Leveraging SIMPLIS to Better Predict Server Platform Power Delivery Performance

Voltage regulator (VR) and power delivery network (PDN) models are critical to the design of server platforms. Inaccurate models can result in over-design, leading to increased costs or an under-design, which can lead to performance deterioration. Tools, such as HSPICE and SIMPLIS, are used to predict the performance of server platform power delivery systems. These tools have different strengths and weaknesses, e.g., HSPICE provides more reliable results in the first voltage droop timing of the waveforms but does not accurately model the behavior of more complex nonlinear VR controllers, and SIMPLIS provides better modeling of the VR response and more reliable results in the third voltage droop timing but can have longer simulation times depending on the complexity of the circuit. A methodology whereby VR models and lab results can be leveraged to improve the overall system level models in both SIMPLIS and HSPICE, through a series of measurement-simulation correlation processes is presented in this paper. Different case studies, including server processor core maximum current (Imax) prediction and Vmin predictability, using this methodology and resulting models, are also presented.