NBTI Caused Aging Prediction Technology Based On RISC-V Processor

As the semiconductor manufacturing process enters the nano-age,the impact of the failure mechanisms such as Negative Bias Temperature Instability(NBTI)on the circuit can not be ignored.Under the long-time influence of these mechanisms,the aging in the circuit will be intensified.In order to detect circuit aging,some researchers have made full use of the advantages that machine learning algorithms can learn from existing data to achieve high accuracy in the determination o f tested data,and the fault analysis has got ideal results.Based on the circuit aging caused by NBTI effect,an open-source RISC-V processor soft core is used as a sample circuit,and an integrated circuit aging detection platform based on the Prime Time power simulation and machine learning algorithm is successfully designed to complete the prediction of aging in RISC-V processor.The basic principle of the aging of the circuit caused by the NBTI effect is studied in this paper.Then,the influence of the NBTI effect on the circuit delay is analyzed,and the research scheme based on the power consumption information analysis is determined to analyze circuit aging.With the help of Spectre,the delay in 15 years of the standard unit used in the design is obtained.In order to obtain the power consumption data in the aging circuit,an integrated circuit aging detection platform based on the Prime Time power simulation and machine learning algorithm is designed.And the PCA algorithm is used to process data.Then the analysis and training of power consumption data can be completed on the data analysis platform composed of neural network algorithms.By comparing and analyzing the best machine learning models obtained from the training,the BP neural network model is chosen to be implemented on FPGA.In this paper,RISC-V is selected as the sample circuit for aging test.And then the digital circuit design process based on RISC-V processor soft core is completed and the processor layout is obtained.Timing checking is finished with Prime Time and the verification of the function and timing can be done through post-simulation.Then the aging prediction of the circuit can be completed by using the built prediction platform.Experimental results show that when the tolerance limit of the output is set as 0.5,the classification accuracy of BP-ANN algorithm and the PSO-ANN algorithm are both ideal.While the running time of PSO-ANN algorithm is about 3 times of the BP-ANN algorithm,and the activation function used is more complex,so the BP-ANN algorithm is chosen to complete the model building on FPGA.When different data are tested,the model can get the same output as the simulator and successfully predict the circuit aging.