Techniques are provided for reducing the effects of short-term charge loss while programming charge-trapping memory cells. Short-term charge loss can result in a downshift and widening of a threshold voltage distribution. A programming operation includes a rough programming pass in which memory cells are programmed close to a final threshold voltage distribution, for each target data state. Subsequently, a negative voltage is applied to control gates of the memory cells. Subsequently, a final programming pass is performed in which the memory cells are programmed to the final threshold voltage distribution. Since the negative voltage accelerates charge loss, there is reduced charge loss after the final programming pass. The rough programming pass can use incremental step pulse programming for the lowest target data state to obtain information regarding programming speed. An initial program voltage in the final programming pass can be set based on the programming speed.