AD5254 power-on problem


I'm facing a problem with digipot AD5254 on my board where I use them as a part of programmable amplifiers in op amp feedback path. Usually during power-on sequence (after turning power supply on) digipot wrongly writes values 0 or 255 in some or all of RDAC registers, even the EEPROM registers are previously programmed to other values between 0 and 255. After I have assured that value between 0 and 255 is actually written to EEPROM, by sending read command and getting expected value, and after switching power supply OFF, and then again ON (after arbitrary delay), digipot doesn't set in proper position according to the written EEPROM value. I have figure that out by giving some input reference voltage to amplifiers, and by observing its output. In these problematic situations amplifier's output is either strictly 0 regardless of the input voltage or it is rising very fast with the increase of input voltage which I would expect if the written EEPROM/RDAC value was highest (255). So, I can't rely on the written EEPROM values in digipot which should set the gains for current and voltage measurement circuit in my application (interface board for three-phase inverter).

Digipot is powered with bipolar power supply VDD=+2.5V and VSS=-2.5V. It communicates with FTDI chip (FT4232H port B set in I2C mode) through voltage translator IC TXS0102. This is because FTDI chip is supplied with +3.3V and digipot with VDD=+2.5V. ~WP pin is permanently tied to +2.5V because it is not used. SCL and SDA line are externally pulled-up on +3.3V side (also, TXS0102 includes internal pull-up resistors on both +3.3V and +2.5V sides). Address lines AD0 and AD1 are connected to ground AGND. Following figure shows this part of the circuit.

Digipot (U5) is used as part of the amplifier circuit in following manner (digipot channels 0, 1 and 2):

Firstly I have checked if there are overshoots on power supply lines, and everything is ok. Something that could be a problem but I'm not sure, is that there is sometime a overshoot on digipot SDA line (at +2.5V side) after turning power supply on, looking like this (yellow line CH1 is signal on digipot SDA pin, and blue line CH2 is +2.5V signal on VDD pin):

I want to emphasize that I have more I2C devices on I2C bus at +3.3V side, but only the digipot on +2.5V I2C bus. In problematic situations, all of I2C devices are readable/writable on +3.3V side, but digipot is not.

Did somebody face similar problem with AD5254 digipot, and is willing to share his/her experience?