A statistical approach to the design and modeling of high speed transmission line networks in high performance high volume systems

The system interconnect and packaging designs around high performance CPUs are becoming the major bottleneck in the overall performance of high speed, high volume systems. Given the reduced signal swings on the external buses the required modeling accuracy is now on the order of tens of picoseconds and tens of millivolts. As such many of the electromagnetic modeling techniques developed for high end supercomputer systems are quickly finding their ways into the PC industry. On the other hand, unlike high end systems, the key to the success of any high volume design is its cost effectiveness. In this paper, a Monte Carlo design and modeling approach to high volume high speed transmission line systems is presented. The sensitivity of the accuracy of any design parameters to the overall system performance is evaluated in terms of its effects on the statistical distribution of the optimized parameters or yield. It is shown that the traditional “worst case” design can be very misleading and under-estimates the worst case. We also show that the uncontrolled nature of the design makes it virtually impossible to find a “representative” signal and therefore detailed analysis of any given signal is only useful in determining the accuracy of the simulations or the phenomena that effect the results, but not the robustness of the entire system