6
\$\begingroup\$

I am trying to stabilize the current in coils in my experiment. For that I am using a circuit as in the picture:

enter image description here

First section is a fast switch. It is also galvanically isolated from the TTL signal with opto-coupler (not in the picture)

Second section is fast discharging of the coils. When we suddenly stop the current from the supply, the coils respond in reverse voltage build-up opening the diode. Then in quarter of a cycle of the LC circuit the energy flows from the coils to the capacitor and the diode closes. Then it slowly discharges through the resistor. Coils inductance is 350 uH.

Third section is linearized MOSFET. I am using OPA277 here. It linearizes the response of the MOSFET by comparing the voltage on the sense resistor with the input and adjusting the gate voltage. 10R resistor is added to reduce gate ringing of the MOSFET. It is connected to the PID control (I am currently using RedPitaya for that). Current measurement for the PID is done independently with a current transducer. In the circuit there are 4 such sections connected in parallel to lower the power dissipation.

What my problem is, that when I am sending larger currents (a few amps) after some time the MOSFET responsible for current stabilization burns. Firstly I thought that it is due to heating up, but it is placed on a water cooled aluminum casing and its temperature doesn't rise. Then I added additional clamping Shottky diodes bypassing the body diode of the MOSFETs (each transistor got one, directly on the terminals of their casings) because I thought it is due to high reverse voltage. The fast discharging setup lets the current discharge in around 100 us and thus large reverse voltage is generated (I calculated it to be around 414 V at 70 A - largest current I want to use).

I have run out of ideas what could be the problem. I would be glad for some explanation what can cause the burning of the MOSFETs and possibly how to protect them.

\$\endgroup\$
3
  • 1
    \$\begingroup\$ Could you explain your choice of device? In particular, why choose one for linear operation, which does not publish an SOA? | Why choose a linear method at all? You have two switches, this is the a perfect setup for an efficient two-switch buck converter with equal up and down [current] slew rates; i.e. with PWM fast enough the coil doesn't care (10s, 100s kHz?), the control loop response is symmetrical (or as symmetrical as it can be given the method), not nonlinear as your RCD snubber will do. \$\endgroup\$ Dec 4, 2023 at 15:55
  • 1
    \$\begingroup\$ As Tim says, please explain why you need/want a linear method. 70 A at 25 V is nearly 2 kilowatts of power getting turned into heat somewhere. Part of that will heat the coil, but a significant part is certainly going to end up in the transistor. Sinking a kilowatt of heat is no easy task, even with water cooling. And the better (lower resistance) your coil is, the more heat ends up in the transistor. \$\endgroup\$
    – TooTea
    Dec 4, 2023 at 16:14
  • 1
    \$\begingroup\$ Try: (1) Zener connected G-S physically & electrically close to FET. Rated at somewhat above max gate drive voltage. This clamps Miller capacitance coupled spikes from drain to gate. Ive found this extremely effective with inductive loads. (2) Add small reverse biased Schottky, also G-S close to device. This clamps negative excursions of any gate ringingand rapidly removes energy. \$\endgroup\$
    – Russell McMahon
    Dec 5, 2023 at 23:22

3 Answers 3

6
\$\begingroup\$

The thermal resistance of the MOSFET junction-case interface, compared with the thermal capacity, coupled with thermal runaway, makes it very easy to burn out these device before the case even gets warm.

Here's an excellent app note on the challenges of MOSFETs in linear mode.

You may well be better using an IGBT for this kind of control. There are some very high current modules with excellent thermal bonding to avoid exactly these kind of problems, but same gate drive as your existing MOSFETs.

Datasheet for a Vishay example of similar physical size here, but for more money they can be had all the way up to 7200A!

Note most IGBT are not designed for linear use, but are more stable in this mode than equivalently rated MOSFETs. As noted in the linked app note, older generation devices (which tend to have bigger dies) and higher voltage ratings (which also tend to have bigger dies) are generally better in linear mode.

\$\endgroup\$
4
  • 1
    \$\begingroup\$ I both love and hate this reference [AP99007]: it comes the closest to explaining the situation, but every single appnote I've seen, this one included, trips and falls flat on its face. The critical point in this one is the line top of page 11: "This means that the temperature coefficient must be known. The latter cannot be easily calculated." Ultimately, no appnote accounts for thermal spreading effect and therefore the runaway exponent; we are left to rely on the SOA curve, as published. \$\endgroup\$ Dec 4, 2023 at 15:51
  • 1
    \$\begingroup\$ @TimWilliams Absolutely agree they all seem to ignore this key factor, and none of the datasheets really help. But this app note does directly refer to the increasing issue with Ron going down, but dies getting smaller, but referencing the use of older tech: "This means that MOSFETs of previous technology generations and/or higher voltage classes will be more suitable for this kind of application." \$\endgroup\$
    – colintd
    Dec 4, 2023 at 15:55
  • 1
    \$\begingroup\$ You can also find MOSFETs which claim to be designed for linear operation, e.g. these. No clue how those fare vs IGBTs in this kind of thing... \$\endgroup\$
    – mbrig
    Dec 4, 2023 at 23:40
  • \$\begingroup\$ @mbrig I've used those bolted to a big fan cooled aluminum block. They survived but I was walking a knife edge. \$\endgroup\$
    – DKNguyen
    Dec 6, 2023 at 2:32
8
\$\begingroup\$

