BQ76920: Highly inaccurate ADC reading

Hi Tom,
The register read description shows individual byte reads. The register values for a voltage should be read sequentially with the same transaction to get an atomic value, note 7.3.1.1 and 7.5.2 in the data sheet. If the cell voltage ADC value is varying around 0x26FF and 0x2700, a separate byte read which crosses the time boundary of the value update could show 0x2600 or 0x27FF, an approximate 200 mV range. This is a common error source at certain voltages in an otherwise apparently OK system, but it is not likely your issue at the voltage stated.
Your schematic is simplified, but shows only cell 5 connected with VC4 at GND. While not specifically stated, the device is not a collection of general purpose ADCs, it is intended for voltage measurement operation with a minimum of 3 cells connected in series with VC5 being approximately the same voltage as BAT. See Figure 8-1 and table 8-1 in the data sheet. The inputs must be operated with sufficient common mode to operate properly. You may have better results if you connect VC5 to the 12V BAT supply with VC4 2.8V below it, or if you apply your single ground-referenced supply at VC1 with respect to VC0, but it would be best to operate the part as intended with input filters similar to figure 8-1 in the data sheet. Resistors can be used to divide the supply voltage to simulate cells as is done on the EVM schematic if power supplies are limited. The part may be happy if REGSRC is 12V, but BAT should approximately follow VC5 (the sum of the cell voltages).

I have implemented the Battery simulation like you suggested, with battery being 16V and each cell being 4V. So  I shorted VC4 to ground. I am now having the problem where every register apart from VC5 is empty. Here is a picture of the I2C communication

Here I am trying to read from VC1. Bytes 2 and 4 are CRC bytes. Why are the data bytes empty. Please help

Hi Tom,
VC4 must not be grounded. See table 8-1 in the data sheet. By grounding VC4 you are shorting out the bottom 3 10k resistors through the balance FET internal between VC4 to VC3. This will put effectively 0V on the lower 3 cells, a negative voltage on cell 4 which violates abs max and probably about 15V on cell 5 which violates its abs max. The cell 5 ADC reads full scale.
Remove VC4 from GND and connect it to VC3, connect VC0 to GND, then check your results. Connecting through filters would be better. There may be a resistive short internally across cell 5 now which will show as an unexpected low voltage on that cell. You may not want to trust this overstressed device.

On further thought cell 5 will see the full 16V rather than 15V mentioned above.

Thank you, I miss understood short as short to ground. I can now read from the other registers, but voltage is still inaccurate, about 2.4V instead of 4V.

Hi Tom,
The part may have been damaged from the wrong connection. Without the filter components the values read from the device should be approximately what you measure across the 10k cell simulation resistors. If there is resistive damage to cell 5 pins it will decrease the top resistor voltage and increase the lower ones. If the ADC inputs are not damaged the sum of the 4 values should be the applied voltage (if all ADCs are in range). I'm glad you were able to make progress, it is probably time to replace the IC.

Ok, I will try with another IC and let you know what happens.

Everything working now with a new IC. Thanks for the help.