Want to build a Mosfet and Jfet tester

[neuralsim]

Hi all, I'm kinda new here, but I've been lurking for a while and I love Dave's videos, I've watched him even before I decided to take up EE as a hobby.

But now here I am and I'm hoping to build something that seems to be highly desired and often asked for but nobody has a black box solution for.

Briefly, my experience level is that I'm a highly skilled software engineer with decades of professional experience, and a highly novice electrical engineer, with no professional experience, but a lot of equipment and enthusiasm. I understand the purpose of all the basic components and how to build simple serial circuits. I'm putting most of my research time lately into learning about amplifiers and SMPS's. I get the impression that if I learn to build a decent amplifier and SMPS, I can do just about anything, because EE is like software in a way, it's all about input and output.

Which brings me to my project. There are numerous test devices that can run basic pass/fail checks on mosfets and jfets, but I'm not aware of a device that can really test them for the characteristics in the datasheet, and certainly don't know of a black box device that can show a performance curve or test a FET device for anything other than forward voltage and activation thresholds.

Which is what I'd like to build. I would like all my tests to be battery powered, and I'm willing to go up to a 20v power supply but I would prefer 12v. I would like to perform at least a few tests at higher voltages than your normal transistor tester, to help determine if the mosfet or jfet meets its stated performance characteristics, to help distinguish between good and bad/counterfeit FETs.

Is this feasible? Or will I need AC/mains power to really do anything useful here? What guidance can you give me for reading and tips for getting started. Any and all comments welcome, including, "y00r n idjit."

[Manul]

Hello. I guess it could have a niche application. Also it seems to me, that it is kind of a project which easily may become very complex, so it needs clear boundaries. For example, it is easy to measure MOSFET's gate threshold. It is harder to measure RDSon curve. It probably becomes really complex if you want to measure things like reverse transfer capacitance vs DS voltage. I suggest to first set your requirements very clearly, what specific characteristics you want to measure.

[neuralsim]

Hello. I guess it could have a niche application. Also it seems to me, that it is kind of a project which easily may become very complex, so it needs clear boundaries. For example, it is easy to measure MOSFET's gate threshold. It is harder to measure RDSon curve. It probably becomes really complex if you want to measure things like reverse transfer capacitance vs DS voltage. I suggest to first set your requirements very clearly, what specific characteristics you want to measure.

Thanks, that's part of what I'm looking for here. Yeah the gate threshold is easy, and that's why standard testers already have that function. I'll have it too, but I'm hoping you guys and gals can give me some guidance on what other characteristics would constitute useful and practical tests to determine whether the mosfet or jfet is fit for purpose and not either bad or counterfeit.

Also, am I crazy to try to pull it off with a battery powered device?

[Kleinstein]

There are instruments to test transistors and measure the characteristic curves. These are usually quite expensive. There were special instruments in the past, today it is more like a 2 channel SMU. One may still need more than 1 instrument to do all the tests. There are just to many parameters and to large a range of parts from tiny one where pA matter to large ones for 100s of A. Other test like noise or temperature effects also need special instruments.

There are some quite impressive universal component testers build a round a µC, that also to the pin identification. However they only do a crude parameter check at a low (e.g. 5 V) voltage.

So chances are one would have to settle for something in between and decide on a few parameters.
Battery operation is OK for the lower power part. It may be a problem for higher current.
I would start with a simple limited version for maybe just one parameter or one curve. It can help to have some hand on experience and get it done in a reasonable time.

If one wants to get more like a single number, more like a complete curve one would likely need some way to send the data to a PC or similar. This usually would something like a µC with ADC and maybe DACs. This would would be one of the first steps - especially when coming from the software side.

[todorp]

There are a few test circuits for various characteristics of BJTs, JFETs, and MOSFETs described in the 3e of the Art of Electronics and The Art of Electronics the X-Chapters...

[T3sl4co1l]

I'm not aware of a device that can really test them for the characteristics in the datasheet

Yeh, which ones?

Here are some broad use classes:
1. General purpose and AF amplification
2. Power amplification
3. RF amplification
4. General purpose switching
5. Power switching

Parts made and tested for these classes will have different combinations of parameters.  All will have some, like maximum voltage and current ratings, but others are specific to one application or another.

For example, small-signal amplifiers usually have transconductance, and noise voltage or noise figure, and may have additional small-signal characteristics (y-parameters for AF, s-parameters for RF).  Switches have resistance, on/off gate voltage, and some parameters related to switching speed (gate charge, internal gate resistance, rise/fall time, capacitance curves, and for MOSFETs, body diode recovery).