What my problem is, that when I am sending larger currents (a few amps) after some time the MOSFET responsible for current stabilization burns.

You are using the MOSFET in a linear circuit application (i.e. not to switch loads) and therefore, you need to be fully aware of what can certainly happen in these circumstances: -

enter image description here

Additionally, you have chosen a MOSFET that is designed for switching applications. It's all there listed on page 1 of the data sheet: -

enter image description here

Do you see anything in the above that gives any hint that it will be suitable for linear applications?

The problem is this: if you operate with low gate-source voltages (in the approximate range of 4.5 volts to 6.5 volts), then, as the junction warms, it will take more current, warm more, take more current etc. etc..

This can take maybe as long as 10 ms to cause parts of the MOSFET junction to reach temperatures of up to 600 °C and melt. A heatsink will not save a MOSFET from failing due to this mechanism. In some cases, it can take less than a millisecond to fail.

It's also called the "Spirito effect" named after the guy who published the first paper on this problem.

G. Breglio, F. Frisina, A. Magri, and P. Spirito, “Electro-Thermal Instability in Low Voltage Power MOS: Experimental Characterization,” IEEE Proceedings ISPSD 1999, Toronto, p233

\$\endgroup\$
8
  • \$\begingroup\$ As a corollary, this is actually one of those applications where a BJT might make more sense; they don't exhibit thermal runaway like that, and are frequently designed for linear applications, unlike power MOSFETs that are almost always optimized for switching. A TIP120 might be suitable, for instance. \$\endgroup\$
    – Hearth
    Dec 4, 2023 at 14:28
  • \$\begingroup\$ Ah, I missed the 70 A requirement--that's a bit much for a TIP120! You'd have to use a handful of paralleled transistors; I'm not aware of any single BJT that's rated for 70 amps (and a search on digikey only turns up obsolete parts). Three or four TIP35s might work, but you'd need some additional driving circuitry (perhaps Darlington them with a medium-power transistor, maybe even a TIP120), and you'd need to ensure proper current sharing (easy since you're already using it for current control). \$\endgroup\$
    – Hearth
    Dec 4, 2023 at 14:41
  • 1
    \$\begingroup\$ Andy, Spirito effect is meaningless here; existence of negative gate tempo is no guarantee of SOA violation of a given part. More to the point: the datasheet simply doesn't publish an SOA at all, which I would consider quite damning. \$\endgroup\$ Dec 4, 2023 at 15:42
  • 1
    \$\begingroup\$ @Hearth Both device types in general exhibit instability. Particularly at high voltages (high power density). It's a combination of electrical characteristics and physical construction that determines stability, the latter of which is never documented in datasheets, so we cannot calculate it, and have to rely on the SOA alone. \$\endgroup\$ Dec 4, 2023 at 15:45
  • 1
    \$\begingroup\$ @TimWilliams This is made worse by the improvements in Ron with MOSFETs, as for a given peak current rating, the dies are generally getting smaller, so the margin when used in linear mode gets smaller. \$\endgroup\$
    – colintd
    Dec 4, 2023 at 15:48
1
\$\begingroup\$

Noting that:

  • I'd expect that your PID controller will not make any very fast gate voltage swings during normal operation, but may at turn on and off.

  • It's not clear if damage occurs during PID regulation before fast discharge.

I suggest trying, in order of likely decreasing effectiveness (ie zener best):

Add a Zener diode connected G-S, physically & electrically close to FET.
Rated at somewhat above max gate drive voltage.
This clamps Miller capacitance coupled spikes from drain to gate.
I've found this extremely effective with inductive loads.

Add a small reverse biased Schottky, also G-S close to device.
This clamps negative excursions of any gate ringing and rapidly removes energy.

Without full circuit and waveforms this is "just an informed gues" as to a possible mechanism.
As your diagram does not show all actual circuit elements and values it is not obvious how adequate your section 3 gate drive change is. While the FET is intended to operate in the linear mode here, if drive is not high enough it MAY be less turned on initially and MAY stay turned on longer than intended during large load energy transitions. If a gate driver is not used, consider using one.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.