Automatic differentiation for gradient-based optimization of radiatively heated microelectronics manufacturing equipment
C.D. Moen, Paul A. Spence, Juan C Meza, Todd D. Plantenga · 6th Symposium on Multidisciplinary Analysis and Optimization · 1996
Automatic differentiation is applied to the optimal design of microelectronic manufacturing equipment. The performance of nonlinear, least-squares optimization methods is compared between numerical and analytical gradient approaches. The optimization calculations are performed by running large finite-element codes in an object-oriented optimization environment. The Adifor automatic differentiation tool is used to generate analytic derivatives for the finite-element codes. The performance results support previous observations that automatic differentiation becomes beneficial as the number of optimization parameters increases. The increase in speed, relative to numerical differences, has a limited value and results are reported for two different analysis codes.