Recently I found some time to have a closer look at the peculiar behavior of my AT8612A - I reported that before but didn't get into too much detail. Actually, I found the cause and I was quite amazed that this slipped through Applent's QC... But let's first have a look what was strange: The first photo shows the load, set to CC mode with a current of 1A dialed in and switched off and nothing connected to it. Everything's fine, the display readings are as one would expect.
Now, all I did was activating the load's input, still nothing is connected to it and nothing else has changed to the previous photo: Mysteriously, the load is reading a current of 212mA now. I did some basic reverse engineering of the passbank circuitry which consists of eight individual current sinks, all controlled by the same control voltage that is generated by the circuitry around the main shunt. This configuration makes sure that the power dissipation is spread evenly across all eight passbank transistors. See the photo of my ugly whiteboard drawing of one channel's simplified schematic.
Some probing around didn't show anything special initially -- except gate voltage at the mosfets at the second glance... It was around 9V with the load enabled on EVERY transistor. Moreover, before the gate resistor(s) I measured something like 10.8V which means that the gate-source protection clamps started to conduct. And that's the cause of the current being displayed when the load is switched on. As soon as the load is actually controlling the current, the gate-source voltage of the passbank transistors will be lowered to a value slightly above the individual transistors threshold voltage -- considerably below the 9V that the clamps become conductive at. That's why the load appeared to work normally when it was controlling an external current. As soon as the external current is lowered below the preset value, the gate voltage is increased and again the clamps become conductive, which results in a current reading offset by the 200 odd milliamps.
The zeners of the gate voltage clamps have apparently been 8.2V type while they probably should have been rated at 12V. Seems like somebody installed the wrong feeder on the p&p machine... Whatsoever, I replaced all 16 zeners with 12V types and the problem was gone.
Since I had to remove the main PCB for this job, I found some ugly details in an otherwise neat design -- probably the "Danger" sign had been placed there to indicate the tiny creepage distance around the mains switch area
- especially to the switch's metal mounting bracket. The pitch of the nearby connectors is 2.54mm for comparison. I'm quite surprised of this since in the other areas where mains power is routed across the PCB, the creepage clearances are ample or even isolation slots are milled through the FR4. Apparently, this ugly detail didn't cause any trouble yet.
I had been in touch with Vera who's really very kind and helpful. I discussed the calibration issue and the benefit of being able to do the job myself and she was so kind to send me a short instruction sheet (see attached PDF - slightly edited by me). With the help of this, I've been able to get the CC mode accurate again. It seems the voltage slope calibration requires a "proper" lab supply with at least 250V output voltage which I haven't got. I experimented with an electrophoresis supply but this is way too picky and shuts the output down all the time. Calibration of the current slope actually requires a power supply that outputs more than 30A - I used a 12V 1kW server PSU for that job and it worked a treat. I'll add the information about how to calibrate the voltage mode and what kind of PSU will do once I succeed.
At least there's some hope for all those of us who messed up the calibration or just want to do the job themselves. I'm most impressed by the amount of cooperation at Applent and can only say "Thank you, Vera".
All the best,
Thomas