A Fortune 500 semiconductor company was using a thermal simulation as part of a chip/housing design process. The high fidelity simulation was too computationally intensive to be useful for optimization or competitively fast design cycles.
Using SmartUQ’s adaptive design technology, the simulation was run a minimum number of times to generate a sufficiently accurate emulator. This process works by using an existing DOE to build an emulator and then using that emulator to choose where new simulations are run. This process is repeated until the emulator is accurate enough for use in further analysis.
An initial hybrid DOE consisting of a combination of a fraction factorial design and a Latin Hypercube Design was used to train the first emulator. Three rounds of adaptive DOEs, each adding 5 test points, were used to increase the emulator accuracy. The 51 simulation evaluations were required to meet accuracy targets compared to the next best method tested, which required 200 simulation evaluations. SmartUQ yielded a 75% reduction while improving performance.
