Accuracy of leakage models: Voltage and temperature de-rating - ForschCV Project Report

Helms, Domenik
In this report, the behavior of a generic 130nm technology under variations of the supply-voltage and temperature is analyzed1 in order to evaluate how accurate a high level leakage model can separate the voltage and temperature dependence to simple de-rating functions. The measurements are based on SPICE3f5[1] simulations using the BSIM4v4.0[2] transistor model with generic transistor data from the default 130nm Berkeley Predictive Technology Model (BPTM)[4]. The result is, that for the generic technology, the voltage and temperature dependence are separable and - in addition - easily predictable from the behavior of the basic transistors (PMOS and NMOS) of this technology. Based on the result of the accuracy analysis, 3 possible leakage macro-models are suggested needing a minimal amount of characterization data. For all technologies, where BPTM files are available, comparisons with real silicon measurements are done with the result, that the SPICE-BSIM-BPTM flow has an average error of just 2% but missing characterization data will usually result in a factor 1.6 overestimation and underestimating of the best suggested model for different technologies.
01 / 2005
Controlling LEAkage power in NanoCMOS SoCs