Even just the definition of some parameters may vary between devices: RF amps usually use design-center ratings, on the expectation that they will be used with an inductive load -- the load develops a peak voltage about double the supply voltage so they rate it by DC supply voltage instead.  Whereas the design-maximum rating is the maximum safe (peak or DC) voltage to apply to the device, with breakdown occurring soon after.

The amusing part, I guess, is that there really isn't that much variation between transistor types.  After you factor out V/I ratings (a matter of scale: how much FET, of what aspect ratio, is put on how big of a die), and the figure-of-merit (more or less, Ft ~ y_fs / Ciss, but also R_G, Coss, etc.), the remaining parameters aren't hugely different: Rds(on) and Vgs(th) tempcos are pretty typical, y_fs ranges from exponential (subthreshold) to linear over the operating range, etc.

Basically, you don't learn all that much from a curve tracer; you can learn more with just a couple of tests, like Rds(on) (Id vs. Vds, Vgs=Vgs(on)), leakage and breakdown (Id vs. Vds, Vgs=Vgs(off)), and transfer (Id vs. Vgs, Vds = fixed).

Whereas back in the days of vacuum tubes, curve tracers were quite handy, because there was quite a bit of difference between types -- some curve upwards, some bend shallow, some even go up and down.  There were a lot of standard types with expected behavior (most of the "up and down" types were obsolete by the 1950s), but still some with enough variation that they're easier to design with using curves; and also other more interesting cases thanks to the richness of free-electron physics.

Tim

[David Hess]

Unfortunately the different sets of tests are not conducive to the same testing circuit so some subset of tests which are must be designed for unless you want to build essentially multiple testers into a single instrument.  Even different DC tests can be difficult because of the range of currents required.

So to keep it simple at the start, I would design the equivalent of a SMU, source measurement unit, which controls voltage and current and reads them back at a single pin.  That will allow measurement of all of the static parameters within the supported range of currents which by itself will be challenging enough.  Achieving a good dynamic range of current will be the most challenging part.

I have looked into the above before and while logarithmic control and read-back of current presents considerable difficulties, it has the advantage of providing excellent dynamic range without range switching.  But range switching is more straightforward and suitable for currents above milliamps.

You might also take a look at the published designs of curve tracers from companies like Tektronix and exactly how they were used to measure various parameters.

[neuralsim]

I'm not aware of a device that can really test them for the characteristics in the datasheet

Yeh, which ones?

Thanks to you and everyone else for the polite and helpful replies! Mainly I'm looking for a good way to separate real from fake, other than cutting open the package. I bought some cheap mosfets before I knew that they are so often faked. I've now cut some open, and they're definitely fake, they're marked as Infineon, but the die is tiny and definitely not the size of the real die. In fact my fakes have even smaller dies than most of the photos of fakes that I've seen! Not a big deal I paid $15 for 50 assorted fake mosfets. I won't make that mistake again, I'll be ordering from digikey or Mouser from now on.

After some consideration, I'm thinking showing performance curves is too ambitious for the project I'm planning and I'm willing to settle for a few tests that will help people (including me) to quickly check if the FET is fake.

Also I wonder if anyone can tell me what my component tester is trying to convey. For a mosfet it shows a value for Vt and C. I'm assuming Vt is the gate threshold. I'm assuming C is some kind of capacitance test, but which? The tester is a cheap thing with crappy documentation, so I'm just left to guess. Any ideas?

[neuralsim]

Unfortunately the different sets of tests are not conducive to the same testing circuit so some subset of tests which are must be designed for unless you want to build essentially multiple testers into a single instrument.  Even different DC tests can be difficult because of the range of currents required.

So to keep it simple at the start, I would design the equivalent of a SMU, source measurement unit, which controls voltage and current and reads them back at a single pin.

Heh I googled SMUs just now. Simple huh? They cost more than most of the cars I've owned. I think I'm going to forgo doing a current sweep and try to pick out some common measurements strategically designed to catch fake mosfets. Heck I may just set up a counterfeit mosfet killer like this one:

I guess if I try to sell this or even distribute the design I'll have to have a disclaimer that it's literally designed to destroy fake equipment, use at your own risk, etc. I'm not sure about the feasibility of being able to test a counterfeit mosfet without killing it.

[Manul]

That is medieval and not very universal

I do not know much about counterfeit mosfets and how much they are different from real ones. But talking about signature of a mosfet, I would probably choose:

1. Gate threshold
2. Rds(on) at datasheet specified gate voltage (or a full curve)
3. Gate capacitance

Such measurements would be relatively easy to achieve, does not need huge power and should reveal if there is a wrong die inside.

