Thermal optimization of electronic systems using Design of Experiments based on numerical inputs

Design optimization of realistic electronic systems is often found to be prohibitive in today's marketplace because of time constraints. Techniques, such as one-variable at a time (OFAT), that prove useful for simple designs, often fail for real-world applications. This paper describes a Design of Experiments (DoE) methodology used in conjunction with numerical inputs from a computational fluid dynamics (CFD) program to demonstrate an efficient method of optimization. As an example, the technique is used to optimize multiple aspects of the thermal performance of a next-generation network server. The outcome was that empirical prototype cycles were reduced from four to two and analytical experimentation cycles were dramatically reduced while producing a far more mature and viable cooling solution. Upon completion of the DoE, the only alterations to the design were the few minor adjustments and settings where the analytical models diverged from actual physical performance. These were rapidly finalized with the initial physical prototype, again using DoE techniques, and the final enclosure cooling architecture was completed.