PIC WDT Temperature accuracy

Jinx says:

Like I said I would, I've tried to quantify the WDT timer against temperature. The results I got surprised me, considering the simple tools I used for the experiment, a freezer and a pot of heated water.

I used 4 PICF84s (different batches) running at 10MHz. Apart from a4/mclr pullups, no external connections, and a well-filtered 5.00V PSU. All four are on the same board, each WDT pulse coming out of their respective b4 pins, to a 4-way rotary switch to select the PIC to measure. The wiper of the switch goes to another PIC which has a 100us IRQ and a 6-digit LED display. Alternate WDT pulses start/ stop the timer, times are shown here to 1/10000th sec. Temperature was measured using a fairly expensive multimeter thermocouple probe stated as being +/- 2% accurate. As the same equipment was used for all four PICs that should cancel out any errors and make the results comparable. There is no 0.1 digit on the TC, so I had to use my best judgement when to take the time reading. That was done when the display stopped fickering between the current and previous temperature reading

Below are the results (untouched, really), and as you can see, remarkable linearity for what I consider careful (but imperfect) measurement. The probe was inserted under a PIC with some heatsink compound, between the chip and the top of the socket. This held it securely and gave good contact to the PIC body.

I chose three stable temperatures as references. -15C, 20C and 82C and calculated the expected time differences on these temperatures to work out the deviations. One PIC was used to gather these figures, the other three came within 0.1% at spot readings.

As the TC probe is metal and the PIC a comparatively big thermal mass, the whole heating and cooling process was done as slowly as I could stomach.

Once a particular PIC is calibrated at two extreme temperatures I would see no reason why it shouldn't be very good sensor, bearing in mind its thermal mass. Putting the PIC on a heatsink would help the response time. Under more controlled conditions you could reasonably expect 100.00 all the way down these figures

82  2.9816   100.00
81  2.9682   99.99
80  2.9573   99.99
79  2.9450   99.99
78  2.9332   99.96
77  2.9158  100.13
76  2.9050  100.08
74  2.8769  100.19
72  2.8489  100.31
70  2.8201  100.45
68  2.8081   99.99
66  2.7818  100.05
64  2.7624   99.86
62  2.7382   99.84
60  2.7210   99.56
58  2.6830  100.04
56  2.6635   99.84
54  2.6436   99.65
52  2.6232   99.48
50  2.5922   99.71
48  2.5618   99.93
46  2.5346   99.98
44  2.5103  100.12
42  2.4942   99.65
40  2.4675   99.72
38  2.4395   99.85
36  2.4136   99.89
34  2.3882   99.70
32  2.3522  100.39
30  2.3324  100.18
28  2.3029  100.38
26  2.2767  100.45
24  2.2562  100.26
22  2.2356  100.08
20  2.2125  100.00
18  2.1902   99.89
16  2.1592  100.17
14  2.1432  100.24
12  2.1133  100.00
10  2.0846   99.82

5     2.0311   100.77

0      1.9601    99.86

-3     1.9284  100.04

-7     1.8809  100.12

-10   1.8581 100.34

-13    1.8067 100.13

-15    1.7782 100.00
----------------------------

See:

See