CN113312868A - Miniaturized partially-fractal electromagnetic band gap structure of high-speed power distribution network and self-adaptive design method thereof - Google Patents
Miniaturized partially-fractal electromagnetic band gap structure of high-speed power distribution network and self-adaptive design method thereof Download PDFInfo
- Publication number
- CN113312868A CN113312868A CN202110584668.5A CN202110584668A CN113312868A CN 113312868 A CN113312868 A CN 113312868A CN 202110584668 A CN202110584668 A CN 202110584668A CN 113312868 A CN113312868 A CN 113312868A
- Authority
- CN
- China
- Prior art keywords
- bridge
- band gap
- fractal
- electromagnetic band
- distribution network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013461 design Methods 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 title claims abstract description 32
- 230000005540 biological transmission Effects 0.000 claims abstract description 9
- 239000002184 metal Substances 0.000 claims description 51
- 238000013528 artificial neural network Methods 0.000 claims description 34
- 238000012549 training Methods 0.000 claims description 34
- 230000002068 genetic effect Effects 0.000 claims description 15
- 238000005457 optimization Methods 0.000 claims description 15
- 239000000758 substrate Substances 0.000 claims description 13
- 239000003990 capacitor Substances 0.000 claims description 11
- 210000000349 chromosome Anatomy 0.000 claims description 9
- 238000004088 simulation Methods 0.000 claims description 9
- 230000000306 recurrent effect Effects 0.000 claims description 8
- 230000006978 adaptation Effects 0.000 claims description 7
- 238000003064 k means clustering Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 230000006399 behavior Effects 0.000 claims description 3
- 230000008878 coupling Effects 0.000 claims description 3
- 238000010168 coupling process Methods 0.000 claims description 3
- 238000005859 coupling reaction Methods 0.000 claims description 3
- 238000009792 diffusion process Methods 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 108090000623 proteins and genes Proteins 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims 1
- 230000001629 suppression Effects 0.000 abstract description 13
- 230000001360 synchronised effect Effects 0.000 abstract description 12
- 230000000694 effects Effects 0.000 abstract description 7
- 238000004806 packaging method and process Methods 0.000 abstract description 3
- 230000007547 defect Effects 0.000 abstract description 2
- 210000004027 cell Anatomy 0.000 description 3
- 230000005764 inhibitory process Effects 0.000 description 3
- 230000000737 periodic effect Effects 0.000 description 3
- 238000012938 design process Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005530 etching Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000003071 parasitic effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/39—Circuit design at the physical level
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Hardware Design (AREA)
- Genetics & Genomics (AREA)
- Physiology (AREA)
- Geometry (AREA)
- Shielding Devices Or Components To Electric Or Magnetic Fields (AREA)
Abstract
The invention relates to a miniaturized partially-fractal electromagnetic band gap structure of a high-speed power supply distribution network and a self-adaptive design method thereof, and compared with the prior art, the miniaturized partially-fractal electromagnetic band gap structure overcomes the defects that the unit size is overlarge, a globally-distributed resonator influences the signal integrity and the EBG structure is difficult to intelligently design in order to meet the performance of a broadband high suppression degree of a traditional EBG structure unit. A square hollow structure is etched in the center of an L-bridge electromagnetic band gap structure unit positioned at the position of a noise source and a noise sensitive circuit on a power distribution network, and a small Koch fractal L-bridge electromagnetic band gap structure is embedded in the square hollow structure. The invention adopts a Koch fractal method to deform the L-bridge EBG embedded unit, improves the effects of wide noise band and high suppression degree of the synchronous switch, reduces the area of the embedded EBG structural unit, further reduces the damage to a power supply layer, and can obviously reduce the influence on the signal integrity of adjacent transmission lines in three-dimensional system-level packaging.
Description
Technical Field
The invention relates to the technical field of high-speed power supply distribution networks, in particular to a miniaturized partially-fractal electromagnetic band gap structure of a high-speed power supply distribution network and a self-adaptive design method thereof.
