Optimizing Tests for Multiple Fault Models

We present a method for deriving minimal tests to cover multiple fault models. Integer linear programming (ILP) is used to select a minimum set of vectors without reducing the original coverage. Tests can be initially generated separately for different fault models. All tests are then simulated for each fault model of interest without fault dropping. The fault simulation data are converted into ILP constraints, whose solution guarantees the minimality of test length. As an illustration we have used stuck-at, transition, and IDDQ (pseudo stuck-at) faults. Tests are generated and simulated using Mentor Graphics FastScan and then minimized by the AMPL-CPLEX ILP software. The illustration includes combinational, scan and non-scan sequential circuits. For scan circuits, either type, i.e., launch on shift (LOS) or launch on capture (LOC) can be used. In each case, a minimal set of vectors is obtained in which a small subset of vectors is identified for IDDQ measurement. An interesting aspect of the ILP methods is the trade off they allow between the absolute minimality of test length and the number of IDDQ measurements.