Temperature analysis of electronic devices for reliability design based on Process Neural Network

Aiming at reliability design for electronic devices, process neural networks are proposed to predict the temperature of electronic devices. To avoid errors caused by discrete input data fitting or difference, only discrete data is used when solving orthogonal transformation coefficients. To accelerate the learning speed of the gradient descent algorithm, a parameter- independent adaptive learning algorithm is developed. The results show that this model has better accuracy and generalization ability compared with artificial neural networks and linear regression method, and the parameter-independent adaptive learning algorithm has quicker convergence rate compared with parameter-fixed algorithm and adaptive learning algorithm.