Cell voltages are ~60% of true values, and CRC passes

Question

We have a daisy-chain configuration with two LTC6813 chips, each managing 14 cells. The measured cell voltages from the first chip in the daisy chain look like a normal distribution centered around 2.0V, even though the true values are around 3.5V: 
 
 Meanwhile, the 14 cells from the next LTC chip read correctly, as a normal distribution centered at 3.5V: 
 
 Sometimes, all the cells voltage reads will be correct (~3.5V), while other times, all of them will be wrong, but still normally distributed around some seemingly arbitrary mean like 2.0V or 2.3V. 
 I know this isn't some strange bit shift issue because the CRC is passing, so these numbers are truly what the LTC is reporting. I've replicated this issue on multiple boards as well. 
 I checked the STAT registers, too, and the chip with the strange measurements reports a low analog supply (2.9V) while the good one sees 5.0V. Similar low voltage on the digital supply. We checked REF1 with a multimeter and it appeared to be 5.0V though. 
 
 Any idea what could be causing the ADC reads to be incorrect in such a way? 
 
 Thanks, 
 Matthew

MatthewLHR · Accepted Answer

We're 90% sure the issue is on the manufacturing side, considering we hand-manufactured these boards, and a couple of them worked perfectly fine. We noticed that on the failing boards, there were low impedance paths between nets like VREF1 and VREG or other 5V+ ones that were caused by flux stuck between the components. The flux was acting as a kind of invisible conductor, and we were able to resolve this issue by using an ultrasonic PCB cleaner with IPA. We had tried flux remover, but it was not nearly as effective as ultrasonic cleaning. If we were to design the boards again, we would either use machine PCBA, space the components out a little better, or make sure ultrasonic cleaning is part of the manufacturing process.)

Partner Forums

Partner Libraries

Products Mentioned

Cell voltages are ~60% of true values, and CRC passes