than clock rate constrained.  I will probably adjust clock speed,
and bring processors on or off line, to regulate the temperature of the
operating processors.

> I don't know if you can get away without cover glass on the solar cells. 
> One function is to reflect light that is too short wavelength to help 
> make electricity, the other is to protect the junction from ionizing 
> particles.

I can add some factor to the radiation lifetime of the cell by putting a
thin cover on it; but not orders of magnitude.  The real calculation is
how much end-of-life power-per-mass I can get;  if the shielding adds
more mass than it adds lifetime, it is not a good tradeoff.  I can do
limited communication and orientation maintenance with a small fraction
of the beginning cell power, and that is all I need to move the used-up
server-sats to the graveyard.

Most importantly, whatever gets used as "cover glass" had better be 
damned close to 2.6um/m CTE, or I get a heck of a lot of curl.  In actual
fact, I probably want a thin layer of glass ( 0.5um/m ) and a layer of
ITO ( 10 um/m ? ) with relative thicknesses adjusted (along with the 
aluminum metalization at 23um/m ) to approximate 2.6um/m over larger
distances.  I want to learn more about how ductility and work hardening
will affect the metalization over thousands of temperature cycles, though.
I would rather not use gold - if I did, I would need to add something
like Ti/W barrier metal to keep gold out of the silicon.  But I will if
the ductility is important, and you alluded to that.

I have been thinking a bit about using shaped discontinuous islands of
glass to lens the light away from the topside metallization, and through
the ITO into the silicon.  Chances are, that is too expensive.

> Thin antennas will be interesting. It would be good it they can hang off 
> the edges of the system. If they are close to the silicon you will have 
> large losses from moving carriers around.

Indeed, unlike the drawing I expect the antennas to be complex metallization
patterns on something like an alumina/glass substrate ring around the edge.
I probably want something I can silkscreen and fire like a technical ceramic,
because the processors and other chips will be beam-lead bonded into holes
in that substrate ring.  Or I might start with sheets and cut holes with
lasers.  I don't think the ring will be made of one big piece.  The big
trick will be making it exactly the same height and density and modulus
as the solar cell stack, so that launch vibrations will not stress the
connections to the solar cell and the backside wiring too much.

I will be using pretty low impedance interconnects on that substrate
ring.  I will use stub resonators for impedance matching to the antennas.

> It may be hard to get decent frequency stability with thin resonators. A 
> stub does not have very high Q. I am not sure how thin you can make a 
> SAW device. I think they may make be very sensitive strain transducers. 
> The beat note between two adjacent ones might be a way of compensating.

Overall, I expect to do array-wide frequency control by averaging, then
use GPS-like signals from the ground and GPS itself to calibrate the
array.  I will be feeding the quadrature modulators with an IF frequency
that I can adjust to compensate for local frequency drift.  Better local
calibration means I can save power on a lower frequency I.F. and do the
adjustments less often.

Keith


-- 
Keith Lofstrom          keithl at keithl.com         Voice (503)-520-1993
KLIC --- Keith Lofstrom Integrated Circuits --- "Your Ideas in Silicon"
Design Contracting in Bipolar and CMOS - Analog, Digital, and Scan ICs