1n4148 as temperature sensor

[Martin Czech]

I took one of my beloved 1n4148 into the thermo stream today, glued
to a linear temp NiCr sensor with heat conducting paste.  The diode was
plugged into my 5 digit multimeter Metex M4640A in diode measurement mode.
It was crucial for the measurement to have very good thermal contact
between sensor and diode, otherwise the data was rubish.

This is the result:

T(C) U(mV)
-40 663.5
-30 645.6
-20 627.4
-10 608.8
0   589.9
10  570.4
20  550.9
30  531.4
40  511.9
50  492.8
60  473.7
70  455.0
80  436.5
90  417.9
100 399.3
110 380.6
120 361.9

The diode/multimeter combination is pretty linear, arround 25C to 70C
I can get away with 0.5C linearity error. Since I have values from -40C
to 120C in 10C steps I can interpolate much better, the error should be
0.2C something. The linearity error at both ends (-40C and 120C) is about
1%, so a diode may not be good enough for such a large range, if you need
linear behaviour for a simple display.

I also checked for repeatability and cycled arround from -40C to 120C etc,
no deviation could be detected after settling time.

This way (by compensating the nonlinearity with pocket calculator or with
a table) I should be able to measure the diode temperature in -40C to
120C with an error of about 0.1C. That's ok.

m.c.