EVALSTDRIVE601 Failure and IGBT Destruction at 600V

[christopherdacc]

Hello,

I am using the EVALSTDRIVE601 evaluation board (featuring the STDRIVE601) to try to generate a controllable three-phase signal. I wired the evaluation board exactly as described in the datasheet:

• HV and PGND (J1 connectors): Connected to the DC voltage source.

• SGND and VCC (J3 connectors): Connected to the ground and output of a buck converter that steps down 24V to 12V.

• VDD: Connected to the 3.3V of the MCU.

• Pin 1 (Fault): Connected to the MCU to read this value and detect fault conditions.

• Pins 3, 5, 7, 9, 11, and 13: Connected to the SPWM signals. I am generating these using a Compute Module 4 with a switching frequency of 10kHz and a modulating frequency of 50Hz.

• Pin 23 (Enable): Connected to the MCU, providing 3.3V to activate the driver.

• Pin 2: Connected to the GND of the MCU.

• Pin 28: Connected to the 3.3V of the MCU.

Test 1: 24V DC with LC Filters I initially tested the board using an external 24V DC power supply. At the output (J2 connectors), I wired an LC filter (L = 680mH, C = 47µF) on each of the three phases. This test worked perfectly, and I was able to generate a controllable three-phase AC signal.

Test 2: 600V DC without Load Next, I tested the board at 600V DC. I started with nothing connected to the J2 output connectors. When I activated the Enable pin with a duty cycle of 10%, a signal resembling a sine wave was generated, which I measured using a voltage probe connected to an oscilloscope between phase 1 and 2 of the J2 connectors. However, this signal was not stable. After a couple of seconds, the signal was completely attenuated, and a fault appeared on Pin 1. This correctly disabled the Enable pin and turned off the SPWM signals (which are inverted, as specified in the datasheet).

Test 3: 600V DC with LC Filters and Board Failure I then tried testing at 600V DC with the LC filters connected to the three phases. This time, when I activated the Enable pin with a 10% duty cycle, I immediately read a fault on Pin 1 and no signal was generated.

I retested while observing the VCC pin of the J3 connectors to check if an undervoltage condition (VCC UVLO) was being detected, but the VCC remained perfectly stable at 12V. While investigating this error condition—and importantly, while the Enable pin was inactive—the 6 IGBTs integrated into the evaluation board brutally exploded, destroying the board as seen in the attached picture.

If possible, could I get some assistance regarding this issue? I am available for any clarifications. If you notice any mistakes I made with my connections or my testing procedure, please let me know.

Thank you in advance for your time and help.

[GMA]

Hello @christopherdacc,

If the first probe, connected between phase 1 and phase 2 of the J2 connectors, remained attached, and another probe, connected between the VCC and GND pins of the J3 connectors to observe the VCC pin, was also connected, a short circuit could occur through the oscilloscope ground.

[christopherdacc]

Thank you for your response.

The two probes are differential and isolated and connected to two different oscilloscopes, and each oscilloscope is powered by an isolation transformer. This means the two probes are not connected to each other at all.

[christopherdacc]

For more information, after cleaning the driver and analyzing it under a microscope, the electric arc occurred on the R6 resistor. We can see that the resistor is completely burned, and the bottom side of the PCB is also burned at this location.

The command board (Raspberry Pi Compute Module 4, from Raspberry Pi Foundation) is no longer functional. It is used to generate the PWM, it reads the Fault pin and generates the 3.3V for the EN pin.

The other damaged parts are:

• the shunt resistor section (red selection in the UM2748 datasheet),

• the power bridge section (dark blue selection in the UM2748 datasheet),

• the SGND screw on the J3 connector, which is also burned (not visible in the picture).

Furthermore, the LC filter used in the HV test has the following values:

• L = 12 mH

• C = 4.7 µF

Update: After re-reading the datasheet of the EVALSTDRIVE601 from STMicroelectronics, it is stated:
"High voltage rail up to 600 VDC (limited to 400 VDC due to onboard components)."
This means that we cannot connect a voltage source generating more than 400 VDC.

However, why, for instance, was the C34 capacitor not damaged first, since it is rated for 450 V and had approximately 650 V across it?

The analysis performed after cleaning the board shows that a high current passed through the common ground (SGND and PGND). The remaining question is: where did this current come from?

[RiPeloso]

Dear Christopher,

Indeed, the board is not made to handle 600 V, as you can see from the bulk capacitor rated 450V. Probably, in your first test without load something has already been damaged. Please also notice that you were working at the absolute maximum rating of the STDRIVE601, for normal operation we suggest working below the recommended operating condition (580 V). Is 600 V strictly needed in your application?

As the board show many different damaged parts, it is not easy to find the root cause and reconstruct the chain of events that occurred. About the arc on R6, a possible explanation is that a shoot-through on IGBTs has briefly connected the 600V supply to PGND, blowing up the resistor and igniting the spark.

Thanks and best regards,
Riccardo

[AScha.3 · Accepted solution]

>However, why, for instance, was the C34 capacitor not damaged first, since it is rated for 450 V and had approximately 650 V across it?

Probably just because : a cap, rated for 450V is typically tested at higher voltage, maybe +25% -> 600V , to have <xx uA leakage ; so it might survive 650V or more for short time. 

BUT the IGBTs here are 650V type , so having 650V here...might trigger one, then...yes , you know now.

So this board 600V driver, 650V IGBTs is good up to mains voltage , 320VDC (400V peak).

+

Lesson to learn : usually and from experience  650V switches MOS/IGBT are used up to 350V rail vdc;

at 500...700V rail voltage always 1200V types are used. You seen the effects , not keeping to "old rules" .

(+ When testing with some power, always wear glasses and earplugs - at least if your nervous system is somewhat sensitive, like mine. )

+

You cannot put a resonant load to a inverter - it might find its resonance triggered and ...BUUUMM ! 

If having a load , use just a coil , same as a motor would be here, with high current capability (better > 200% of highest current to run); if this choke coil saturates ... BUOM ! (or you need a good working current limit ! )

+

blown R6 : if you had useful connection power rail+ / - and low level signals and GND , R6 cannot blow up, as there has to be no way from power hi-rail to controller/signal-GND. 

power and signal GND are just connected in one point : R6 , and no current has to flow here !

Otherwise your GND connections anyway bad .

[christopherdacc]

Thank you for your response. I will make the necessary changes to the new EVALSTDRIVE601 I purchased and apply your advice.