A Learning-Based Cell-Aware Diagnosis Flow for Industrial Customer Returns

Abstract

Diagnosis is crucial in order to establish the root cause of observed failures in Systems-on-Chip (SoC). In this paper, we present a new framework based on supervised learning for cellaware defect diagnosis of customer returns. By using a Naive Bayes classifier to accurately identify defect candidates, the proposed flow indistinctly deals with static and dynamic defects that may occur in actual circuits. Results achieved on benchmark circuits, as well as comparison with a commercial cell-aware diagnosis tool, show the effectiveness of the proposed framework in terms of accuracy and resolution. Moreover, the proposed flow has been experimented and validated on industrial circuits (two test chips and one customer return from STMicroelectronics), thus corroborating the results achieved on benchmark circuits.

Cell-Aware Diagnosis of Customer Returns Using Bayesian Inference

“…The most common is when the test sequences used during manufacturing test are available and can be used again for diagnosis purpose so as to mimic the process used initially during manufacturing test. This is the scenario assumed in our previous work [17], in which we considered two successive test sequences used to performed customer return diagnosis. First, a static CA test sequence generated by a commercial cell-aware ATPG tool is applied to the Circuit Under Diagnosis (CUD).…”

“…These results are provided in the form of a fault dictionary containing, for each defect within a cell, the cell input patterns detecting (or not) this defect. An example of training data as used in [16][17] and containing four instances for an arbitrary two-input cell is shown in Fig. 4.…”

“…For this reason, the value '0.5' is assigned to each dynamic defect (D 11 , D 12 ) for all related static cellpatterns, meaning that such a defect is detectable or not depending on whether or not the above conditions are satisfied. As only dynamic instance tables are manipulated in this work, the representation of training data as used in [16][17] can be simplified without loosing information and decreasing the quality of the training phase. This comes from the observation that a static defect is a particular case of dynamic defect (e.g., a full open is a resistive open with an infinite value of the resistance), and that all static cell-patterns for a given defect are embedded in its whole set of dynamic cell-patterns.…”

“…In an attempt to deal concurrently with all types of defects that may occur in customer returns, we proposed a new CA diagnosis flow in [17]. We assumed a test protocol in which two test sequences (static and dynamic) are used successively, each one assuming a dedicated testing scheme, i.e., basic scan and fast sequential.…”

“…In this paper, we propose a new version of the CA diagnosis flow in which both static and dynamic defects can be diagnosed owing to a single dynamic test sequence applied atspeed. Such a flow can be used to address a missing scenario in our former proposals [16][17]. As only dynamic instance tables are manipulated in this work, the representation of training and new data is simplified (a single type of feature vector is used) without loosing information and decreasing the quality of the training and inference phases.…”