Background
The development of circuit systems is faced with the high requirements of continuously increasing the operation speed, continuously increasing the circuit integration level and continuously increasing the data throughput. With the size of the circuitry becoming smaller and smaller, the number of integrated circuits increasing, and the components and traces on the circuit board becoming denser. The dense distribution of components amplifies Power supply noise resulting in Power Integrity (PI) problems. How to effectively suppress noise propagation in a high-speed circuit system on the premise of ensuring that the PI performance of the system is not deteriorated is an important consideration in circuit system design. Furthermore, existing circuitry designs are very slow because of discontinuities and irreparability of the circuitry. How to apply the intelligent algorithm to the circuit design process and improve the design efficiency of a circuit system is also a significant research direction.
When a large number of active devices are Switching simultaneously, Synchronous Switching Noise (SSN), also known as delta-I Noise or ground/power supply bounce Noise (G/PBN), is typically generated simultaneously. Transient switching currents flowing along the parasitic inductances of the Power/ground lines will induce fluctuations or disturbances in the Power Distribution Network (PDN).
In recent years, experts have found that the EBG structure or the application thereof to the PDN can bring high impedance and further attenuate SSN propagation, and the coplanar EBG structure has come into play. Yang Fei-Ran et al, 1999, designed a UC-EBG structure with periodic resonant cells etched only in the power plane of the PCB. Coplanar EBG structures are obtained by etching a one-dimensional periodic pattern of circular, sinusoidal, triangular, square or fractal structures on the power plane, where fractal geometry is often used to design compact electromagnetic structures due to its own space filling and self-similarity, such as Koch fractal which increases the length of the structure in a limited space. Generally, at least 4 or 5 periodic EBG cells are required to provide good bandgap characteristics, and thus a large physical space is required to integrate EBGs into one system. The planar EBG structure only needs to be etched in a power supply layer, the processing mode is simple and easy to operate in practical application, and meanwhile, the strong synchronous switch noise suppression capability can be obtained.
Disclosure of Invention
The invention aims to solve the defects that the traditional EBG structure unit in the prior art has the performance of meeting the high broadband suppression degree, the unit size is overlarge, the overall distribution resonator influences the signal integrity and the EBG structure is difficult to intelligently design, and provides a miniaturized partially fractal electromagnetic band gap structure of a high-speed power supply distribution network and a self-adaptive design method thereof to solve the problems.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a miniaturized local fractal electromagnetic band gap structure of a high-speed power distribution network comprises a dielectric substrate, wherein a metal grounding plate is printed on the back surface of the dielectric substrate, and a power distribution network is printed on the front surface of the dielectric substrate; the power distribution network is 9L-bridge electromagnetic band gap structural units which are placed on a dielectric substrate in a 3X 3 mode, feed through holes are formed in the centers of any two of the 9L-bridge electromagnetic band gap structural units, the two feed through holes are connected with a metal grounding plate and a metal patch, each L-bridge electromagnetic band gap structural unit comprises a metal patch, four L-shaped bridges are connected to the periphery of each metal patch, and adjacent metal patches distributed in the positive direction of an X axis and a Y axis are connected through the L-shaped bridges;
the power distribution network is distributed by using a local topological network, a square fractal hollow structure is etched only at the positions of a noise source and a noise sensitive circuit, and a miniaturized and deformed Koch fractal L-bridge electromagnetic band gap structure is embedded in the square fractal hollow structure;
the small Koch fractal L bridge electromagnetic band gap structure comprises a polygonal metal patch, wherein an L-shaped Koch fractal bridge extends from each edge of the polygonal metal patch, the long edge of the L-shaped Koch fractal bridge is parallel to the edge of the polygonal metal patch extending out of the L-shaped Koch fractal bridge, and the adjacent L-shaped Koch fractal bridges are arranged at an angle of 90 degrees.
Four sides of the small Koch fractal L bridge electromagnetic band gap structure are inwards sunken towards the center of the small Koch fractal L bridge electromagnetic band gap structure to form a triangular shape, wherein each side of the polygonal metal patch is embedded towards the center of the small Koch fractal L bridge electromagnetic band gap structure to form a triangular shape, and the long side of the small Koch fractal L bridge is embedded towards the center of the polygonal metal patch to form a triangular shape.
