AW: [En-Nut-Discussion] PT100 Sensor on ethernut

James A van Zee javanzee at
Mon Sep 19 02:23:17 CEST 2005


   So many solutions!  I just recently finished a design needing some internal and external temperature sensors, so I will share some of my thinking with you.  In the end I decided to use the TMP35 sensor from Analog Devices (Digikey carries them, they're about $1.25 if I remember correctly).

   The TMP35 comes in a TO92 pkg, which makes a fairly convenient 'probe' if you connect it to some small shielded 'microphone' cable (2 wires+shield, pretty easy to find) and use heat-shrink tubing and some water-proof adhesive to seal it.  The TO92 shape isn't necessarily the best, but I suppose you could grind it to a more cylindrical shape if you needed to.  (I haven't done that, so it's just a 'wild' suggestion.)

   But there's another reason I chose this device: it's the same as the LM35 from National Semiconductor (also available from Digikey, but more expensive than the TMP35).  However, the LM35 comes in (yes!) a TO220 power transistor package, which means you can screw one down onto a surface if that's what you need to do!

   The sensitivity of the LM/TMP35 is 10mV/degC, with 0,0C = 0,0V.  So that's a problem if you really MUST measure down to -10degC.  The solution to this problem is to use the TMP36 (Analog Devices/Digikey: only available in a TO92 pkg), which has a 400mV offset (but the same sensitivity), so you can go from -40degC to +150.  There's one more option: the LM34 from National/Digikey is a Fahrenheit sensor: 10mV/degF, which means it has a 32deg = 320mV offset, and so can measure down to -10degC.

   These devices run on 2,7-5V supplies (not critical) and (obviously) have a 1,20V (TMP35) or 1,60V (TMP36) output at 120degC.  You -could- use any common opamp to give you a 2x or 3x gain, but if you use the ATMega128's internal 2.56V reference, each A/D bit is 4,00mV = 0,40degC, which meets your requirements without doing anything but wiring them up!  If you use the LM34 you get an automatic gain of 1,8 due to the difference between the Fahrenheit and Celcius scale, hence your effective resolution with just a 'straight hookup' is 0,22degC.  Assuming you could adapt the TO92 pkg to your needs, the LM34 would appear to easily meet all your specifications (and they are also widely available in 'surplus stores' as well as Digikey).

   However... there are many other options: for a mere $71.75, Analog Devices will sell you an "AC2626", which is a very nicely packaged stainless steel probe with a 1,00uA/K output, which meets your range and accuracy requirements, but does require some analog circuitry to turn the current into a voltage.  I'll assume you don't want to spend that much per probe and leave it there.  More info from <> if you need it.

   Then there are the new breed of 'diode sensors', using the tempco of a Si junction as a thermometer.  Many companies are active in this area, but Maxim and National are the first places to check.  What I like about this approach is that you just buy a little 'processing chip', which is typically TWI (I2C/SMB) or SPI based, to which you connect 'remote diodes' that are typically 2N3904 diode-connected transistors, and viola' you have a bunch of thermometers!  National's LM95231 measures two external diodes, and, if I recall correctly there are devices that will measure 4 or 5 (or maybe even 6) diodes.  The circuitry in these chips is pretty sophisticated in order to produce "noise-free" results from signals that are in the microvolt range, but, hey, they did all the work, so you don't have to!  Obviously you can use a screw-down power transistor instead of a TO92-style if that's appropriate.  However a word of caution: all Si junctions are not created equal, i.e. they don't all have the same tempco; these sensor chips are typically designed primarily for a specific junction found in Intel processors, put there for just this purpose.  They typically also work for junctions in SMD parts, but don't necessarily have the specified accuracy with big old power transistors.  However, if you are willing to do the calibration yourself, it's an option.

   Finally we come to what you probably really have, which are the PT100 sensors.  I found an interesting ApNote from Intersil <> which discusses some of the vagaries of Pt RTDs, and provides a circuit that claims to reduce the inherent non-linearity (the 'curvature correction I mentioned earlier) by a factor of 100.  I wouldn't call it a 'simple circuit', but it's probably a 'realistic circuit'.  You could replace the two digipots (that's what Intersil was trying to sell!) with real pots and have a pretty decent thermometer with a net 10mV/degC characteristic, i.e. just the same as the TMP35, etc. I mentioned above.  Note, however, that you are going to need a much better amplifier than an LM324!  I think you could replace the 2,5V ref used in the Intersil design with the buffered reference output of the ATMega128, or with the less-expensive MCP1525 2,5V reference from Microchip.

    That's about all I have to offer.  Perhaps someone else has some better ideas!

                                                                  Jim van Zee
                                                                  Seattle, WA

----- Original Message -----
From: "Ole Reinhardt" <ole.reinhardt at>
To: "Ethernut User Chat (English)" <en-nut-discussion at>
Subject: Re: AW: [En-Nut-Discussion] PT100 Sensor on ethernut
Date: Sun, 18 Sep 2005 22:39:02 +0200

> Hello Jim,
> sorry, I missed to mention these detailes... I only want to measure a
> temperature between -10°C and +120°C with a resolution of 0.5°C is 
> sufficient. I also don't need any type of correction other than the one I can 
> do in software.
> I would appreciate a simple design using some simple operational amplifier 
> like lm324.
> I need to measure four temperatures at the same time and need long distances 
> between the sensor and the ethernut board (min. 3 meter).
> As an alternative the ds1822 seem a good idea, but I would need it in a 
> "closed" design. If it's not available I would need to construct something by 
> my self.
> Best regards,
> Ole Reinhardt
> -- kernel concepts    Tel: +49-271-771091-14
> Dreisbachstr. 24   Fax: +49-271-771091-19
> D-57250 Netphen    E+ : +49-177-7420433
> --
> _______________________________________________
> En-Nut-Discussion mailing list
> En-Nut-Discussion at

More information about the En-Nut-Discussion mailing list