Research On Aging Sensing Of Integrated Circuits Based On NBTI Effect

With the development of semiconductor technology,the aging of CMOS integrated circuits has become a problem of increasing concern.The performance of the circuit will slowly degrade due to aging and eventually fail.Among all the aging mechanisms,Negative Bias Temperature Instability(NBTI)is the main factor affecting the circuit.Therefore,it is important to assess circuit aging due to NBTI,in order to develop aging mitigation measures at the circuit design stage.Existing research on the evaluation of aging mainly focuses on three directions:transistor-level model,which has high accuracy but low speed;The Look-up Table(LUT)model has high speed but low precision;The machine learning model captures the potential relationship between variables through the machine learning algorithm,with high accuracy and fast speed.In this paper,two aging perception models based on machine learning are designed for circuit aging caused by NBTI:(1)linear regression aging perception model based on the random three-layer sub-circuit;(2)Linear regression aging perception model based on the novel critical gate.The linear regression sensing model based on the random three-layer sub-circuit simulates the aging delay trend of the random three-layer sub-circuit,through the linear regression algorithm.Firstly,HSPICE software and MOSRA model were used to obtain the increment of cell gate aging delay and establish the gate aging model.Then,static time sequence analysis was used to obtain the random three-layer sub-circuit aging delay,and the linear regression algorithm was used to simulate the aging delay trend.The accuracy of the model is verified on the reference circuits of ISCAS’85 and ISCAS’89 of the 32 nm transistor process.The worst RMSE,MAE and MAPE of the model are8.4231 ps,23.9850 ps and 3.4440% respectively.It is significantly improved compared with other aging perception models based on machine learning.Considering the randomness of the random three-layer sub-circuit,the quality of the training set of the linear regression perception model based on the random three-layer sub-circuit is not high,this paper also designs a linear regression aging perception model based on the novel critical gate.Firstly,the number of gates,contained in the three-layer sub-circuit area of each gate in the circuit,are calculated,and it is taken as the weight of the gates.The gate with the largest weight value is the novel critical gate.Then,on the basis of the linear regression aging perception model based on the random three-layer sub-circuit,the linear regression algorithm was used to simulate the aging delay trend of the three-layer sub-circuit formed by the novel critical gate,and the model was established.Taking the ISCAS’85 and ISCAS’89 reference circuits of 32 nm transistor technology as experimental samples,the results show that the worst RMSE,MAE and MAPE of this model are 2.1922 ps,6.2415 ps and 0.8963% under different reference circuits,which is a reliable solution.Figure [29] Table [15] Reference [70]...