Four sides of the small Koch fractal L bridge electromagnetic band gap structure are inwards sunken towards the center of the small Koch fractal L bridge electromagnetic band gap structure to form a pentagram shape, wherein each side of the polygonal metal patch is embedded towards the center of the small Koch fractal L bridge and forms a pentagram shape, and long sides of the small Koch fractal L bridge are embedded towards the center of the polygonal metal patch and form a pentagram shape.
Four sides of the small Koch fractal L bridge electromagnetic band gap structure are positioned on two sides of a pentagram shape and are all sunken in to form a triangular shape, wherein each side of the polygonal metal patch is positioned on two sides of the pentagram shape and is sunken in to form the triangular shape, and a long side of the L-shaped Koch fractal bridge is positioned on two sides of the pentagram shape and is sunken in to form the triangular shape.
The lumped parameter of the 9L-bridge electromagnetic band gap structural units comprises a parallel plate capacitor Cp,lParallel plate inductor Lp,lBridge microstrip capacitor CbrLocal inductance L of bridgebrAnd the gap coupling capacitance C between adjacent metal sheetsg(ii) a The lumped parameter of the small Koch fractal L-bridge electromagnetic band gap structure containing element is provided with a polygonal metal patch plate capacitor Cp,mlPolygonal patch board capacitor Lp,mlKoch fractal L bridge local inductance Lb,ml。
The small Koch fractal L-bridge electromagnetic band gap structures are arranged at the 1-1 position and the 3-3 position of the power distribution network.
A miniaturized partial fractal electromagnetic band gap structure of a high-speed power supply distribution network and a self-adaptive design method thereof are designed based on an improved generalized recurrent neural network and a genetic algorithm, the genetic algorithm is utilized to simulate the selection and genetic process in the nature and continuously search the optimal solution in a region to be optimized, and the improved generalized recurrent neural network is used to replace simulation software to model the designed initial electromagnetic band gap structure; after obtaining the simulation database, the genetic algorithm is used for optimization to obtain the optimal structural parameters to meet the design requirements, which includes the following steps:
81) obtaining sample data: selecting an initially optimized circuit structure and selecting subsequently optimized size parameters according to a circuit design basic principle, and determining a numerical range of the parameters to be optimized;
82) training the K mean generalized regression neural network: generating a sample in a parameter value range to be optimized, and training a K mean generalized regression neural network;
83) randomly generating 20 individuals according to a set coding rule within a parameter value range to be optimized to construct a first generation population, wherein the coding rule is as follows:
the length k of the chromosome can be calculated using the following formula:
2k-1＜(U2-U1)×10n≤2k-1
wherein [ U ] is1，U2]Is the value range of the structural parameter, n represents the decimal place number;
the decimal value is then converted to a binary chromosome by the following equation:
84) then, decoding chromosomes in the population into decimal values, importing the decimal values into the trained improved generalized regression neural network, and outputting corresponding S parameters of individuals;
84) writing a fitness function according to design requirements, importing an S parameter of an individual into a fitness function equation, and calculating a fitness value corresponding to the individual;
85) in order to evaluate individual performance, the adaptation degree of individuals in the population to the environment is calculated, namely an adaptation function value, the adaptation degree value of the individuals is calculated through a set target function, after sequencing and comparison, the excellent degree of the individuals is determined, and then whether the individuals are used as a parent gene to be inherited to a next generation population is selected; the fitness function is as follows:
86) evaluating the individual performance according to the fitness value, and determining the optimal structural parameters when the fitness value reaches the highest value and meets the expected design performance; otherwise, generating a population by selecting a cross and variation method; and after the new generation population is generated, returning to the step 84) until an optimal solution meeting the design target is found.
