"AI-Driven Cross-Domain Debugging in Embedded Systems: A Systematic Mapping Study and Hybrid Synthesis"

"The verification of heterogeneous embedded systemsis hindered by a critical abstraction gap between high-levelsoftware intent and low-level hardware execution. We presenta Systematic Mapping Study (SMS) and hybrid synthesis of 53primary studies (2020\u20132025) focusing on AI-driven cross-domaindebugging. Adhering to PRISMA 2020 guidelines, we identified871 records, screened 210 full-text papers, and included 53studies after quality assessment. We establish a hybrid synthesisframework stratifying studies into high-confidence (Class A:Quantitative + Sample Size), narrative (Class B: Quantitativeonly), and thematic (Class C: Qualitative\/Hardware-focus) tiersusing a custom Engineering Risk of Bias tool. Quantitativemeta-analysis of high-confidence (Class A) studies (n=4) yields apreliminary weighted pooled accuracy of 98.51% [95% CI: 97.2\u201399.4] for AI-driven fault localization. However, the small samplesize necessitates caution in interpretation. Meta-regression indicatesthat platform heterogeneity explains only 4.79% of thevariance in debugging performance (R\u00b2=0.0479), with no significantpenalty for downscaling to resource-constrained platforms.The systematic map identifies a dominant cluster of GraphNeural Networks (GNNs) in fault diagnosis, yet highlights acritical research gap in RISC-V platform coverage (only 5.7%of studies, n=3). Furthermore, resource analysis indicates thatonly 13.2% of TinyML solutions (7\/53) successfully meet hardreal-time deadlines (< 10 ms). While AI enables causal reasoningacross system layers to reduce Mean-Time-To-Repair (MTTR),the lack of ISO 26262-compliant explainability remains a barrier.We propose a Neurosymbolic Safety Framework to bridge thisgap by combining deep learning with formal runtime monitors."