PIC10F322 - Internal Temperature Indicator

[viki2000]

Memory free for temperature? 10%, worst case 15%.
Unfortunately on production site there is no chance to calibrate using external communication with a PC or another device. There is no option to write or read the firmware. The PICs will be used as they come with my programmed code done by Microchip. The end product is a tiny board which is tested at the end only for functionality.
Microchip said they do not calibrate for me prior programming. 50K is too small quantity.
The temperature detection is not “a must”, but a “good to have” feature.

I asked Microchip about +/-19% FVR tolerance.
The answer:

[temtronic]

hmmm given say 50 words of code space I don't think it is possible to use the temp sensor. If say 10 words are used for the 'real' temp. function( read the sensor, if (cold,Ok,hot), flash LED ?), that only leaves 40 words for the 'auto calibration' routine.
As you've already invested a LOT of time into this, you might as well see if you can cut the 'auto calibrate' code as Mr. T. suggests. Show us that and perhaps we can reduce or modify it to fit. Obviously only use unsigned integers, NO division, etc. As you only need a 'rough' result (too cold, too hot) the math would be small and simple ie: ... if adc_result<90 flag=too_cold, if adc_result>192 flag=too_hot).

I'd cut the 'real' temp function first to see how much code is needed. Don't worry about the 'calibration' function, just the read adc, cold/ok/hot ?, flash LED and see what code space that takes up. Hopefully not much, less is better giving you more for the 2nd part, the 'auto calibrate' function.

Bottom line, you'll need a LOT of clever 'tricks' to do it.

Jay

[viki2000]

Me:

[viki2000]

Since almost 2 weeks I have on my desk the setup with PIC10F322 communicating with PIC18F4550 attached to the small display showing ADCti, ADCfvr and temperature without calibration. The VDD is 4.9xxx not very stable for the second and/or third decimal point.
The ADCfvr is jumping up/down 1 digit 52/53.
The ADCfti is jumping up/down 1 digit 139/140.
The calculated temperature jumps between 27-39°C.
By calculation, and also practical, I noticed a jump of 1 digit error of ADCfvr leads to 8°C error and 1 digit error of ADCti leads to 4°C error.
It does not matter the value or if it is calibrated, important is the range of imprecision.

If we trust the datasheet and consider the FVR tolerance +/-19%, then we are in trouble when the VDD is not constant and we want to determine the VDD value. The error is too big to be usable.
It is enough to take 1 example of VDD=5V and supposing we do not know that is 5V, so we want to find how much is it by measuring it using FVR as described in AN1072
http://ww1.microchip.com/downloads/en/AppNotes/01072A.pdf
ADCfvr=1.024*((2^n)-1)/VDD=1.024*255/5=52.224, so we can say ADCfvr must show 52 or 53 as my system shows.
With FVR +/-19%, we get instead of 1.024 two values 0.82944 and 1.21856.
Then the ADCfvr may be ADCfvr1=0.82944 *255/5=42,30 or ADCfvr2=1.21856*255/5=62.14, so we can take ADfvr1=42 up to ADCfvr2=62.
If we calculate back the VDD, supposing you do not know it is 5V, then you get VDD1=1.024*255/42=6.21V and VDD2=1.024*255/62=4.21V
With this big error range of 4.21-6.21V finding the VDD when is not stable or known we arrive to such big errors in measuring the temperature, only due to FVR tolerance, that makes it totally unusable.
For stabilized VDD=5V , mode 4 and 27°C we should get for ADCti the value 139 – calculated.
Assuming the ADCti is stable 139 with zero tolerance, because the VDD is in reality 5VDC even if was measured with error 4.2-6.2V, then for the calculated temperature we get for VDD1 and VDD2 between -76°C up to +96°C. Just use ADCti=139 and ADCfvr1=42 and ADCfvr=62 in the formula from previous page assuming we have VDD=5V and 27°C in reality. That is 172°C range.
That leads clear to the fact that even with calibration we cannot use temperature indicator on PIC10F322 with unknown VDD and AN1333 makes no sense at all for this PIC as long as FVR tolerance is +/-19%.
The temperature indicator inside PIC10F322 must be used at known VDD without FVR.