The training of the improved K-means generalized regression neural network comprises the following steps:
91) obtaining sample data: according to the circuit principle analysis, the internal length d of the polygonal patch board and the length L of the L-bridge short edge embedded in the polygonal patch board are selected2Distance L from the starting point of the long side of the L-bridge to the edge of the polygonal patch3Width w of L bridgel2Koch fractal iteration number i, operating frequency fcS of electromagnetic bandgap structure as input sample of model21Considered as output samples of the model;
92) sample pretreatment: normalizing original sample data to be in the range of [ -1,1] through a Sigmoid function, and then taking the sample as a training sample;
93) k-means clustering of training samples: randomly selecting K initial clustering centers from the training samples, calculating Euclidean distances between the rest sample data and the clustering centers, dividing the training samples into K groups by using a K mean value clustering method, then finding the clustering center closest to the target sample data, and distributing the sample data to the cluster corresponding to the clustering center; then, calculating the mean value of sample data in each cluster as a new cluster center, and calculating the error square sum of all clusters; finally, if the sum of the squares of the total errors is not changed, clustering is finished; otherwise, recalculating the Euclidean distances between the rest sample data and the clustering center and performing next iteration;
94) carrying out improved generalized recurrent neural network training:
respectively calculating the distance between each query point of a group of test data and the center of a K cluster, and selecting K training data as training samples in the cluster closest to the query point, wherein the group of samples are associated with the training of the improved generalized regression neural network; then, the root mean square error RMSE between the NN output and the actual S parameter is continuously reduced by adjusting the expansion factor;
if the RMSE is smaller than the expected value, the expansion factor meeting the simulation precision can be determined, and the accuracy and the reliability of the trained improved generalized regression neural network are correspondingly verified;
otherwise, continuously adjusting the diffusion factor to train a new improved generalized regression neural network model until the precision meets the preset condition;
and completing the establishment of a complete electromagnetic modeling database based on the electromagnetic band gap structure parameters and the transmission behaviors.
Advantageous effects
Compared with the prior art, the miniaturized partially fractal electromagnetic band gap structure of the high-speed power supply distribution network and the self-adaptive design method thereof have the advantages that the L-bridge EBG embedded unit is deformed by adopting a Koch fractal method, the broadband and high-suppression-degree effects on the noise of the synchronous switch are improved, meanwhile, the area of the embedded EBG structure unit is reduced, further, the damage to a power supply layer is reduced, and the influence on the signal integrity of adjacent transmission lines can be obviously reduced in three-dimensional system-level packaging. In addition, the K-mean generalized neural network is adopted to carry out small-sample and high-precision modeling on the designed EBG, so that the self-adaptive optimization capability of the power distribution network can be obviously improved.
The design method provided by the invention adopts a K-means clustering method to improve the simulation precision of the traditional GRNN. The modified GRNN models the initial EBG structure of the design. After the simulation database is obtained, genetic algorithms are used for optimization to obtain the optimal structural parameters to meet the design requirements. The design method can effectively improve the optimization efficiency and obviously reduce the optimization time. The final EBG structure is automatically designed through K-GRNN and GA, and has the ultra-wideband SSN inhibition performance of 360MHz to 20GHz and the synchronous switching noise inhibition level of-50 dB; the designed structure also reduces the geometric area of the embedded unit of the power supply layer, reduces the damage to the power supply layer while having the effects of wide noise band and high suppression degree of the synchronous switch, and can obviously reduce the influence on the signal integrity of the transmission line of the adjacent layer in the three-dimensional system-in-package.
Drawings
FIG. 1 is a perspective view of a three-dimensional structure of the present invention;
FIG. 2 is a side cross-sectional view of the present invention;
FIG. 3 is a top view of a 2L-bridge electromagnetic bandgap cell of the present invention;
FIG. 4a is a top view of the structure of the embedded small Koch fractal L bridge electromagnetic bandgap structure of the present invention;
FIG. 4b is a top view of the structure of the first embodiment of the electromagnetic bandgap structure with embedded small Koch fractal L-bridge;
FIG. 4c is a top view of a second embodiment of an electromagnetic bandgap structure with embedded miniature Koch fractal L-bridges according to the present invention;
FIG. 5 is an equivalent circuit diagram of the synchronous switching noise suppression structure of the present invention;
FIG. 6 is a flow chart of modeling based on an improved generalized regression neural network in a synchronous switching noise suppression structure according to the present invention;
FIG. 7 is a flow chart of an intelligent optimization design based on an improved generalized regression neural network and a genetic algorithm in a synchronous switch noise suppression structure according to the present invention;
FIG. 8 is a graph comparing the transmission performance results of the synchronous switching noise suppression structure with and without intelligent optimization, respectively;
the antenna comprises a metal grounding plate 101, a metal patch 201, a 202-L-shaped bridge, a polygonal patch 301, a Koch fractal L-bridge 302, a dielectric substrate 401 and a feed through hole 402.
