Usually we calculate junction temperature from collector dissipation (easy to measure) and Rth from datasheets. Once we have working prototype we can measure case temperature and infer the junction temperature.
Can we instead calculate Tj from difference in Vbe measured immediately after a cold start (when Tj is supposedly equal to the ambient temperature) and again when the thermal equilibrium is achieved?
This is the formula:
Tj = Tamb + (Vbe1 – Vbe2) / tc
Tj - operating junction temperature
Tamb – ambient temperature
Vbe1 – base emitter voltage measured immediately after a cold start
Vbe2 – base emitter voltage when operating temperature is reached
tc – temperature coefficient, ~ 2mV/C for silicon BJT
I suppose measuring the Vbe1 is the most challenging part. Allowed time to make the first measurement depends on the physical size of the device. For small devices thermal time constant might be too low for reliable measurement with common multimeter.
BTW the same method could be considered also for MOSFET devices.
Your comments are welcome.
Can we instead calculate Tj from difference in Vbe measured immediately after a cold start (when Tj is supposedly equal to the ambient temperature) and again when the thermal equilibrium is achieved?
This is the formula:
Tj = Tamb + (Vbe1 – Vbe2) / tc
Tj - operating junction temperature
Tamb – ambient temperature
Vbe1 – base emitter voltage measured immediately after a cold start
Vbe2 – base emitter voltage when operating temperature is reached
tc – temperature coefficient, ~ 2mV/C for silicon BJT
I suppose measuring the Vbe1 is the most challenging part. Allowed time to make the first measurement depends on the physical size of the device. For small devices thermal time constant might be too low for reliable measurement with common multimeter.
BTW the same method could be considered also for MOSFET devices.
Your comments are welcome.
Last edited:
i have found an ir thermometer to be invaluable when designing amps.
sometimes the circuit is too complex to accurately work out power dissipation.
upto 50 degrees c is good, much more then a heatsink is required
sometimes the circuit is too complex to accurately work out power dissipation.
upto 50 degrees c is good, much more then a heatsink is required
That,s a good practical method and chip_mk your comment on this manufacturer ?
[url]www.semikron.com/dl/service-support/downloads/download/semikron-application-note-calculating-junction-temperature-using-a-module-temperature-sensor-en-2020-01-29-rev-02/ [/URL]
[url]www.semikron.com/dl/service-support/downloads/download/semikron-application-note-calculating-junction-temperature-using-a-module-temperature-sensor-en-2020-01-29-rev-02/ [/URL]
Obviously. I forgot to mention that in OP. Still some variation in quiescent current will not make big difference. For example +-10% variation will result in about +- 1C error (but rises exponentially).
Never used IR thermometer. Does it measure only the case temperature (not the junction)?
It's a quite sophisticated technology. I guess more suitable for hard core professionals.
Where I worked, we used the Mosfet Bulk body diode to test for voids under the bond between the chip and the die header. If there was a void, the diode would read lower than it should, ie it was hotter than spec. That diode was also used to test a few other parameters during testing ( these were devices for the auto mkt - high rel etc).
You may find This interesting Some Ideas on Temperature Compensation for Audio Amplifier EF Triples
You may find This interesting Some Ideas on Temperature Compensation for Audio Amplifier EF Triples
Last edited:
Not only that, does the base current have to be constant, or the emitter current? Should the Vcb also be constant to avoid Early effect and saturation issues? It seems clear than using diodes for temperature sensing is less complicated with only the one junction.
If the current is large, the emitter resistance will act as another temperature
varying (and device varying) component, complicating the response curve. Unless you can separately calibrate the response it all sounds pretty rough and ready.
Well, the gain is temperature dependent, so a constant base current is undesirable. That makes the collector current the one to keep constant (or emitter current in a high gain device). And yes, the Early effect can change the gain, so yes the collector voltage should also be kept constant.
But otherwise, measuring Vbe as a means of temperature of the device, it is perhaps the most accurate method. So investigating drift between switching on and warming up where the currents might be or are stable would be possible.
Also, thermal impedances can be small so you may also need a fast sampling voltmeter or scope to measure instantaneous potentials.
But otherwise, measuring Vbe as a means of temperature of the device, it is perhaps the most accurate method. So investigating drift between switching on and warming up where the currents might be or are stable would be possible.
Also, thermal impedances can be small so you may also need a fast sampling voltmeter or scope to measure instantaneous potentials.
Ic / Icob = (e)exp( Vbe / {kT/q} )
Only k and q are absolute constants, the rest are variables.
So, to measure T and have Ic or Vbe as indicator, the rest should be kept at a constant value too. Notice the exponential element, that is an easy runoff.
As suggested in above posts, it is a challange.
Only k and q are absolute constants, the rest are variables.
So, to measure T and have Ic or Vbe as indicator, the rest should be kept at a constant value too. Notice the exponential element, that is an easy runoff.
As suggested in above posts, it is a challange.
