We are using a TXS0104E to level shift some signals from 3.3V to 1.8V. They are mix of push pull and open drain. The signal of interest is only ever one direction and is constantly driven (high or low) output from an Microchip AVR microcontroller. The other side is routed directly to an IO pin of an Intel FPGA. The signal is a SPI CS_L that activates for approximately 400uS every 5ms. What we believe we are observing is the part is incorrectly detecting a low on the FPGA (1.8V) side which in turn drives back to the AVR, which then contends with the TXS0104E and the circuit oscillates a few times before becoming stable again. Adding a weak pull up internally to the FPGA (25Kohm) helps reduce the frequency of occurrence and degree of oscillation but doesn't eliminate it completely.