SIMULATION OF RAPID THERMAL PROCESSING IN A DISTRIBUTED COMPUTING ENVIRONMENT

Abstract

Rapid thermal processing (RTP) has become a key technology in the fabrication of advanced semiconductor devices. As wafers get larger and chip dimensions become smaller, the understanding of the highly coupled physics, such as radiative heat transfer, transient fluid flow, heat transfer, and chemical reactions through numerical modeling using high-performance computing, is the key to the design, optimization, and control of RTP reactors. In this study, an accurate and efficient simulation tool for RTP in a distributed computing environment is developed by implementing various new models and algorithms. Thegoverning equations for highly coupled and transient transport phenomena inRTP are solved by anunstructured finite volume method (FVM). Surface radiative heat transfer is the most dominant mode of heat transfer in RTP and it is modeled by the modified discrete transfer method (MDTM). The radiative properties on the patterned wafer are quite different from those on the bare silicon and they are predicted by the matrix method. To enhance thecomputationefficiency, anefficient parallelalgorithmis implemented in the solution procedure. Data communication among the processors is carried out by the Parallel Virtual Machine (PVM). To evaluate the present simulation tool, an actual commercial RTP reactor is investigated under various conditions. The accuracy of the present model is validated through the comparisons of wafer temperature profile between different models for steady state andtransient flow andheat transfer cases. To demonstrate the importance of the pattern effects in RTP systems, a transient case containing the patterned wafer is investigated. The temperature profile and its uniformity for the patterned wafer are found to be quite different from those for the unpatterned wafer. To examine the performance on parallel computation, the previous transient case is studied with different processor numbers. As the processor number increases, the computationtime is seen to reduce; however, the parallel performance is seen to degrade A larger solution iteration number and higher communication overhead are believed to be the major reasons for the degradation of the parallel performance. The present case studies indicate that the simulation tool developed in this study can be used to systematically investigate various effects in RTP systems because of its high accuracy and efficiency.