The LPCVD Furnace Model consists of two coupled 1-D models for the axial and radial directions. Full two-dimensional reacting-flow simulations that resolve the geometry of all of the wafers in the furnace would require substantial computing resources. To allow more practical simulation, therefore, the LPCVD Furnace model uses a simplified description of the gas flow. This allows simulation of complex process chemistries relatively quickly on PC’s or on single-processor workstations. The approximations used, however, are reasonable, due to the fact that the subsonic, rarefied, hot gases in LPCVD reactors are highly diffusive. For this reason, the gas composition varies little over distances comparable to the wafer spacing. High resolution of the flow field is therefore usually unnecessary, as it would capture effects only of secondary importance to the deposition process.

The axial 1-D model predicts the transport in the annular region between the wafers and the furnace wall and predicts deposition rates on the furnace wall along the entire length of the reactor. The radial 1-D model describes the transport between the wafers and yields deposition profiles in the radial direction across the surface of the wafers. The model ignores buoyancy, so that the tube can have either a vertical or a horizontal orientation. It does, however, allow for multiple gas injectors, which are treated as radially symmetric gas sources. Temperature distributions within the furnace can be input based on independent knowledge of the reactor, or they can be obtained using the LPCVD Thermal Analyzer.

The LPCVD Furnace Model allows general description of the gas-phase and gas-surface chemistry, as described in the Chemkin Input Manual Input Manual. The LPCVD Furnace Model also accounts for multicomponent gas transport properties, with user options to specify mixture-averaged or multicomponent formulations. These options are described in more detail in both the Chemkin Theory Manual and the Chemkin Input Manual Input Manual. The governing differential equations are solved over the computational grid using the boundary-value steady-state solver, Twopnt.