Research On The Aging Fault Diagnosis Of SRAM Memory Circuit Based On Neural Network

Electronic systems used in the aircraft becomes more integrated and intelligent than ever before due to the rapid development of aerospace industry during the last few years.Static random-access memory(SRAM)and other storage circuits in the avionics system have been exposed to harsh environments such as high temperature and impact of cosmic high-energy particles for a long time,and are susceptible to hardware failures such as aging and radiation effects.Sex is difficult to guarantee.The traditional experience-based fault diagnosis model has been difficult to meet the reliability requirements of modern complex and huge storage circuits.With its excellent function fitting and pattern recognition capabilities,neural networks have gradually become a new technology for contemporary integrated circuit fault diagnosis.means.The thesis adopts the methods of theoretical research,model analysis,data simulation and software verification to design the fault data simulation platform and fault diagnosis model based on RBF neural network in the context of aging and radiation faults.First,the degradation mechanism of common aging radiation faults is analyzed and the degradation model of the faults to be tested is established.Based on the physical model,the degradation mechanism of several common faults is analyzed,and the negative temperature bias instability(NBTI)and single event effect(SEE)effects,which have serious effects on the storage circuit,are determined as The fault to be tested.According to the reaction diffusion model,the functional relationship between the threshold voltage drift of the transistor affected by the NBTI effect and time,stress,gate voltage and other factors is fitted.The PMOS under the degradation of the NBTI effect is established by the adjustable voltage source and the gate in series Transistor model;according to the transient injection pulse model,the relationship between the transient current under the single-particle effect and the injected particles and the device process is obtained.The NMOS tube model under the degradation of the single-particle effect is established using verilog A language.The established model can be directly embedded into the storage And simulate the data in the circuit.Secondly,the fault simulation circuit was designed and constructed and analyzed.In the Cadence environment,a 128x64-bit SRAM circuit was built based on the TSMC40 nanotechnology library as a system platform for integrated circuit aging and radiation fault injection and simulation,including basic 6-tube memory cells,logic control circuits,address input circuits,and row and column decoders And sensitive amplifiers.The accuracy of the memory circuit function is verified by continuously completing the write0,write 1 and read 1 operations within 100 ns.By embedding and simulating the fault model,the relationship between circuit delay,voltage,power supply current and other information and the degree of degradation is obtained.Finally,I built a Radial Basis Function(RBF)neural network diagnostic platform and performed diagnostic verification on the circuit under test.A radial basis neural network with 20 hidden layer nodes is designed as a diagnostic tool,Softmax function is selected as the output classification function,and the software implementation of the RBF network is completed using the Tensor Flow neural network toolkit in the Py Charm environment.The neural network fault analysis model and the SRAM simulation circuit are collaboratively verified.The bit line voltage,storage voltage and power supply current under different fault injection conditions are collected as characteristic parameters.The training set is used to adjust the network parameters and the test set is used to test the training.Neural network accuracy and diagnostic accuracy.The experimental results show that when the data training set is 7000 and the number of network training iterations reaches 1000,the diagnostic accuracy of circuit faults can reach 97%.