Detailed Description
So that the manner in which the above recited features of the present invention can be understood and readily understood, a more particular description of the invention, briefly summarized above, may be had by reference to embodiments, some of which are illustrated in the appended drawings, wherein:
as shown in fig. 1, fig. 2, fig. 3 and fig. 4, the miniaturized partially fractal electromagnetic bandgap structure of a high-speed power distribution network according to the present invention includes a dielectric substrate 401, a metal ground plate 101 printed on a back surface of the dielectric substrate 401, and a power distribution network printed on a front surface of the dielectric substrate 401. The power distribution network is 9L-bridge electromagnetic band gap structural units which are placed on a dielectric substrate in a 3X 3 mode, feed through holes 402 are formed in the centers of any two of the 9L-bridge electromagnetic band gap structural units, the two feed through holes 402 are connected with the metal grounding plate 101 and the metal paster 201, each L-bridge electromagnetic band gap structural unit comprises a metal paster 201, four L-shaped bridges 202 are connected to the periphery of each metal paster 201, and adjacent metal pasters 201 distributed in the positive directions of the X axis and the Y axis are connected through the L-shaped bridges 202.
A square hollow structure is etched in the center of an L-bridge electromagnetic band gap structure unit located at the position of a noise source and a noise sensitive circuit on the power distribution network, a small Koch fractal L-bridge electromagnetic band gap structure is embedded in the square hollow structure, and the small Koch fractal L-bridge electromagnetic band gap structures are arranged at the 1-1 position and the 3-3 position of the power distribution network.
As shown in FIG. 5, the small electromagnetic band gap structure is etched in the large electromagnetic band gap structure unit, an LC resonator is additionally added on the power distribution network, and the effect of suppressing noise is influenced by the polygonal metal patch plate capacitor Cp,mlPolygonal patch panel inductor Lp,mlAnd Koch fractal L bridge local inductance Lb,mlInfluence. As the physical length of the L-bridge increases, the bridge inductance increases, and the suppression of synchronous switching noise is enhanced.
As shown in fig. 4a, the small Koch fractal L-bridge electromagnetic bandgap structure includes a polygonal metal patch 301, an L-shaped Koch fractal bridge 302 extends from each side of the polygonal metal patch 301, the long side of the L-shaped Koch fractal bridge 302 is parallel to the side of the polygonal metal patch 301 extending from the L-shaped Koch fractal bridge 302, and the adjacent L-shaped Koch fractal bridges 302 are arranged at 90 °.
As shown in fig. 4b, as a first embodiment, after the intelligent design is performed by using the method of the present invention, based on the structure shown in fig. 4a, after a first iteration, four sides of the designed small-sized Koch fractal L-bridge electromagnetic bandgap structure are recessed towards the center thereof to form a triangular shape, wherein each side of the polygonal metal patch 301 is embedded towards the center thereof to form a triangular shape, and the long side of the L-shaped Koch fractal bridge 302 is embedded towards the center of the polygonal metal patch 301 to form a triangular shape.
As shown in fig. 4c, as a second embodiment, after the intelligent design is performed by using the method of the present invention, based on the structure shown in fig. 4a, after a second iteration, four sides of the designed small-sized Koch fractal L-bridge electromagnetic bandgap structure are all recessed towards the center thereof and take the shape of a five-pointed star, wherein each side of the polygonal metal patch 301 is embedded towards the center thereof and takes the shape of a five-pointed star, and the long side of the L-shaped Koch fractal L-bridge 302 is embedded towards the center of the polygonal metal patch 301 and takes the shape of a five-pointed star.
