Aiming at the problems that the fault definition standard of analog circuit board chip is not clear and it is difficult to realize fast and accurate classification, this paper proposes a fault diagnosis model based on binary convolution logistic atom search algorithm to optimize BP neural network. Firstly, the temperature of circuit board chip in different states is collected and feature extracted, and the features are fused by Euclidean distance to establish a fault feature model containing chip fault definition criteria. Then, the binary convolution logistic map is used to initialize the population size and location of the atomic search algorithm to improve the convergence speed and accuracy. Then, the optimization process of BP neural network is optimized by BCL-ASA to obtain the optimal weight and threshold. Finally, input the chip fault characteristic model into the BCL-ASA-BP neural network for training and testing to complete the circuit board chip fault diagnosis. The experiment uses the power supply circuit board for reliability analysis, and the results show that the accuracy of BCL-ASA-BP's comprehensive diagnosis of chip faults reaches 98.35%, which is 13.9% higher than the traditional BP algorithm.