[temtronic]

You've put a lot of time and effort into your project and reporting your observations here should help others doing similar work! As you've found out an unknown or unstable VDD makes using the internal temp. sensor, well, useless. It requires a very stable VDD and that means a 'strong' power supply.One capable for 10X the normal current draw using a precision Vref as the regulator. Normal '7805' types are at best +-5%, NOT suitable for the task.As well a batch of 10 units will not be even close to the same VDD!
I do applaud your effort of this !!

Jay

[joergn]

I am considering to use the 16F1765 internal TEMP_INDICATOR also and find this discussion quite late...

The main function I do with my pic project is to measure the LED junction temp with my new patented principle - so I have done lots of work related to measure the temp with power diodes, quite precisely. On my main project I resolve into the mV range doing some more samples summing into a 16bit var.

To obtain the final resolution of temp reading and probably also accuracy with this internal PIC diode, a single ADC conversions is simply not fine enough, you may need some statistics, doing an average of 16 or 32 samples.

There are also some tricks possible to increase the ADC resolution by giving statistic noise on the ADC reference. If there is no such noise in the system present the resolution can not be increased by the averaging.

I also avoid float math and if possible I devide by power of 2 : >> ;-)

VDD I regard as to be not that important, since it must be simply high enough, like it is written inside the datasheet. More important is the internal current regulator of the temp indicator itself, which I can not find any tolerance on it. This is what determines the quality ;-) So you measure differential from VDD just over the diodes, if VDD is put as Vref+ to the ADC. This is the best possible ADC setup, because you do not depend on VDD

[viki2000]

At that time, I opened a similar discussion on Microchip Forum only to get more input from different experts.
http://www.microchip.com/forums/m881560.aspx
I tried the code with CCS and XC8 compilers.
As conclusion from different sides, also based on my measurements: don’t lose your time with the internal temperature indicator – it is not accurate, don’t rely on it unless you have no other option, but then you must calibrate each chip and even then you have tolerances for measurement. So use it if you can live with those tolerances.
If you find any other method to improve the temperature measurement using the internal temperature indicator, then it would be nice if you can share your findings.

[temtronic]

I've never understood why anyone would put an internal temp indicator inside a micro. From a design view, it can only measure the internal temp which will widely vary depending upon processor speed and what it's doing. IE : more I/O or operations will raise the temp. It's NOT designed or useful as a 'local room air' temperature indicator or similar process controller.
I think some big mfr wanted it in THEIR product and Microchip thought it might be a nice peripheral to offer to the general public.
If anyone has actually used it, please post 'how and why'. tnx.

Jay

[joergn]

[temtronic]

Thanks for the explanation ! The 'cheap like auto' ALSO means not precise or accurate so you NEED to calibrate every unit or accept a WIDE range as being 'valid'.
I can also see that the physical layout of the PCB and it's orientation in the real world will affect the int. temp sensor. Somethings NOT usually considered until rev 3 or 4 and months have gone by....BTDT !
I'm 'old school' and always have a free pin or two. One for a diagnostic LED / serial to PC / other use. Also in the 'old school' department is accurate, repeatable measurements. It's often cheaper to add a component (like an LM34 type device in this case) than spend days doing R&D to get the internal temp sensor working satisfactorly. Having to adjust code, resample a sensor 16 times or more, to be sure the reading is OK really does cost $$.

Jay

[joergn]

[temtronic]

With R&D costing $100/hr I can easily justify spending 20 cents for an MCP9700 that is easy to code, known to work and far more accurate 'out of the box' versus the hours (or days) to get that internal temp sensor working. From what I've read on the Web, a LOT of time has been wasted getting marginal performance out of the internal sensor. It's one of those 'sounds good' but when you go to use it, it just isn't worth the time and effort.
As a 'school project' or something to do on a rainy day, yes, I can see experimenting with it but I just can't see using it, especially if the software to get it to run 'nice' is big, too big for a small PIC!
It reminds me of the TPMS in cars, nice technology but not required.
Jay

[joergn]

[temtronic]

I don't use the 16F1765 but am curous if there's a 'bug in the die' that causes the bad readings. Is it only the internal temp sensor or seen on other ADC inputs or perhaps the method(sequence of code) used to acquire the reading ?
I also read any analog 16 times and average the results,though I've never seen really 'bad' ADC results.

Jay

[PCM programmer]