In order to further increase the length, after the method of the present invention is used for intelligent design, based on the structure shown in fig. 4a, after a third iteration, four sides of the electromagnetic band gap structure of the small Koch fractal L-bridge are designed to be positioned at two sides of the pentagram shape and are both recessed to be in a triangular shape, wherein each side of the polygonal metal patch 301 is positioned at two sides of the pentagram shape and is both recessed to be in a triangular shape, and the long side of the L-shaped Koch fractal bridge 302 is positioned at two sides of the pentagram shape and is both recessed to be in a triangular shape.
The Koch fractal can increase the structure length in a limited space, and the fractal change is carried out on the L bridge embedded with the electromagnetic band gap structure, so that the area of the embedded unit can be reduced, the physical length of the L bridge is increased, the bridge connection inductance is further increased, and the inhibition effect is improved. The iteration times of the Koch fractal can also be used as a parameter to be optimized, and in the subsequent optimization design process, the shape of the electromagnetic band gap structure meeting the design requirement is intelligently selected.
As shown in FIG. 5, the lumped parameter of the 9L-bridge electromagnetic bandgap structure units comprises the parallel plate capacitance Cp,lParallel plate inductor Lp,lBridge microstrip capacitor CbrLocal inductance L of bridgebrAnd the gap coupling capacitance C between adjacent metal sheetsg(ii) a Lumped parameter polygonal metal patch plate capacitor C of small Koch fractal L-bridge electromagnetic band gap structure containing elementp,mlPolygonal patch panel inductor Lp,mlKoch fractal L bridge local inductance Lb,ml。
As shown in fig. 6 and 7, there is also provided a method for adaptively designing a miniaturized partially fractal electromagnetic bandgap structure of a high-speed power distribution network, which is characterized in that the method is designed based on an improved generalized recurrent neural network and a genetic algorithm, the genetic algorithm is used to simulate the selection and genetic process in the nature and continuously search for an optimal solution in a region to be optimized, and the improved generalized recurrent neural network is used to replace simulation software to model the designed initial electromagnetic bandgap structure. After the simulation database is obtained, genetic algorithms are used for optimization to obtain the optimal structural parameters to meet the design requirements.
The invention provides a design method of a miniaturized partially fractal electromagnetic band gap structure of a high-speed power distribution network, which is characterized in that a neural network is utilized to simulate the nonlinear relation between the size parameter and the electric response of an electromagnetic structure, and further the function of simulation software in the genetic algorithm optimization process is replaced. Neural networks are intended to achieve the same high accuracy as simulation software, requiring a large number of samples. For electromagnetic structures, the sample acquisition of complex models is more complex and takes a lot of time. Therefore, a K mean value clustering algorithm is added on the basis of the traditional generalized regression neural network, the simulation precision is improved, and the sample demand is reduced.
Which comprises the following steps:
step one, obtaining sample data: and selecting an initially optimized circuit structure and selecting subsequently optimized size parameters according to a circuit design basic principle, and determining a numerical range of the parameters to be optimized.
Secondly, training a K mean generalized regression neural network: and generating a sample in a parameter value range to be optimized, and training the K mean generalized regression neural network.
The training of the K mean generalized regression neural network comprises the following steps:
A1) obtaining sample data: according to the circuit principle analysis, the internal length d of the polygonal patch board and the length L of the L-bridge short edge embedded in the polygonal patch board are selected2Distance L from the starting point of the long side of the L-bridge to the edge of the polygonal patch3Width w of L bridgel2Koch fractal iteration number i, operating frequency fcS of electromagnetic bandgap structure as input sample of model21Are considered as output samples of the model. The corresponding numerical ranges are shown in table 1.
