Robust interconnect communication capacity algorithm by geometric programming

This paper proposes a new model for interconnect communication capacity in the presence of process variations. Different from previous research works, this model, for the first time, reveals the dependency of interconnect communication capacity on its parasitic parameters. A new method based on Uncertainty Ellipsoid Method (UEM) is applied to optimize the interconnect capacity considering random parameters variations. This new approach incorporates both spatial correlations of intra-die width and parameters variations in the optimization procedure. As is well known, process variation introduces perturbations in the transfer function of interconnect networks. The perturbed transfer function in turn causes variations in the Bit Error Rate (BER). Becoming random, the perturbed BER leads to a changing communication capacity. Based on robust communication theory, we propose a new capacity model which is a function of interconnect geometric parameters. With the help of Geometric Programming (GP) procedure, we use the new model to conduct optimization with regard to the design parameters. Experimental results show that the new model provides less than 7:3% mean square error in capacity prediction comparing with Monte-Carlo method. Based on this bit error value, GP technique is applied to determine the optimal solution, which in return guides the fabrication of interconnects.