[neuralsim]

That is medieval and not very universal

I do not know much about counterfeit mosfets and how much they are different from real ones. But talking about signature of a mosfet, I would probably choose:

1. Gate threshold
2. Rds(on) at datasheet specified gate voltage (or a full curve)
3. Gate capacitance

Such measurements would be relatively easy to achieve, does not need huge power and should reveal if there is a wrong die inside.

You're right but I kinda like it as an option, testing to (potential) failure, up to 5-6 amps. With suitable dire warnings that you're basically intentionally going to destroy it if it is fake. I thought maybe for some people that's maybe better than it failing inside their amp when it gets driven hard?

[David Hess]

The test to measure power capability is not too difficult and is non-destructive.  Measure the body diode forward voltage drop at ambient temperature, pulse the device at a given power by knowing the drain voltage and current, and then measure the forward voltage drop again to get the junction temperature.

Rinse and repeat and you get get an idea of the die size and power capability which is usually completely different on fakes.  Of course measuring the capacitance works for this also.

[T3sl4co1l]

Thanks to you and everyone else for the polite and helpful replies! Mainly I'm looking for a good way to separate real from fake, other than cutting open the package. I bought some cheap mosfets before I knew that they are so often faked. I've now cut some open, and they're definitely fake, they're marked as Infineon, but the die is tiny and definitely not the size of the real die. In fact my fakes have even smaller dies than most of the photos of fakes that I've seen! Not a big deal I paid $15 for 50 assorted fake mosfets. I won't make that mistake again, I'll be ordering from digikey or Mouser from now on.

After some consideration, I'm thinking showing performance curves is too ambitious for the project I'm planning and I'm willing to settle for a few tests that will help people (including me) to quickly check if the FET is fake.

Also I wonder if anyone can tell me what my component tester is trying to convey. For a mosfet it shows a value for Vt and C. I'm assuming Vt is the gate threshold. I'm assuming C is some kind of capacitance test, but which? The tester is a cheap thing with crappy documentation, so I'm just left to guess. Any ideas?

Now, if you're a. not so much concerned with reproducing the complete list of datasheet specs and b. mostly interested in power switching transistors, there's quite a lot you can glean from one or two related tests...

The gate charge switching test can be performed two or three ways, and if a dual channel digitizing oscilloscope is handy, one can measure Qg, Ciss, Coss, Crss, Rds(on), y_fs, Vgs(th), Vdss, td(on/off), tr, tf and probably some other things.  The equivalent circuit is common-source DUT, with switchable strong driver or low current CCS to the gate, and switchable load resistor, inductor or CCS to the drain.

The gate charge curve is measured with CCS drive, and either resistor or CCS load (if resistor, then R = V+ / Id; if CCS, then I = Id, with V+ and Id stated as parameters for the test).  When Ig(on/off) is small, voltage drops on inductances and stray resistances don't matter, and the system is quasi-static: every bit of charge we move into/out of the gate, is another increment to the change in gate and drain voltages.  Thus, Vgs and Vds are swept through real load conditions, and the rates (dV/dt) are proportional to capacitances (especially Crss).

When a drain load resistor is used, we can also measure y_fs, from the dVds / dVgs (transfer) slope.

When a drain CCS is used, we can also measure Coss.  We use a small current, and a strong gate driver this time.  Vgs is pulsed on, pulling drain to GND, then turned off sharply.  Vds hovers near zero (actually it's pulled slightly negative, due to Crss), then rises at Id = Coss(Vds) dVds/dt.  Since Id is constant, Coss is measured by the slope, dV/dt.

And of course we can measure Rds(on) when Vgs=Vgs(on) and Id is enough to measure a sensible Vds.

Note that for testing full load switching, the drain CCS must be quite large.  Datasheets usually give an equivalent circuit using two identical devices, one strapped to fixed Vgs so that it acts as a current limiter.  (Such "suicide biasing" is discouraged, and probably a real circuit would use a beefier MOSFET -- one with a wider SOA -- and source degeneration, and maybe active control as well, to stabilize the current.)  This dissipates a lot of heat, and can only be done so slowly before one or both transistors blow up!

The main thing that varies with load current, is Vgs(plateau), so it's not a big deal to have full load current in a Qg test.  Also, the main difference between a load resistor or CCS, is the slope of the plateau, which is flatter for the CCS.  (Note that no real device has the square corners shown in the datasheet -- for whatever reason, it's become tradition to draw it in straight segments, but it is of course rounded over, real devices don't have cusps*.  IXYS datasheets for example, I think tend to show real curves.)

*More specifically, real phenomena are infinitely differentiable, among other things.

