ADC7172-2 randomly converts 0x000000 or 0xFFFFFF


We are using our AD7172-2 AD converter in our prototype board ( AVDD1 = +5V, AVDD2 = +5V, IOVDD=+3.3V, AVSS = 0V) single ended mode (+=CNH0, GND=CHN1) with continuous conversion, running it from internal clock and internal reference voltage. The AD7172-2 is driven from the Silicon Labs’ Giant Gecko 32-bit ARM Cortex M4 MCU starter kit using its’ USART0 on 12MHz, in SPI3 mode. The problem is that our long-term stability test is failed. We have an error counter set to see how many times we have 0x000000 or 0xFFFFFF conversion result. (We have 660mVRMS  sine voltage, shifted to +1.25VDC, with +2.5V reference voltage, so the conversion result would never be 0x0000000 nor 0xFFFFFF) This counter and the elapsed time are printed on a small LCD screen of starter kit with the elapsed time and a small scope which shows the last 512 samples. Sometimes the AD converter works for 30-35 minutes without error, sometimes the error code is incremented within 1 minute. Once it happens, then we have to reset (re-initialize) the ADC7172-2 chip again in order to work for another XXX minutes. If we don’t reset the chip it converts forever 0x000000 or 0xFFFFFF. (We checked this state with scope) We have two prototype boards, both have this bug. Can you help? Thanks, Louis. (see attachment for the details)