I'm building a capacitor discharger with a thyristor for high-voltage impulse generation into an inductive load. To make the circuit easier to use, I wired the thyristor as a high-side switch. Then I built a floating gate driver circuit around it. The isolated ground is connected to thyristor's Cathode, and the driver output is connected to the Gate. The idea is that, when the thyristor turns on, the potentials of both its cathode and gate rise up to a high value, but the isolated driver always applies a bias to its gate, in spite of the rising potentials.
The isolated high-side thyristor gate driver is built using an isolated DC-DC converter with 3000 Vpk functional isolation, and an isolated MOSFET driver.
The 12 V isolated voltage is generated by an isolated DC/DC converter module with 3000 Vpk functional isolation, Murata NMV0512SAC (26 pF isolation capacitance). The gate driver is generated by an isolated MOSFET gate driver (kind of an abuse, I know, but as long as it can generate a driving pulse within its ratings, it should work), TI UCC5304 (0.5 pF isolation capacitance).
According to an ABB application note, for high-current, high di/dt applications, it's important to generate a momentary surge of high peak current to the gate (as high as 100% of rated absolute maximum gate current) to trigger the thyristor to its fullest degree. Thus, R26 and C41 are used to provide a surge of 2.3 A of peak current. After a few microseconds, R27 takes over, reducing the steady-current gate drive to 0.4 A.
Clamping TVS diode D7 has been removed during the test.
The core of the circuit is shown in the follow schematic.
The actual device prototype is fairly complex and includes multiple subcircuits and an expensive 1000 V thyristor. But for the purpose of this question, I've simplified the circuit to its core and downscaled both voltage and components, and I could still replicate this problem. The capacitor is charged by a high-voltage supply to 130 V, then the thyristor is triggered by the control logic to discharge the capacitor.
The component values of the load are required by design of the pulse-shaping circuit and cannot be changed.
Two tests have been done, I performed the first test with the switch SW1 open, disconnecting the rightmost part of the load (actually by uninstalling R2, there's no physical switch). The next test was performed with SW1 closed, discharging high impulse current into a low-impedance load. Calculations have been done to ensure that both the maximum di/dt and the maximum non-repetitive surge current ratings of the thyristor were not exceeded.
Unfortunately, tests showed that the circuit does not work as expected.
During the first test, the switch SW1 is open. Channel 1 is a 10x oscilloscope probe connected between the left side of R26/R27 (driver side) and DRIVE_GNDC, Channel 2 is a 10x oscilloscope probe connected between the right side of C41 (DRIVE_G) and DRIVE_GNDC, and the measured waveforms behaved as expected - a microsecond voltage spike at the leading edge by the RC differentiator, followed by a steady-state low drive voltage, lasting for 1 second.
During the second test, the switch SW1 is closed and the same measurement setup was repeated. Now the waveform no longer behaves as expected. An undershoot as high as -33 V is generated in this test (look at the change of scale!). What also immediately followed by the test was a noise similar to a "zap" or "bang". After the test, the thyristor is destroyed - it became very leaky and the capacitor can no longer be charged.
This is actually not the worst result. During the process of testing this circuit, an undershoot as high as -100 V was experienced at higher voltages. Here's another example.
If a 15 V bidirectional TVS diode is installed across the gate driver, the circuit can survive much longer. It was tested up to 800 V before the TVS diode was destroyed. But it was entirely accidental, during the design the TVS diode was only meant to handle small transients, not a huge undershoot I just described. Naturally, the TVS diode was eventually destroyed after a couple more tests and failed as a short-circuit. But it seemed to save the thyristor in this case, as the circuit worked again after replacing the diode, thus thyristor was not (completely?) destroyed.
It's also worth noticing that this is already the second prototype on a PCB. The first prototype was constructed on a perfboard with an improvised gate driver but similar in principle. Mysterious driving transistor destructions were also observed, but the problem went away after a high-voltage diode was installed in series with the gate. And that prototype was successfully tested up to 1000 V and 500 A per design without any problems. On second thought (unconfirmed), that prototype may (or may not) have the same problem, and the diode may have blocked this destructive undershoot, and only the initial peak voltage arrived at the thyristor's gate. It may work fine for a while, but probably at the risk of incomplete thyristor turn-on - this may reduce its reliability.
This layout is clearly sub-optimal from the perspective of high-speed electronics. Unfortunately, high-voltage electrical spacing requirements created a lot of layout constraints. Although there's no regulatory requirements for functional isolation, good design practices call for around ~5 mm of electrical spacing for uncoated through-holes at 1 kV for reliability. I found it was difficult to use ground planes at the output side around the thyristor and its load, Thus, all nets are routed as traces.
The output drive and ground of the gate driver was routed to the thyristor as two individual traces. Power and gate driving nets were routed as different traces as Kelvin connections, with no common path. The pulse-shaping network at the output was routed by traces as well.
The oscilloscope probe placement was also sub-optimal. Due to the lack of test points, ground springs are not an option. The probe ground was clipped at the labelled through-hole in the picture. The probe input was clipped at the "Gate" pin of the thyristor.
And this is how the load current is returned to its source.
Update: Cathode, Anode and Gate with Respect to GNDB
Could you show us the voltage waveforms at the thyristor's anode and cathode with respect to GNDB? I have a feeling that the cathode voltage might be rising above the anode voltage. - Jonathan S.
Here are the traces.
Channel 1 (yellow) is Cathode, Channel 2 (blue) is Anode. I don't think there's anything unusual. Sure, there's a small inductive spike at turn-off, but I don't think it has anything to do with the initial gate undervoltage.
Channel 1 (yellow) is Cathode, Channel 2 (blue) is Gate. This still clearly shows the gate undervoltage.
What's going on? What is responsible for the massive undershoot when the isolated driver is supposed to maintain a constant bias voltage on the thyristor's gate?
My current hypothesis is that high-voltage thyristor has an internal resistor across the gate and cathode (10 Ω, as I already measured), thus, the isolation of the driver is essentially breached by this low-value resistor, making the assumption of isolation no longer valid. Is it the correct culprit? If so, how can an isolated high-side thyristor driver be constructed? If not, what is the real problem that I've missed?