Design and Implementation of Fault-Tolerant Convolutional Neural Network for Safety-Critical Embedded Systems

Convolutional neural networks (CNNs) have gained significant attention in safety-critical embedded systems, such as those deployed in autonomous vehicles, medical devices, and industrial control systems, for their remarkable capabilities in image and pattern recognition tasks. In addition, the fault-tolerance capability is also appreciable. This chapter proposes a fault-tolerant approach for implementing CNNs on field-programmable gate arrays (FPGAs) to enhance the reliability and robustness of the embedded system. The fault injection mechanism is also applied to create a faulty environment, yielding reduced effort for testing the system’s fault tolerance mechanism. A comprehensive analysis of memory-fault sources and their impact on CNN-based systems has been discussed. Experimental results demonstrate that the fault-tolerant 24 approach effectively corrects faulty output, enhancing the reliability and robustness of CNN-based embedded systems. Additionally, this chapter investigates the trade-offs between fault tolerance and resource utilization in the FPGA implementation.