Reservoir computing based on the ferroelectric FET (FeFET) technology offers a computational platform for information processing of time-series data with a low computational cost by leveraging the nonlinear polarization/charge dynamics. While hafnia/Si FeFETs for a memory application encounter critical challenges on the poor endurance caused by polarization-induced interface degradation, the reservoir computing operation of hafnia/Si FeFETs exhibits a high tolerance to the interface degradation particularly when the system is frequently re-trained. The degradation tolerance can be attributed to the polarization dynamics not being canceled out by the trap dynamics in the time domain during operation of reservoir computing. A degradation-robust FeFET reservoir can be trained to classify spoken-digit dataset, where more than 10⁴ of bipolar voltage inputs were applied during data processing, with high classification accuracy.