Improving indirect test efficiency using multi-directional tessellations in the measure space

Abstract

Indirect test strategies have risen as a promising solution to overcome the challenges encountered in analog and mixed-signal circuit testing and the ever increasing device verification costs. This work explores the possibilities of using multidirectional tessellations in the indirect measure space aiming to reduce false positive test outcomes. The key idea of the proposal is to use several rotated 2n-ary trees in order to tessellate and encode the indirect measure space along multiple directions. Such tessellations create a refinement in the highly non linear test decision boundary without the need of including more circuit samples within the training set. The trained trees, together with a strict test decision criterion, serve as a classifier during the production testing phase. The proposed multi-directional tessellation methodology has been applied to test a Biquad filter under the presence of parametric variations. Promising simulation results report considerable improvements in lowering test escape metrics by an average factor of 50 as compared to a single octree tessellation as well as low computational overhead.