Using statistical analysis of optimal experimental designs, the dependence of oxide thickness, and oxide thickness variation within a wafer and wafer-to-wafer on process variables was studied in rapid-thermal processing systems that differed in chamber configuration and construction, incoherent light source, and pyrometers used for temperature measurement. Mechanisms for oxide growth and oxide thickness variation in rapid-thermal oxidation are discussed. In general, the percent standard deviations of oxide thickness within a wafer and wafer-to-wafer were less than 10%, but in unmodified linear lamp array systems, too large for manufacturing, which requires that three times the percent standard deviation is less than 10% of the mean oxide thickness. A thermally induced stress, lamp configuration, and convective cooling affected the oxide thickness variation within a wafer. Wafer-to-wafer oxide thickness variation depended on the material of chamber construction, quartz or metal, and was related to residual heating for longer oxidations. The process latitude and uniformity improved between a manual system and a production model that had more advanced features, such as a slip ring which enhanced temperature uniformity, for the same processing conditions, the oxide thickness was different for different systems. The differences were caused by temperature error and a photonic component to rapid-thermal oxidation. Analysis of empirical oxide thickness models revealed a silicon orientation effect and a mechanism related to oxidant transport, that was common to rapid-thermal oxidation in different systems. Before rapid-thermal oxidation becomes a manufacturable process, the general rapid-thermal processing problems of temperature measurement and uniformity must be solved. In addition, the design specifics of rapid-thermal processors have a major effect on the oxide thickness and oxide thickness variation. © 1992 IEEE