Application Of Improved Generalized Regression Neural Network In Power And Signal Integrity Of High-speed Circuit

The development of circuit systems is faced with the high requirements of increasing operating speed,increasing circuit integration and increasing data throughput.Following this,the size of the circuit system is getting smaller and smaller,the number of integrated circuits is also increasing,and the components and traces on the circuit board are more densely.The unreasonable layout of interconnects in high-speed systems has brought serious Signal Integrity(SI)problems.The dense components amplify the power supply noise and lead to Power Integrity(PI)problems.How to effectively suppress the noise propagation,while ensuring SI and PI performance of the system does not deteriorate is a problem needs to be considered in the design of the circuit system.In addition,the efficiency of the existing circuit system design method is very slow because of the discontinuity and non-differentiability of the circuit system.How to apply intelligent algorithms to the circuit design process and improve the efficiency of circuit system design is also a significant research direction.This article focuses on how to combine the improved Generalized Regression Neural Network(GRNN)and Genetic Algorithm(GA)to automatically design the Electromagnet Band Gap(EBG)structure and the compensation inductance structure for different target performance,so as to solve the SI and PI problems in high-speed circuit systems.The main innovations of this article are as follows:(1)K-means clustering is added to the traditional GRNN and the training samples are classified first.This way improves the problem of insufficient accuracy of local nodes caused by the global prediction of the traditional GRNN.After comparison and analysis,the improved GRNN has higher simulation accuracy of the structure and the advantages of high efficiency and high generalization ability.On this basis,the improved GRNN and GA are combined,and an automatic circuit structure design scheme based on the combined algorithm is proposed.(2)Two Koch fractal L-bridge Electromagnet Band Gap(EBG)structures are designed using the improved GRNN and GA combined algorithm for two design requirements,and the combined algorithm is used to optimize the design of the number of fractal iterations and four structural parameters.The first automatically designed Power Distribution Network(PDN)composed of an EBG structure can suppress Simultaneous Switching Noise(SSN)in the frequency range of 360 MHz to 20 GHz with a suppression depth of-50 d B.The automatically designed PDN composed of the second EBG structure maintains the same suppression capability,while reducing the area of the etched structure inside the unit and improving the SI problem caused by the destruction of the PDN power layer.(3)The combined algorithm of improved GRNN and GA is used to design the compensation inductance structure for the curved differential transmission line with two angle corners.The combined algorithm is used to optimize the design of the compensation structure's coil turns and four structural parameters.The two compensation inductance structures obtained are embedded into the 90° and 135° corner bend differential transmission lines respectively,which made the electrical length difference between the inner and outer wires of the two is significantly reduced.The two structures both can suppress the differential common-mode conversion noise in the frequency range of DC to 15 GHz with a suppression level of-25 d B.The compensation structure also reduces the differential mode return loss and insertion loss of the bent differential transmission line without needed additional wiring space.Time-domain analysis and eye diagram simulation were performed on the two curved differential transmission lines.The comparison results show that the common mode noise and time delay of the differential line are significantly reduced and the eye diagram performance is better by adding the compensation inductance structure.The two EBG structures and the two curved differential transmission lines with compensation structures obtained in the research are processed by single-layer printed circuit technology,and the physical properties are tested with a vector network analyzer.The comparison found that the simulation and test results are basically the same,and this performance verifies the proposed technology.