Abstract-Measuring the steady state leakage current (IDDQ) is very successful in detecting faults not discovered by standard fault models. But vector dependencies of IDDQ decrease the resolution.We propose deterministic ATPG algorithms to create test vectors within predefined leakage ranges. Even when random pattern generation does not find test vectors, the proposed algorithms identify vectors within the desired range. Experimental results confirm that leakage constraints are effectively handled during test pattern generation without decreasing fault coverage.Keywords-Algorithms, deterministic ATPG, IDDQ
I. INTRODUCTIONThe steady state leakage current (IDDQ) is a good indicator to decide whether a circuit contains failures introduced during production. Even faults that remain undiscovered using functional testing based on fault models are detected by IDDQ measurements [1].With continuously shrinking feature sizes the IDDQ current of devices increases. At the same time the IDDQ current of good devices changes due to process variations and test vector dependencies. Consequently, differentiating good and bad devices by using a simple threshold value for the IDDQ current becomes infeasible.Instead, post-processing techniques are typically applied to handle IDDQ variations. Current signatures [2] are a sorted plot of measured IDDQ values. Discontinuities in this curve typically indicate a fault. Delta-IDDQ [3] is an improvement that compares the differences between measurements and yields more accurate information. These techniques and similar approaches [4] help to remove certain effects coming from process variations and from test vector dependencies.In contrast to these approaches the technique of [5] is applied before the measurement during Automatic Test Pattern Generation (ATPG). By this, leakage variations coming from test vector dependencies are drastically reduced. An IDDQ model predicts the expected leakage current for a given test vector. Then, a small range for IDDQ is defined. Only test vectors within this range are created by the ATPG tool. Figure 1 shows the resulting leakage signatures. No restrictions (α = ∞) yield test vectors across a wide range of leakage values. Tight restrictions (α = 0.5) yield an almost linear curve with a small slope while keeping high fault coverage. Consequently, good and bad devices can be differentiated more easily by IDDQ testing. But no complete ATPG algorithm was given, instead a simple heuristic was applied to generate test vectors. That approach cannot decide whether no test vector within the defined range exists.Algorithms for input vector control [6], [7], [8] search for the input assignment causing the lowest quiescent current for a circuit. Thus, a single optimization problem is solved.