Mosfet Junction Temperature from Rds On Measurement

[Etesla]

Hi all,

I'm curious about something. I notice in many mosfet datasheets there is a curve relating the relative RDSon to a junction temperature like the one shown below.


When I measure the Rds on of a mosfet in application, I often see a waveform like the following:

Blue Vds in volts, red Vgs in volts, black Ids in amps, pink calculated Rds on in milliohms

Just from looking at this Rds on (pink), over the duration of a pulse, it starts pretty low around 1 mOhm, comes up to about 2 mOhms, then starts to come back down again. I am very confident that the probing in this example was adequate.

My question is this. Does the Rds on measured in that waveform actually reflect the junction temperature? If it does, that would imply that my junction temperature gets up to about 175C (2mOhm/1mOhm -> 2 -> about 175C) before tailing off in this example. If not, what are the timescales at which the Rdson -> junction temperature chart is applicable? What is the explanation for the varying Rds on with the gate voltage being fairly ideal?

[Kleinstein]

A reason why R_on is going up with temperature is that the electrons get less mobile at higher temperature. This is much like with normal pure metals like copper of iron, that get increasing resistgance with temperature. There are silicon based RTDs  (e.g. KTY82 series) that are mainly based on this.

Another small effect is from some effective additional voltage for the gate control. One gets something like a 2 mV/K shift of many voltage levels in semiconductors (not just the normal diode, but also varicaps and similor other parts). This can also contribute a little to the temperature dependence, especially in parts with a low gat voltage (e.g. 2 V operation). It would be less of an issue with 10 V at the gate.

The time scale this would be effective would be essentially instant, so down to the speed of the FETs in the 10s of ns.

[PartialDischarge]

What is the mosfet in question? What is the starting Vds?
Sorry but I doubt that your Vds measurement is correct, it is a difficult one, where offsets matter and you also have a negative undershoot in the current, and in the Vds, those don’t seem real to me.
Are you using a diode blocking approach to measure this?

[David Hess]

During characterization the junction temperature is measured using the body diode.

[Jay_Diddy_B]

Etesla and the group,

The graph that you show of RDSon versus Tj is normally measured using a pulsed technique. The gate source voltage of the MOSFET is fixed, in your case the Vgs is set to 10V.

A pulsed current, in the graph you show is 40A, is applied to the Drain source connections. The voltage across the Drain Source is measured with Kelvin connections. The current source is turned on for between 80us and 500us.
The duration of the current pulse is short to avoid self-heating. The duty-cycle of the current will be kept low.

The MOSFET is externally heated, on a hot plate or similar device and the RDSon can be measured as the temperature is changed. It assumed that Tj = Tcase.

Tektronix produced this video showing how to measure RDSon using a Keithley Sourcemeter (SMU).

Link:

Other instruments have similar capabilities, one early one was the Tektronix 176 pulsed fixtured in a 576 curve tracer.

The measurement that you performed is very difficult.
When the MOSFET is off there is significant voltage across the MOSFET, enough to overload the amplifier being used. In fact your oscilloscope waveform shows 'Channel overrange' in the top left corner.

You need perfectly compensated voltage probes when doing this.

You also need to be aware of the oscilloscopes overdrive recovery.

You could be looking at how the over-driven voltage amplifier recovers and / or the transient response of the scope probe.

There are some techniques using a diode, hinted at by MasterTech, to block the high voltage from the voltage channel.


Regards,
Jay_Diddy_B

[Jay_Diddy_B]

Hi group,

Here is a link to a Microsemi (APT) paper that illustrates how to measure Vce or Vds.


https://www.microsemi.com/document-portal/doc_download/14710-vds-on-vce-sat-measurement

This illustrates some techniques.

Jay_Diddy_B

[Wolfram]

My question is this. Does the Rds on measured in that waveform actually reflect the junction temperature? If it does, that would imply that my junction temperature gets up to about 175C (2mOhm/1mOhm -> 2 -> about 175C) before tailing off in this example. If not, what are the timescales at which the Rdson -> junction temperature chart is applicable? What is the explanation for the varying Rds on with the gate voltage being fairly ideal?

Rdson is not only a function of temperature and Vgs, it also depends on drain current, rising with higher currents. This could be what you are seeing in your measurements.

Edit: I had a look at some data for representative devices (I assume these are in the 20 - 30 V Vdsmax range, given the ratio of nominal current to Rdson?), and the effect is not as dramatic as with the higher voltage devices I'm used to working with, especially not with your 10 V Vgs. The datasheets I found also only show the Rdson change with current at room temperature, the slope can be steeper at elevated temperature. If you have the actual device part number, it should be possible to make some more accurate estimates.

[PartialDischarge]

The thing is, I did spend a lot of time researching circuits to measure dynamic saturation Vds. And my conclusion is that there is nothing really good out there, bad linearity around 0, high offsets.... Since many rely on limiting resistors to the diodes then the higher the VDS voltage the slower the response time. Most of them have a settling time around 10us which is too slow and undervoltage swings are typical (as seen in the OP traces), this problem getting worse the faster the dV/dt of the negative going Vds.

So eventually I did design something different, attached is a measurement done with this circuit of mine, sorry can't post it because it will become a product soon.

CH2 is a 200V rail being switched to 0 with a mercury relay, measured with a x100 passive probe.
CH3 is the saturation voltage 1:1 scale, starting at 15V.
CH3 is again measured in the next graph alone, without CH2 as I tested that the x100 probe did affect the measurement.
The settling time is less than 100ns, and there is no negative going peak given the fast decay of the Vds. This is the kind of speed needed in todays fast inverters that have <1us ON time.

The linearity is tested with a triangle wave around 0, but I don't have any graphs at hand. Also there will be a x10 times output for devices that have very low saturation Vds.

[Jay_Diddy_B]

Hi,

I found a nice paper on measuring the  on-state voltage here:
https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8438422

Regards,
Jay_Diddy_B