Assessing the Effectiveness of Software-Based Self-Test Programs for Cell-Aware Test

Software-Based Self-Test (SBST) is a widely adopted technique for ensuring the in-field reliability of safety-critical systems, particularly in the form of Software Test Libraries (STLs). While the effectiveness of SBST is traditionally assessed using classical fault models such as stuck-at and transition delay faults, these models may not fully capture the behavior of modern nanometric technologies. In contrast, Cell-Aware Testing (CAT) has emerged as a defect-oriented approach that more accurately reflects physical defects at the cell level, and is gaining traction in manufacturing test flows. This paper presents a structured framework for evaluating SBST programs under defect-oriented Cell-Aware Test (CAT) models and interpreting the results at the defect level. The proposed approach combines defect-to-conditional fault mapping, a simplified observability-based taxonomy, and a post-processing flow to derive accurate defect-level coverage. In addition, a table test percentage (TT%) metric is introduced and exploited to quantify intrinsic defect detectability and guide test improvement. Together, these techniques enable a consistent defect-level interpretation of fault simulation results, reduce the overestimation inherent in raw conditional-fault coverage, and provide actionable insight for identifying and prioritizing hard-to-detect defects. The proposed framework is validated on a RISC-V System-on-Chip using multiple Software Test Libraries and two technology libraries. Experimental results show that, while existing STLs achieve substantial coverage on CAT defects, a subset of defects remains systematically hard to detect due to intrinsic observability constraints. The proposed framework identifies critical hard-to-detect defects, enables targeted STL refinement, and remains compatible with industrial defect-oriented validation flows.