TABLE 1 sensitive structural parameter value Range Table for training and testing samples
A2) Sample pretreatment: normalizing original sample data to be in the range of [ -1,1] through a Sigmoid function, and then taking the sample as a training sample;
A3) k-means clustering of training samples: randomly selecting K initial clustering centers from the training samples, calculating Euclidean distances between the rest sample data and the clustering centers, dividing the training samples into K groups by using a K mean value clustering method, then finding the clustering center closest to the target sample data, and distributing the sample data to the cluster corresponding to the clustering center; then, calculating the mean value of sample data in each cluster as a new cluster center, and calculating the error square sum of all clusters; finally, if the sum of the squares of the total errors is not changed, clustering is finished; otherwise, recalculating the Euclidean distances between the rest sample data and the clustering center and performing next iteration;
A4) carrying out improved generalized recurrent neural network training:
respectively calculating the distance between each query point of a group of test data and the center of a K cluster, and selecting K training data as training samples in the cluster closest to the query point, wherein the group of samples are associated with the training of the improved generalized regression neural network; then, the root mean square error RMSE between the NN output and the actual S parameter is continuously reduced by adjusting the expansion factor;
if the RMSE is smaller than the expected value, the expansion factor meeting the simulation precision can be determined, and the accuracy and the reliability of the trained improved generalized regression neural network are correspondingly verified;
otherwise, continuously adjusting the diffusion factor to train a new improved generalized regression neural network model until the precision meets the preset condition;
and completing the establishment of a complete electromagnetic modeling database based on the electromagnetic band gap structure parameters and the transmission behaviors.
And thirdly, randomly generating 20 individuals according to a set coding rule in the numerical range of the parameter to be optimized to construct a first generation population. The encoding rule is as follows:
the length k of the chromosome can be calculated using the following formula:
2k-1＜(U2-U1)×10n≤2k-1
wherein [ U ] is1，U2]Is the value range of the structural parameter and n represents the number of decimal places.
The decimal value may then be converted to a binary chromosome by the equation:
and fifthly, decoding chromosomes in the population into decimal values, importing the decimal values into the trained improved generalized regression neural network, and outputting corresponding S parameters of the individuals.
And sixthly, writing a fitness function according to design requirements, importing the S parameter of the individual into a fitness function equation, and calculating the fitness value corresponding to the individual.
And seventhly, calculating the adaptation degree of the individuals in the population to the environment, namely a fitness function value, in order to evaluate the performance of the individuals. Calculating the fitness value of the individual through a set objective function, determining the excellent degree of the individual after sequencing and comparison, and further selecting whether to transmit the individual as a parent gene to a next generation population. The fitness function is as follows:
and eighthly, evaluating the individual performance according to the fitness value, and determining the optimal structural parameters when the fitness value reaches the highest value and meets the expected design performance. Otherwise, populations will be generated by selection, crossover and mutation methods. And after the new generation of population is generated, returning to the fourth step until an optimal solution meeting the design target is found.
As shown in fig. 8, the miniaturized fractal electromagnetic bandgap structure of the high-speed power distribution network provided by the present invention is compared with the transmission characteristic results after algorithm optimization and without algorithm optimization, and the shape and geometric parameters of the EBG built-in unit are changed by using the K-GRNN and GA combined design algorithm, so that the surface impedance of the PDN can be increased, and the propagation of the SSN is suppressed. The best solution obtained by GA and K-GRNN optimization is to reduce the edge length of the embedded elements from 18.5mm to 13.7 mm. By using the K-GRNN and GA combined design algorithm to design the EBG structure, the suppression performance of the SSN is obviously improved under the condition of keeping a small area of the embedded unit. Compared with the initial EBG structure, the low-frequency cut-off frequency of the final structure is reduced to 360MHz, the bandwidth is expanded to 20GHz, and the suppression level is-50 dB.
