The phenomenon of time-dependent polarization loss in poled ferroelectric capacitors, also known as retention, represents one of the most important reliability issues for ferroelectric nonvolatile memories. In a number of publications different ways to control retention by varying ferroelectric material composition or processing have been proposed, but no quantitative physical model of this phenomenon is available so far. The goal of the present work is to fill this gap by proposing a retention model that describes the polarization loss as a function of time and temperature. This model considers polarization switching to be driven by the depolarization field occurring in a poled ferroelectric film capacitor. For this purpose the earlier-proposed nucleation-limited switching concept was extended to enable the description of polarization reversal versus temperature, time, and voltage. Temperature-dependent performance of ferroelectric nonvolatile memories is another important issue. The proposed approach employs the same concept for modeling the temperature dependence of polarization switching and retention loss. The theoretical predictions were verified using experimental data measured on Pb(Zr,TiO)(3) ferroelectric film capacitors. Based on the proposed model we formulate a device-oriented algorithm that enables the following important predictions for the ferroelectric film capacitors based on relatively fast and simple tests: (a) Prediction of switching curves characterizing the reversed polarization versus time at different temperatures, based on room-temperature data only. (b) Prediction of retention properties for long periods up to ten years extrapolated from short 24-h test results. (c) Prediction of retention properties at elevated temperatures based on the room-temperature results, or, alternatively, use of the accelerated test at high temperature for characterization of the retention performance at room temperature. (c) 2005 American Institute of Physics.