The testing circuit, then, would have a couple of driver circuits, a supply voltage or a few, and probe points.  Supplies can be continuous duty or pulsed (toss on some big capacitors and use just a few watts; heck, you could probably make a USB tester if you really want ).  Scope probes must be carefully compensated.  Scope is connected to computer and set to the ranges and settings required, and waveforms are downloaded for processing.

The scope could be integrated as onboard ADCs, with careful attention paid to the flatness of the analog front ends.  Hack: an ideal differentiator could be used instead of, or in addition to, measuring the voltage directly, to get a more accurate slope reading.

Hmm, might be possible to do much of this with an MCU; the fast sampling for rise/fall is the biggest hurdle I think.  Equivalent time sampling can certainly be used; if accurate timing can be generated internally, and if the internal ADC has a snappy S&H, that'll do; otherwise, some amount of this can be shifted outside as analog hardware.

Tim

[neuralsim]

The test to measure power capability is not too difficult and is non-destructive.  Measure the body diode forward voltage drop at ambient temperature, pulse the device at a given power by knowing the drain voltage and current, and then measure the forward voltage drop again to get the junction temperature.

Rinse and repeat and you get get an idea of the die size and power capability which is usually completely different on fakes.  Of course measuring the capacitance works for this also.

This sounds like the perfect answer. Keeping in mind I'm a beginner, would you mind elaborating a bit? Let me repeat it in my own words and let's see if I understand it. The body diode is between source and drain, so first I'll measure the voltage drop across them.

Then I'll use this voltage drop to calculate a resistance, and pulse the device by sending a square wave to the gate with a given voltage and resistance on DS and measure the drop between DS again afterwards, which will give me a junction temperature... how?

Also since I cannot simply detect the type of mosfet in advance, nor am I going to have a database of them, the test power level and frequency sounds like maybe a user selectable option? Also the heat calculation, should it just be displayed to the user after the test and then they should be able to figure out whether it's good? Or can pass/fail somehow be automated?

[Kleinstein]

A diode can make a good temperature sensor. The common approximation is to have a change of some -2 mV per K of temperature rise.  Another approximation is having a linear curve that extrapolates to 1.24 V at 0 K. This slightly less known form also works for more unusual cases where the diode drop is small (e.g. large transistors with a small test current) or large (e.g. low leakage diodes).
The diode voltage drop is measured as a voltage, not as an apparent resistance. With a DMM this is the diode test mode, for the extra test circuit it would take a reasonable constant current. This could be something like a resistor to not too small fixed voltage (e.g. some -5 V).

The suggested thermal test would check the thermal resistance from the die to the case / heat sink. Usually fakes use cheaper and thus smaller dies that usually have a higher thermal resistance. For comparison it would help to also have a known good one, as the resulting temperature rise is not so easy to connect to a typical curve shown. It is  related to the transient thermal resistance popular in more modern data-sheets.

With a N-MOSFET and the source as ther reference / ground level, it would need a negative voltage (current) at the drain. For easy an efficient heating one would need a positive voltage and current. So the test has to be transient: something like 0.1 - 1 seconds of heating (e.g. some 20 V and 1 A) and than some 0.01 to 1 seconds for testing the temperature in reverse with maybe some 0.1 mA.  To actually measure the die temperature and not so much the quality of the heat sink, one would have to measure relatively fast, ideally within only ms from turning off the heating.

For a defined heating one could use the DUT MOSFET as a constant current sink circuit. So an OP to control the gate and a small resistor (e.g. 0.1 Ohm) on the source side to measure the current.  A simpler version could use a larger resistor (1-3 Ohms, gets hot) and a limited gate voltage. 

The gate would also be used to turn off the current. One would still need an additional switch/transistor to also turn off the main voltage, so the negative test current can work.

Because of the relatively high power needed here, this test is not so suitable for battery operation.

[David Hess]

Then I'll use this voltage drop to calculate a resistance, and pulse the device by sending a square wave to the gate with a given voltage and resistance on DS and measure the drop between DS again afterwards, which will give me a junction temperature... how?

The temperature coefficient of the body diode is about -2.1 millivolts per degree Celsius, with a variable offset of about 0.6 volts.  So if a measurement of the forward voltage drop is taken at ambient temperature before the chip is heated up, then a later measurement of the forward voltage drop will indicate the junction temperature.  So some power can be run through the chip while monitoring the junction temperature to determine the thermal capability.

Real testers measure the junction temperature using the "delta Vbe" method which requires no measurement at ambient to establish the forward voltage drop but that is not strictly necessary in this case.  The delta Vbe method uses the ratio of two test currents to make an absolute measurement of temperature.

