Semiconductor Process Simulation - Process Simulation Methods

Process Simulation Methods

The process steps most often associated with process simulation are ion implantation, annealing (diffusion and dopant activation), etch, deposition, oxidation, and epitaxy. Other common steps include chemical-mechanical planarization (CMP), silicidation, and reflow.

All commercial process simulators use a combination of the finite element analysis (FE) and/or finite volume methods (FV) methods. A complete description of FE/FV method is out of the scope of this article but there are many fine books which describe the topic thoroughly. However, it is important to discuss requirements for process simulation for achieving accurate results. These requirements are based on the same requirements as generic to FE/FV techniques with an additional difficulty coming from the changes in the geometry during the simulated fabrication of the device. Process simulation uses an FE/FV mesh to compute and store the dopant and stress profiles. Each geometrical change in the simulation domain requires a new mesh which fits to the new boundaries. As will be described below, the large number of geometry modifying steps involved and the nature of process simulation where each step depends on the cumulative results of all previous steps, make process simulation an especially challenging application of the FE/FV technique.

One of the most important results of process simulation is the dopant profile after processing. The accuracy of the profile strongly depends on maintaining a proper density of mesh points at any time during the simulation. The density of points should be just enough to resolve all dopant and defect profiles but not more because the computation expense of solving the diffusion equations increases with the number of mesh points. A typical full flow CMOS process simulation can have more than 50 mesh changes and the number of mesh changes can increase dramatically if adaptive meshing is performed. For each mesh change, interpolation is used to obtain data values on the new mesh. It is important to manage the mesh changes in such a way to avoid accuracy degradation due to interpolation error. The easiest way to do this is to always keep points once they are introduced into the mesh, but this has the drawback of producing very many mesh points which can be computationally expensive. Maintaining a balance between interpolation error, computational expense, and minimization of required user input is important for obtaining accurate results with a minimum of computational expense. This is especially true when simulating devices in 3D. Without careful placement of mesh either the accuracy will suffer unacceptably, or the computational expense will be too great to be useful. Process simulation tools so far have had limited success in completely automating mesh adaptation such that no user intervention is required. This places a requirement of the user to understand meshing and how it affects simulation accuracy and run time and the burdens the user to track mesh changes during the simulation to ensure proper mesh is maintained.

One of the most important uses of TCAD tools is to explore new device technology where many exploratory simulations are performed in order to give the device designer a better understanding of possible benefits as well as drawbacks of a given technology. This use case demands sequential simulations with some analysis in between. In order to be useful, many simulation cycles must be run within the time allotted for exploration, putting a high priority on minimization of simulation run time. Currently, full flow standard CMOS simulations are most often accomplished with a combination of 1D and 2D simulation and take less than a few hours on a 2.6 GHz Pentium 4. To perform these simulations in 3D (from gate formation on) would take a minimum of 24 hours for minimum accuracy simulation. Most of the information desired from TCAD simulations can be extracted from the simplification that the device can be treated uniformly in depth (i.e. a 2D simulation). To include the effects device shape along the depth or to investigate implant shadowing, 3D simulations must be performed.