The temperature dependence of Icob dominates. That's why VBE drops with temperature at a given collector current. When you extrapolate the VBE versus temperature graph to absolute zero temperature, you get a value around 1.23 V, which is the bandgap voltage of silicon plus some correction term - but that's only true at moderate current densities, where high injection and emitter bulk resistances don't mess things up.
To summarise, it's a "yes, but..."
Actually, I've made a very good thermometer from a diode-connected BC547 and run from a constant current. But I changed to an old BC107 in a metal can as it had a faster response than the plastic one.
Actually, I've made a very good thermometer from a diode-connected BC547 and run from a constant current. But I changed to an old BC107 in a metal can as it had a faster response than the plastic one.
I think the correct statement about Vbe, assuming the same part number, same current and same temperature , is that
1. Vbe will vary from transistor to transistor by up to a a few (10's) of mV about the nominal 600mV at room temp
2. The dVbe will vary by a few hundred uV/deg from unit to unit about the nominal 2.2mV
3. For each transistor, the repeatability of dVbe with temperature cycles will be very accurate and to within fractions of a %
Once you calibrate out the initial Vbe and the dVbe slope, you can indeed make a fine thermometer. When I was involved in industrial instrumentation (decades ago) we used a BC237 as a CJC compensator in T/C amplifiers. The calibration process dialled out the initial Vbe and the slope and after that instrument accuracy and repeatability were excellent. The alternative was to use a temperature IC at about 20x the cost.
For audio, you dial out the differences in output transistor Vbe when you adjust the Iq. If you fixed the bias current and made it non-adjustable, there would be a large spread in Iq.
(You can try these things out BTW in LTspice)
1. Vbe will vary from transistor to transistor by up to a a few (10's) of mV about the nominal 600mV at room temp
2. The dVbe will vary by a few hundred uV/deg from unit to unit about the nominal 2.2mV
3. For each transistor, the repeatability of dVbe with temperature cycles will be very accurate and to within fractions of a %
Once you calibrate out the initial Vbe and the dVbe slope, you can indeed make a fine thermometer. When I was involved in industrial instrumentation (decades ago) we used a BC237 as a CJC compensator in T/C amplifiers. The calibration process dialled out the initial Vbe and the slope and after that instrument accuracy and repeatability were excellent. The alternative was to use a temperature IC at about 20x the cost.
For audio, you dial out the differences in output transistor Vbe when you adjust the Iq. If you fixed the bias current and made it non-adjustable, there would be a large spread in Iq.
(You can try these things out BTW in LTspice)
Last edited:
Yes quite right. My circuit had those adjusters for the absolute Vbe and differential. And, as you say, it required re-cal on changing devices.
Interesting. That implies low dispersion of parameters within batch, so the difference due to quality of bond (glue?) is dominant.
Basic idea is to use this method to check junction temperature in-situ. Precision is unavoidably somewhat compromised due to small fluctuations of Iq and Vce. But it gives some insight of the chip temperature (e.g. is there 70C or 95C).
We can measure case temperature and infer junction temperature from the thermal resistance junction-to-case. This is another, more direct source if information. I just wonder how reliable it is.
It is perhaps the best guide to a junction temperature. But it needs to be measured under more or less static conditions. A bipolar transistor base voltage will depend on whether there is current being shunted in or out in a dynamic environment which will upset the readings. If the voltages and currents are stable, then it is excellent. But it will also need to be calibrated for a given device as Bonsai mentioned because of manufacturing variations if you want the accurate temperature (as I did in a thermometer).
It would be possible to use Vbe as a guide to an "instantaneous" temperature but after high speed transients had stabilised - so could be fast but not perhaps below microseconds, depending on how the device was being operated. And whether you had a high speed scope.
It would be possible to use Vbe as a guide to an "instantaneous" temperature but after high speed transients had stabilised - so could be fast but not perhaps below microseconds, depending on how the device was being operated. And whether you had a high speed scope.
FWIW my homemade desktop heatsink temperature meter is a ground 1N4002 so 1 side is flat (body now resembles a D) and epoxied to a 1 x 1 cm , 1mm thick piece of aluminum.
I feed constant 5mA through it a nd measure voltage drop.
I calibrated it by reading V drop when submerged in a glass of water and floating ice (actually stirring it with diode 😉 ) which guarantees 0 deg C and submerged in boiling water which guarantees 100C.
Draw both points on a graded paper sheet, join them by a straight line and staple it to the wall.
Works like a charm.
I press or clamp sensor to heatsink, add some thermal grease if you wish but not really needed, sensor mass is very low.
I feed constant 5mA through it a nd measure voltage drop.
I calibrated it by reading V drop when submerged in a glass of water and floating ice (actually stirring it with diode 😉 ) which guarantees 0 deg C and submerged in boiling water which guarantees 100C.
Draw both points on a graded paper sheet, join them by a straight line and staple it to the wall.
Works like a charm.
I press or clamp sensor to heatsink, add some thermal grease if you wish but not really needed, sensor mass is very low.