Efficient integrated circuit

摘要

Novel algorithms for computing the responses of circuits to multi-tone excitations. The new algorithms are efficient and robust for large, strongly nonlinear circuits excited by multi-tone (quasi-periodic or envelope-modulated) signals. Hence they are particularly useful for integrated RF applications. The multivariate representation captures features produced by strong nonlinearities (such as spikes or pulses) much more compactly than traditional frequency- or time-domain representations. The new algorithms compute these functions efficiently by solving a partial differential equation (PDE) in the time or mixed frequency-time domains. Frequency-domain spectra or time-domain waveforms are generated from the multivariate functions as cheap post-processing steps. Two methods, multivariate FDTD and hierarchical shooting, are purely time-domain techniques suitable for the general strongly nonlinear circuit problem. They differ in their memory and computation needs. A new mixed frequency-time method is more efficient for circuits that are moderately (but not strongly) nonlinear in one or more tones (e.g, switching mixers, switched-capacitor filters). The linear systems at the core of all three techniques are especially well suited for iterative solution. This is exploited in the methods to achieve linear growth of computation and memory with respect to circuit size. The mixed frequency-time method of this work avoids the ill-conditioning problem by computing with the slow harmonic components directly, rather than with time-domain samples at sets of points close to each other.