To achieve both excellent analog switching for training and retention for inference simultaneously, we investigated elevated-temperature (ET) training of Pr0.7Ca0.3MnO3-x (PCMO) electrochemical random access memory (ECRAM). Improved weight update characteristics can be obtained by thermally reduced ionic resistivity of the HfOx electrolyte at ET (413 K). Furthermore, excellent retention characteristics (108 s) were observed at room temperature, which can be explained by enhanced ion storage within the reservoir (or channel) layer via ET training. By adopting ET training on PCMO ECRAM, we can achieve both training and inference accuracy of neural networks (NNs).