Development and Implementation of Large Area, Economical Rotating Disk Reactor Technology for Metalorganic Chemical Vapor Deposition

The vertical, high speed, rotating disk reactor (RDR) has, in recent years, found broad application in the Metalorganic Chemical Vapor Deposition of a variety of material systems. These applications include epitaxial films of III-V and II-VI compound semiconductors, oxides (such as YBCO superconductors/ferroelectrics and SiO2, amongst others), Group IV materials (such as diamond and SiC), and metals (such as copper and tungsten). As production of these material systems increases, so too does the need for economical, high yield equipment capable of producing these materials with high levels of uniformity and repeatability. We have used computational fluid dynamic modeling to investigate the complex flow and thermal dynamics required for scaling existing RDRs (as large as a 7.25″ diameter disk handling up to 3×3″ wafers) to larger dimensions (11″ and 12″ diameter disks for multiple 4″ and 15.5″ diameter disk for 3×6″ wafers). The scaling parameters predicted by the modeling codes are reviewed and correlate well with experimental results. Materials results on GaAs films using TBAs, TMGa, and TMA1 for the 11″ diameter system routinely demonstrate within wafer thickness uniformities of <1.1% for 3×4″ wafers, as well as for 6″ or 8″ diameters, wafer to wafer uniformities <1% and run to run repeatabilities within 1%. These results are verified by SEM analysis, as well as with GaAs/AJGaAs Bragg reflectors. The excellent results on the 11″ and 15.5″ diameter platters combined with modeling indicated that 4×4″ wafers on a 12″ diameter platter would produce ideal films which, indeed, is the case. The 11″ diameter results have been surpassed, demonstrating <0.9% for >9″ diameters (4×4″ wafers) on a 12″ diameter susceptor. With high reactant efficiencies (>3 6%), short cycle times between growths using the loadlock, and minimal maintenance requirements, the costs per wafer in a cost of ownership model are found to be dramatically less than in competitive technologies.