We report in this work a Si-nanocrystal (Si-nc) MOS basic cell which shows, at the same time, fast writing times and long charge retention. This has been achieved by optimizing a structure reported previously that exhibited excellent retention characteristics. For the new structure, 15 keV Si ions have been implanted in a 40 nm thick oxide at high doses in order to obtain Si excess ranging from 10 to 20 atomic % at projected range (25nm). An annealing step at 1100 °C has been performed to precipitate the nanocrystals. We show that there is a Si excess compromise (density depth profile of Si-nc) in which write times are improved by at least 3 orders of magnitude (to the submillisecond range) while still maintaining a virtually infinite retention time. Such behavior has been correlated with structural characterization by EFTEM, which reveals a control oxide completely free of Si clusters and thick enough (11 nm) to prevent tunnelling from/to the gate electrode. The Si-ncs are located around the projected range and show a mean size of 2.7 ± 0.2 nm. The tunnel oxide is not completely free of Si-nc or clusters, as observed by EFTEM, but there is a significant reduction in mean size and density when approaching to the p-type substrate. We believe that these small Si-nc or clusters in the tunnel oxide play an important role in improving the performance of the devices. For charging (writing), when a gate bias is applied to the structure, these clusters assist like traps when tunneling to the central region. However, when the Si-nc are already charged, these nanoclusters do not similarly enhance the discharge process because they have larger band-gaps (due to quantum confinement) than the bigger Si-nc in the center of the layer, and therefore act as an insulating material. This simple model, based on the correlation of the most important electrical memory parameters and the structural information, has allowed us to engineer the implantation dose as a technological parameter when a trade-off between write and retention times is required. For our samples, this dose is about of 15% Si excess. Finally, endurance tests have been performed, showing a completely flat and stable programming window after 106 Write/Erase programming pulses.