Method and apparatus for designing and manufacturing electronic circuits subject to process variations

Abstract

Methods and apparatus are described in which, at design-time a thorough analysis and exploration is performed to represent a multi-objective 'optimal' trade-off point or points, e.g. on Pareto curves, for the relevant cost (C) and constraint criteria. More formally, the trade-off points may e.g. be positions on a hyper-surface in an N-dimensional Pareto search space. The axes represent the relevant cost (C), quality cost (Q) and restriction (R) criteria. Each of these working points is determined by positions for the system operation (determined during the design-time mapping) for a selected set of decision knobs (e.g. the way data are organized in a memory hierarchy). The C-Q-R values are determined based on design-time models that then have to be 'average-case' values in order to avoid a too worst-case characterization. At processing time, first a run-time BIST manager performs a functional correctness test, i.e. checks all the modules based on stored self-test sequences and “equivalence checker” hardware. All units that fail are deactivated (so that they cannot consume any power any more) and with a flag the run-time trade-off controllers, e.g. Pareto controllers, are informed that these units are not available any more for the calibration or the mapping. At processing time, also a set of representative working points are “triggered” by an on-chip trade-off calibration manager, e.g. a Pareto calibration manager, that controls a set of monitors which measure the actual C-Q-R values and that calibrates the working points to their actual values. Especially timing monitors require a careful design because correctly calibrated absolute time scales have to be monitored.