Why MOSFETs fail in Solid State TC duty
After emptying many pounds worth of dead MOSFETs from the trash can in my workshop, I decided that it might be worth actually compiling a list of reasons why MOSFET devices might fail in solid state tesla coil applications.
There are quite a few possible causes for device failures, ranging from obvious reasons to some mechanisms which are not so immediately obvious. When investigating a fault, this list can be used as a check-list:
MOSFETs have very little tolerance to overvoltage. Damage to devices may result even if the voltage rating is exceeded for as little as a few nanoseconds. MOSFET devices should be rated conservatively for the anticipated voltage levels, and careful attention should be paid to suppressing any voltage spikes or ringing.
High average current causes considerable thermal dissipation in MOSFET devices due to the relatively high on-resistance. If the current is very high and heatsinking is poor, then the device can be destroyed by excessive temperature rise. MOSFET devices can be paralleled directly to share high load currents.
Short duration, massive current overload can cause progressive damage to the device with little noticeable temperature rise prior to failure. (Also see shoot-through and reverse recovery sections below.)
If the control signals to two opposing MOSFETs overlap, then a situation can occur where both MOSFETs are switched on together. This effectively short-circuits the supply and is known as a shoot-through condition. If this occurs, the supply decoupling capacitor is discharged rapidly through both devices every time a switching transition occurs ! This results in very short but incredibly intense current pulses through both switching devices.
The chances of shoot-through occuring are minimised by allowing a dead time between switching transitions, during which neither MOSFET is turned on. This allows time for one device to turn off before the opposite device is turned on.
When switching current through any inductive load (such as a Tesla Coil,) a back EMF is produced when the current is turned off. It is essential to provide a path for this current to free-wheel in the time when neither switching device is carrying the load current.
This current is usually directed safely back to the supply rails by means of a free-wheel diode connected anti-parallel with each switching device. When MOSFETs are employed as the switching devices, the designer gets the free-wheel diode "for free" in the form of the MOSFET's intrinsic body diode. This solves one problem, but creates a whole new one...
A high Q resonant circuit such as a Tesla Coil is capable of storing considerable energy in its inductance and self capacitance. Under certain tuning conditions, this causes the current to "free-wheel" through the internal body diodes of the MOSFET devices as one MOSFET turns off and the other device turns on. This behaviour is not a problem in itself, but a problem arrises due to the slow turn-off (or reverse recovery) of the internal body diode when the opposing MOSFET tries to turn on.
MOSFET body diodes generally have a long reverse recovery time compared to the performance of the MOSFET itself. If the body diode of one MOSFET is conducting when the opposing device is switched on, then a "short circuit" occurs similar to the shoot-through condition described above.
This problem is usually eased by the addition of two diodes surrounding each MOSFET. Firstly, a Schottky diode is connected in series with the MOSFET source. The schottky diode prevents the MOSFET body diode from ever being forward biased by the free-wheeling current. Secondly, a high speed (fast recovery) diode is connected in parallel to the MOSFET/Schottky pair so that the free-wheeling current bypasses the MOSFET and Schottky completely.
This ensures that the MOSFET body diode is never driven into conduction. The free-wheel current is handled by the fast recovery diodes which present less of a "shoot-through" problem.
If the MOSFET gate is driven with too high a voltage, then the gate oxide insulation can be punctured rendering the device useless. Gate-source voltages in excess of +/- 15 volts are likely to cause damage to the gate insulation and lead to failure. Care should be taken to ensure that the gate drive signal is free from any narrow voltage spikes that could exceed the maximum allowable gate voltage.
MOSFET devices are only capable of switching large amounts of power because they are designed to dissipate minimal power when they are turned on. It is the responsibility of the designer to ensure that the MOSFET devices are turned hard on to minimise dissipation during conduction. If the device is not fully turned on then the device will have a high resistance during conduction and will dissipate considerable power as heat. A gate voltage of between 10 and 15 volts ensures full turn-on with most MOSFET devices.
Little energy is dissipated during the steady on and off states, but considerable energy is dissipated during the times of a transition. Therefore it is desirable to switch between states as quickly as possible to minimise power dissipation during switching. Since the MOSFET gate appears capacitive, it requires considerable current pulses in order to charge and discharge the gate in a few tens of nano-seconds. Peak gate currents can be as high as an amp.
MOSFETs are capable of switching large ammounts of current in incredibly short times. Their inputs are also relatively high impedance, which can lead to stability problems. Under certain conditions high voltage MOSFET devices can oscillate at very high frequencies due to stray inductance and capacitance in the surrounding circuit. (Frequencies usually in the low MHz.) This behaviour is highly undesirable since it occurs due to linear operation, and represents a high dissipation condition.
Spurious oscillation can be prevented by minimising stray inductance and capacitance around the MOSFETs. A low impedance gate-drive circuit should also be used to prevent stray signals from coupling to the gate of the device.
MOSFET devices have considerable "Miller capacitance" between their gate and drain terminals. In low voltage or slow switching applications this gate-drain capacitance is rarely a concern, however it can cause problems when high voltages are switched quickly.
A potential problem occurs when the drain voltage of the bottom device rises very quickly due to turn on of the top MOSFET. This high rate of rise of voltage couples capacitively to the gate of the MOSFET via the Miller capacitance. This can cause the gate voltage of the bottom MOSFET to rise resulting in turn on of this device as well ! A shoot-through condition exists and MOSFET failure is certain if not immediate.
