INTRODUCTION To achieve the highest product quality, Cell-Aware (CA) test has become mandatory for semiconductor industry. In this methodology, a cell-internal-fault dictionary or CA model, describing the detection conditions of each potential defect affecting a cell, is used [1-2]. However, the generation of CA models for all standard cells is a time- and resource-consuming task that limits the deployment of CA test. Typical CA model generation flow starts with a SPICE netlist representation of a standard cell. This representation is used by an electrical simulator to simulate each potential defect against an exhaustive set of stimuli. The stimuli detecting defects are synthetized into a CA model. As thousands of standard cells, with various complexities, are used for a given technology, the generation time of CA models for complete standard cell libraries may reach up to several months, thus drastically increasing the library characterization process cost. To improve the generation run time of CA models and ease the characterization, this work proposes a methodology to predict the behavior of cell-internal defects using Machine Learning (ML) [3]. More widely, the goal is to use existing CA models from various standard cell libraries developed using different technologies to predict CA models for new standard cells independently of the technology.