Novel non-volatile memory devices are under intense investigation to revolutionize information processing for ultra-energy-efficient implementation of artificial intelligence and machine learning tasks. Ferroelectric memory devices with ultra-low power and fast operation, non-volatile data retention and reliable switching to multiple polarization states promise one such option for memory and synaptic weight elements in neuromorphic hardware. For quick adaptation by industry, complementary metal oxide semiconductor process compatibility is a key criterion that led to huge attention to hafnia-based FE materials. Designing a high endurance hafnia-based FE is crucially important for online training applications in neuromorphic hardware. In this work, we report on the physical origins of fatigue and recovery mechanisms in back-end-of-line compatible ferroelectric Hf