Also since I cannot simply detect the type of mosfet in advance, nor am I going to have a database of them, the test power level and frequency sounds like maybe a user selectable option?

Perhaps not.  I was thinking you could raise the power until the junction temperature reaches a fixed point of 75 to 100 Celsius and use that to determine the rough power handling capability which is related to die size.  For power packages, this would need to be done while mounted to a heat sink; otherwise the package to ambient thermal resistance is what would be measured.

Measuring the power capability like this is more complicated than some of the other tests but I thought I would mention it as an option anyway.

Also the heat calculation, should it just be displayed to the user after the test and then they should be able to figure out whether it's good? Or can pass/fail somehow be automated?

A pass/fail test would require some type of reference part.

[neuralsim]

After some consideration, I'm thinking showing performance curves is too ambitious for the project I'm planning and I'm willing to settle for a few tests that will help people (including me) to quickly check if the FET is fake.

Now, if you're a. not so much concerned with reproducing the complete list of datasheet specs and b. mostly interested in power switching transistors, there's quite a lot you can glean from one or two related tests...

Yep. I want to mainly test mosfets and jfets for suitability in power and/or amplifier circuits, with a focus on testing for fake or broken parts, more emphasis on fake but ideally something that could tick both boxes would be perfect.

I'm carefully reviewing the rest of your reply along with reference material including a PDF copy of The Art of Electronics that fell off a data truck. It's semi-over my head but not entirely, with time and reference materials and some fooling around in the lab I think I can decode all of it. I really appreciate the insight and the help anyway though, even though I have to study to understand it completely.

The gate charge switching test can be performed two or three ways, and if a dual channel digitizing oscilloscope is handy, one can measure Qg, Ciss, Coss, Crss, Rds(on), y_fs, Vgs(th), Vdss, td(on/off), tr, tf and probably some other things.  The equivalent circuit is common-source DUT, with switchable strong driver or low current CCS to the gate, and switchable load resistor, inductor or CCS to the drain.

I do have a digital oscilloscope (Hantek 5102p), and I think it has logic analyzer capability, but haven't hooked it up to the PC yet to check. I'm just getting my lab built out, I have a basic setup minus a bench power supply, which I'm planning to build next time I can get my hands on an ATX power supply. Alternately I have a couple very powerful laptop supplies I can build a converter for (19.5v 16.9a, 330W).

I'll probably be messing around with the scope and a couple of my (as I now realize after cutting them open, fake) mosfets this weekend. Thanks again for the detailed information, now I just need to put in the work to understand all of it to make it worth your effort!

[neuralsim]

Also since I cannot simply detect the type of mosfet in advance, nor am I going to have a database of them, the test power level and frequency sounds like maybe a user selectable option?

Perhaps not.  I was thinking you could raise the power until the junction temperature reaches a fixed point of 75 to 100 Celsius and use that to determine the rough power handling capability which is related to die size.  For power packages, this would need to be done while mounted to a heat sink; otherwise the package to ambient thermal resistance is what would be measured.

Measuring the power capability like this is more complicated than some of the other tests but I thought I would mention it as an option anyway.

This sounds like a really good method to get the information I am after: basic pass/fail for if the device is up to spec. When you say you don't think battery powered is feasible, what kind of power are we talking? I'm OK with a fairly powerful battery, I'm planning for the eventual product to have a powerful rechargeable DC power supply. I don't want a laptop sized battery but a large battery pack or two is OK with me.

Also the heat calculation, should it just be displayed to the user after the test and then they should be able to figure out whether it's good? Or can pass/fail somehow be automated?

A pass/fail test would require some type of reference part.

[Kleinstein]

The power needed for thermal testing depends on the transistor size. SOT23 / TO92 would be easy on battery, but with larger parts in an TO3 or TO247 case one would likely want something like 10 or even 50 W to get a significant temperature rise. With lower power one would need more sensitive temperature measurement.

It is possible with a battery, but more like a large one.

An easy test to check the basic function (e.g. detect broken MOSFETS) and measure the gate charge / capacitance would be switching the MOSFET via a relatively large resistor at the gate. This could be as little as an small oscillator with an NE555 some 1 K to the gate and something like a  9 V batteries and 1 K as a load at drain. The measurement would be with the scope at the gate. Already looking at the gate charge could tell the difference between a large and small MOSFET (e.g. fake). With a separate test with a higher drain voltage one could also get extra info on the gate drain capacitance separately.

For JFETs the usual tests wanted for an amplifier are the threshold voltage for matching and possibly checking the noise to find low noise ones (kind of a check of quality).