Binary collision simulations of high-fluence 1 keV Si + ion implantation into 8 nm thick SiO 2 films on (001)Si were combined with kinetic Monte Carlo simulations of Si nanocrystal (NC) formation by phase separation during annealing. For nonvolatile memory applications, these simulations help to control size and location of NCs. For low concentrations of implanted Si, NCs form via nucleation, growth and Ostwald ripening, whereas for high concentrations Si separates by spinodal decomposition. In both regimes, NCs form above a thin NC free oxide layer at the SiO 2 /Si interface. This, self-adjusted layer has just a thickness appropriate for NC charging by direct electron tunneling. Only in the nucleation regime the width of the tunneling oxide and the mean NC diameter remain constant during a long annealing period. This behavior originates from the competition of Ostwald ripening and Si loss to the Si/SiO 2 interface. The process simulations predict that, for nonvolatile memories, the technological demands on NC synthesis are fulfilled best in the nucleation regime.Recently, nonvolatile memory concepts based on nanocrystals (NCs) embedded in the gate oxide of MOS transistors have attracted much interest [1]. For that aim, NCs have been synthesized by a variety of techniques like chemical vapor deposition [2], ion implantation [3,4], and Si aerosol deposition [5]. Ion implantation followed by thermally activated precipitation of the implanted impurity atoms is most compatible with current silicon technology. By low-energy Si + ion implantation into thin SiO 2 layers on (001)Si, NCs of Si were formed a few nanometers above the Si/SiO 2 interface [3]. This allows charging of the NCs by direct electron tunneling, which is a prerequisite for high endurance and low operation voltages [6]. Further optimization of location and size of ion beam synthesized NCs for memory application requires a deeper understanding of the mechanisms involved, which determine (i) the built-up of Si supersaturation by high-fluence ion implantation and (ii) NC formation by phase separation. For that aim, process simulations were divided into two steps. The Si implantation was studied using the binary collision code TRIDYN [7], which includes dynamic target changes. The phase separation of Si from SiO 2 during subsequent annealing has been simulated with a kinetic lattice Monte-Carlo code, which describes the thermally activated processes.The TRIDYN depth profiles are shown in Fig. 1 for 1 keV Si + ion implantation into SiO 2 . TRIDYN takes into account dynamic target changes due to ion deposition, ion erosion and ion beam mixing. The input parameters required by the simulation include the displacement and surface binding energies of target atoms. The displacement energies E d for both, Si and O, were assumed to be 8 eV. This value proved to yield satisfactory agreement between earlier TRIDYN simulations of ion mixing and experiments [8] and is also consistent with the choice of Sigmund and Gras-Marti [9] in their theoretical treatment of ion m...