The Miller effect can be minimised by using a low impedance gate drive which clamps the gate voltage to 0 volts when in the off state. This reduces the effect of any spikes coupled from the drain. Further protection can be gained by applying a negative voltage to the gate during the off state. Eg. Applying -10 volts to the gate would require over 12 volts of noise in order to risk turning on a MOSFET that is meant to be turned off !
Imagine the effect of connecting just 1pF of capacitance from the top of your sparking tesla coil to each of the sensitive points in your solid state controller. The hundreds of kilovolts of RF present would have no problem driving significant current through the tiny capacitors directly into the control circuit.
Well this is exactly what happens in practice if the controller is not placed in a screened enclosure !
It takes little stray capacitance to high impedance points of the control circuitry to cause abnormal operation. Bare in mind that a controller which is not operating correctly could attempt to turn on two opposing MOSFET devices at the same time. Effective RF screening of the control electronics is essential.
It is also highly desirable to segregate power and control circuitry. Rapidly changing currents and voltages present in the power switching circuit still have the ability to radiate significant interference.
Rapid switching of large currents can cause voltage dips and transient spikes on the power supply rails. If one or more supply rails are common to the power and control electronics, then interference can be conducted to the control circuitry.
Good decoupling, and star-point earthing are techniques which should be employed to reduce the effects of conducted interference. The author has also found transformer coupling to drive the MOSFETs very effective at preventing electrical noise from being conducted back to the controller.
Antistatic handling precautions should be used to prevent gate oxide damage when installing MOSFET or IGBT devices.
(I am not an RF Engineer, so I don't fully understand this one. However, Jim Lux kindly offered this excellent explanation:
In a pulsed system, VSWR isn't as big an issue as in a CW system, although it's still an issue.
In a CW system (or, for that matter, something where you put RF power out for many cycles of the RF, so almost any form of transmitter is in this class).
The typical transmitter is designed for a 50 ohm resistive output impedance (at least, the designer dreams of this), then is connected via some sort of transmission line to a load. Hopefully, the load and the line are also 50 ohms, and power flows nicely down the wire. However, if the load impedance is not 50 ohms, then some amount of the power is reflected back from the impedance discontinuity. The reflected power causes several potential problems:
1) the transmitter looks like a load and absorbs it all... If you have a high efficiency transmitter (say, running class C) the power devices (tubes, transistors, or whatever) are dissipating a small fraction of the output power. If your amp were, say, 80% efficient, and you're putting in a kilowatt, normally, the devices dissipate about 200 W, and 800W goes down the line. IF all that 800W gets reflected back, now all of a sudden, your devices are dissipating the full kW.
2) The combination of forward and reflected waves causes standing waves in the transmission line, where the voltage can get quite high at points 1/2 wavelength apart, resulting in breakdown or other bad things. This is essentially the result of the apparent load impedance (at the transmitter) not being what is expected. The transmission line effectively transforms the load impedance to some other value (1/4 waves are neat because Zin*Zout = Zline*Zline... see what happens if Zout goes to zero). If your transmitter is a constant current source, and the load impedance goes higher than expected, then the voltage gets higher than expected, etc. (by the way, for grins, if you have a "tough" RF power source at a few 10's of MHz, you can rig up an open parallel wire transmission line, drive it with an open circuit on the other end (or a short, it doesn't matter) and see the arcing at the voltage peaks along the line...)
In pulsed systems (probably more representative of switchers, tesla coils, etc.) you have a problem with the pulse propagating down the line, hitting the discontinuity in impedance, reflecting back, and summing with the next pulse being sent. Whether the reflected pulse is the same or different polarity depends on the distance, and the relative impedances. If you have several mismatches, you can get lots of pulses moving back and forth which reinforce or cancel as the case may be. (This is a real big problem on commercial power distribution, because the propagation time down the line is a significant fraction of the line frequency period, causing problems when circuit breakers open and close and for lightning strikes... the impulse from the lightning whips on down the line, the protective gap fires, shorting the line to ground, causing another impulse to propagate back, etc...)
All those cool pulse forming networks based on transmission lines work (like Blumlein, Guillemin, etc.) all make use of this idea... charge the line up, short the end, and you get a nice square pulse out the other end. Charge up a whole raft in parallel (say, though HV chokes), short one end in parallel, hook the other ends in series, and you can make a big HV pulse with a lower voltage source.
Folks talk about VSWR because that's what's easy to measure with an RF wattmeter or a bridge. In reality, what's important is the reflection coeffcient, which can be complex, by the way. VSWR can be calculated from the magnitude of the reflection coefficent VSWR = (1+mag(Gamma))/(1-mag(Gamma)), where Gamma is the reflection coefficient. (mag(Gamma) is always in the range 0 to +1).
When a device does fail...
One other thing worth mentioning here, is that MOSFET devices usually fail short-circuit, as opposed to "burning" open-circuit.
Q: Why does a MOSFET always fail short-circuit ?
A: So that the opposing MOSFET is also destroyed of course ! (It must be a Murphy law thing.)
But, seriously MOSFETs usually fail short circuit, causing the opposing device to fail as well, (also short-circuit.) This short-circuits the supply, so current limiting or suitable fuse protection should be employed to prevent damage to the supply. Electrical isolation from the control circuitry is also desirable in these circumstances. Small isolation transformers prevent damaging power from flowing back into the control circuitry when MOSFETs fail with all 3 leads shorted together.
The list provided here is by no means complete, but is meant to represent a starting point for fault finding. If you have discovered another failure mode please let me know so that I can add it to the list.
Back to home page