I am having a problem with leakage currents through a circuit board.
A product I am working on has a dc-dc converter that converts 12V from a battery to 350Vdc. The two power buses are mostly isolated, where the only connection between them is a differential amplifier with a
1MOhm input impedance. The input impedance of the amplifier centers the high volt output relative to the battery ground. So relative to battery ground the high volt supply is +/- 175V.I recently added a safety circuit to detect leakage currents between the high volt lines and battery ground. This looks at the voltages on the high volt bus relative to battery ground, and if they are not centered, there is a leakage current. This way the system can detect a fault.
This worked great through development, but now in production I am seeing a 3-5% fallout due to this fault. The boards that pass are dead nuts on. There is no discernible leakage current to the limits of my measurement equipment. The boards that fail show an obvious leakage current, the impedance of the defect is within 100kOhm to a couple MOhm. This only happens with high voltage, a regular DVM shows an open circuit. This appears to be due to a defect within the circuit board itself. When I look at it with a scope I can see the voltages moving around, maybe due to partial discharges. On a couple of the boards I have watched the fault clear itself, and have not been able to get it to reappear.
I am looking for high voltage specifications for PCBs. The design follows the IPC-2221A guide lines for creepage and clearance. Are there other material specs i should be calling out?
Does anybody have an idea of what the failure mechanism is? If a board shows a fault that clears itself, presumably burning out whatever contamination is causing the fault, is it more likely to fail in the future? Are there processing issues that could cause a large leakage current? I have seen a board go through a lead free soldering process and come out charred, but I am pretty sure that is not what is going on here.
Thanks,
Ethan