>
> Date:    Mon, 17 May 1999 07:45:16 -0500
> From:    "Kurr, Martin" <[log in to unmask]>
>
> Resistance is proportional to temperature, so measuring resistance over a
> circuit (through a diode, for example) can be mapped to temperature.

Sorry, things are not that simple.  Referring to SAMS Publishing's "Reference
Data for Engineers:", Eq. 21, page 18-10:
                                             qV/kT
            I = qA[(D /L )p  + (D /L )n   ](e      - 1)
                     p  p  n     n  n  p

Note that the current I is approximately exponential to the voltage V and
to 1/T.

If I've solved correctly for T (absolute temperature in degrees K or R)

                                       qV
             T = -------------------------------------------
                                        I
                   kln(--------------------------------  + 1)
                           qA[(D /L )p  + (D /L )n ]
                                p  p  n     n  n  p

Lumping the constants into 2 K's

                        Ka
            T = --------------------- * V
                  ln( I/Kb   +  1 )

This implies that you need a good, steady, voltage regulator to get a
decent temperature reading.  Alternatively, you could use a constant-
current diode as a source of constant current, and measure the voltage
across the diode in the  CPU.

> Whatever you'd use to measure resistance would have to be calibratable (use
> digital temp. probe on die surface, and insulate entire CPU surface to
> simulate interior of die temps.).

As long as you do not have to use the CPU while you are calibrating it, just
connect to the diode you will be using, and put it in a covered thermos with
some hot or cold oil.  Measure the temperature of the oil with an accurate
thermometer.  Stir gently with a wire bent so that it avoids bumping
anything.  Use a small or intermittent current to reduce self-heating.
(Like a transistor in thermal runaway.)  After a few minutes, the readings
should stabilize, with a very slight trend towards room temperature.

>                                   In addition to a zero point you must take
> temps over the entire range you'll measure, and correlate to the resistance
> readings (Excel can help with this).

Rather than plot Temperature versus Resistance (V/I), or even I, holding
V constant; plot Temperature versus Voltage, holding I constant.  You should
get a good "Least Squares" fit.  I would expect that CPUs from the same
foundry, or at least the same production run, should have "similar"
calibration curves.

>                                      The key problem here is hooking your
> resistance meter up to the unused CPU element you will use.  Of
course, you > must identify an unused CPU element first...

Agreed.  On many small-scale integrated circuits, there are "protection"
diodes.  Also, the substrate may form a diode with another lead...

SUGGESTIONS WANTED on where to find this information on various CPUs.

> This is not really a worthwhile endeavor IMHO.  Better I think would be
> several temp probes (one on each side of CPU?) and use average, then
> correlate to actual CPU temp by measuring when CPU is fully insulated.
> Martin Kurr
> email [log in to unmask]
> > ----------

Suppose the CPU fan falls off the CPU, or the heat sink develops a "fur coat"
of dust and lint.  The fan will still be spinning, but it won't cool the CPU
properly.  Temperatures measured around the CPU and heat sink may appear
normal (The same amount of heat - watts or calories - is still being
removed.), but the chip is frying.  When the die gets too hot, connections
fail, or diffusion hastens the demise of the CPU.

If you can measure the temperature right at the source, then you can
monitor it to ensure that everything keeps working properly.

> > From:         Kenneth Alan Boyd Ramsay[SMTP:[log in to unmask]]
> > I understand that certain semiconductor diodes can be used as a sensitive
> > thermometer (although they may be tricky to calibrate).  If you could use
> > two legs of an unused transistor, or some other diode on the CPU die, you
> > should (at least in theory) be able to follow the actual die temperature.
>
> ------------------------------
> On 14 May 99, at 17:13, Max Timchenko wrote:
>
> to the CPU to be adjusted.
>
> right on the CPU die, which is carried out through the CPU slot to optionally
> connect to a monitoring chip which some motherboards provide.  Look for a
> program called "MBMxx.exe" ("MotherBoard Monitor") which can read data from
> this chip; I believe it has a web page that lists many supported boards, and
> yours might be on the list.

Boyd Ramsay

[log in to unmask]

               PCBUILD mailing list is brought to you by:
                            The NOSPIN Group
                            http://nospin.com