The rapid evolution of embedded and automotive systems has introduced unprecedented complexity, driven by the integration of multi-core processors, real-time operating systems, and software-defined functionalities. This complexity has significantly increased the vulnerability of such systems to transient and permanent faults, particularly radiation-induced soft errors and memory safety violations. This research develops a comprehensive, theoretically grounded framework for fault tolerance that integrates hardware-based lockstep architectures, software-level fault detection and recovery mechanisms, and cyber-physical safety models. Drawing on foundational and contemporary literature, the study critically examines the limitations of software-only approaches in error detection coverage, the effectiveness of dual-core lockstep systems in mitigating soft errors, and the role of architectural diversity and safety frameworks such as the Simplex architecture and time-triggered systems. The methodology employs a conceptual modeling approach to analyze fault propagation, detection latency, and system recovery across heterogeneous computing environments, including automotive zonal controllers and high-performance embedded platforms. The findings demonstrate that hybrid architectures combining hardware redundancy with selective software-based mechanisms significantly enhance fault coverage and system resilience while maintaining manageable performance overhead. Furthermore, the incorporation of safety-oriented architectural paradigms effectively limits fault propagation and ensures predictable system behavior. The study highlights the importance of integrating memory safety mechanisms and control flow integrity techniques to address emerging software vulnerabilities. The discussion explores the implications of these findings for next-generation automotive and cyber-physical systems, emphasizing scalability, energy efficiency, and real-time constraints. Future research directions include adaptive fault-tolerance strategies and the integration of intelligent monitoring systems. This work contributes a unified perspective on resilient system design, bridging the gap between hardware reliability, software correctness, and system-level safety.