Dear Power by Linear,
I've designed the LTC3649 into a medical device whose tip is connected by a cable that could be two meters long (by my estimate - my customer cannot be more exact.) A footswitch enables a relay which connects or disconnects the load, and the load itself is fairly inductive. The output voltage is user adjustable, from near-zero volts to a maximum of 13.0VDC. That's why I chose the LTC3649, in fact - for the near zero Vout capability. Here's what I can share of the schematic:
Two prototypes have failed as I tested switching the load on and off via the relay with Vout set to its maximum of 13V, with a 1A load. This was a 1A CC, using an electronic load. When things go wrong, the system loses the ability to set the output voltage below around 6VDC with no load. If I load the output, then Vout will go down to near-zero. Also, when the relay is off (therefore always no-load for the 3649), after the damage the Vout oscillates if set higher than about 11V.
Ch.1 = ISET pin voltage, Ch.2 = VoPre (output voltage prior to relay)
I expected a lot of transients when turning the relay on and off - not surprisingly it's worse with higher output current. Here's a shot of the relay turning on into the 1A CC load:
Ch.1 = ISET pin voltage, Ch.2 = VoPre
However, I don't see anything potentially fatal in these images. After one prototype died, I added a 30V, 1A schottky from VoPre to GND right at the relay pins to protect against negative transients, but the prototype with that schottky died in the same way as the first.
Perhaps ISET needs negative transient protection, too? A simple impedance check seems to indicate that the ISET pin is OK. What other pins might be sensitive?
In the schematic you don't have a compensation network for Ith pin. This network is necessary for loop stability. you can use LTPowerCAD for a reasonable value.
I have a single 100 nF cap from the ITH pin to GND. Normally I just start with that value, test the loop, and adjust. But when I tested, the transient response and gain/phase plots looked so good that I didn't change it. The Vout of this circuit can change from about 1.0V up to 13.5V, so I took some more plots. The input is a 30W AC-DC PSU, so the max load current changes. Here are some plots. Note - my network analyzer adds 180º to the phase so that phase margin can be read directly as the difference between zero degrees and the value on the graph. VIN is always 17.0VDC.
This is Vout = 1.0V with a 2A load, the strange response at low frequency is likely measurement error.
This is Vout = 3.3V with a 3A load.
This is Vo = 10V with a 2.5A load
This is 13.5V with no load. After taking this plot I applied a 2A load and the Vout collapses to around 8V, and when I disabled the load (an electronic load) the LTC3649 failed - it shorted out the input supply. I removed the connection and the AC-DC supply was OK, so I know the LTC3649 is what died.
So, overall, the failures occur when Vout is near the maximum and there are load transients, either by enabling and disabling the load or by turning the output on and off with the mechanical relay.
If you think a different control loop compensation would help, I can try with values calculated by LTCPowerCAD, but the bandwidth and phase margin are good enough for me. I still think that one or more pins of the 3649 are being pulled negative - maybe too high in the positive, but given the 60V abs max, that seems unlikely.