The power distribution network is composed of an L-shaped bridge electromagnetic band gap structure, a square fractal hollow structure is etched only at the positions of a noise source and a noise sensitive circuit by using local topological network layout, and a miniaturized and deformed Koch fractal L-shaped bridge electromagnetic band gap structure is embedded in the square fractal hollow structure. The invention adopts a Koch fractal method to deform the embedded unit of the L-bridge EBG, improves the effects of wide noise band and high suppression degree of the synchronous switch, reduces the area of the embedded EBG structural unit, further reduces the damage to a power supply layer, can obviously reduce the influence on the signal integrity of adjacent transmission lines in three-dimensional system-level packaging, and adopts a K-mean generalized neural network to carry out small-sample and high-precision modeling on the designed EBG, thereby obviously improving the self-adaptive optimization capability of the power distribution network.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are merely illustrative of the principles of the invention, but that various changes and modifications may be made without departing from the spirit and scope of the invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (9)
1. A miniaturized partial fractal electromagnetic band gap structure of a high-speed power distribution network comprises a dielectric substrate (401), wherein a metal grounding plate (101) is printed on the back surface of the dielectric substrate (401), and a power distribution network is printed on the front surface of the dielectric substrate; the power distribution network is 9L-bridge electromagnetic band gap structural units which are placed on a dielectric substrate in a 3X 3 mode, feed through holes (402) are formed in the centers of any two of the 9L-bridge electromagnetic band gap structural units, the two feed through holes (402) are connected with a metal grounding plate (101) and a metal patch (201), each L-bridge electromagnetic band gap structural unit comprises a metal patch (201), four L-shaped bridges (202) are connected to the periphery of each metal patch (201), and adjacent metal patches (201) distributed in the positive directions of the X axis and the Y axis are connected through the L-shaped bridges (202); the method is characterized in that:
a square hollow structure is etched in the center of an L-bridge electromagnetic band gap structure unit located at the position of the noise source and the noise sensitive circuit on the power distribution network, and a small Koch fractal L-bridge electromagnetic band gap structure is embedded in the square hollow structure.
2. The miniaturized partially-fractal electromagnetic bandgap structure of high-speed power distribution network as claimed in claim 1, wherein: the small Koch fractal L bridge electromagnetic band gap structure comprises a polygonal metal patch (301), an L-shaped Koch fractal bridge (302) extends from each side of the polygonal metal patch (301), the long side of the L-shaped Koch fractal bridge (302) is parallel to the side of the polygonal metal patch (301) extending from the long side, and the adjacent L-shaped Koch fractal bridges (302) are arranged at an angle of 90 degrees.
3. The miniaturized partially-fractal electromagnetic bandgap structure of high-speed power distribution network as claimed in claim 2, wherein: four sides of the small Koch fractal L bridge electromagnetic band gap structure are inwards sunken towards the center of the small Koch fractal L bridge electromagnetic band gap structure to form a triangular shape, wherein each side of the polygonal metal patch (301) is embedded towards the center of the small Koch fractal L bridge electromagnetic band gap structure to form a triangular shape, and the long sides of the L-shaped Koch fractal L bridge (302) are embedded towards the center of the polygonal metal patch (301) to form a triangular shape.
4. The miniaturized partially-fractal electromagnetic bandgap structure of high-speed power distribution network as claimed in claim 2, wherein: four sides of the small Koch fractal L bridge electromagnetic band gap structure are inwards sunken towards the center of the small Koch fractal L bridge electromagnetic band gap structure to form a pentagram shape, wherein each side of the polygonal metal patch (301) is embedded towards the center of the small Koch fractal L bridge electromagnetic band gap structure to form a pentagram shape, and long sides of the L-shaped Koch fractal L bridge (302) are embedded towards the center of the polygonal metal patch (301) to form a pentagram shape.
5. The miniaturized partially-fractal electromagnetic bandgap structure of high-speed power distribution network as claimed in claim 4, wherein: four sides of the small Koch fractal L bridge electromagnetic band gap structure are positioned on two sides of a pentagram shape and are inwards sunken to form a triangular shape, wherein each side of the polygonal metal patch (301) is positioned on two sides of the pentagram shape and is inwards sunken to form the triangular shape, and a long side of the L-shaped Koch fractal bridge (302) is positioned on two sides of the pentagram shape and is inwards sunken to form the triangular shape.
6. The miniaturized, partially-divided electromagnetic strip of high-speed power distribution network as claimed in claim 2The clearance structure, its characterized in that: the lumped parameter of the 9L-bridge electromagnetic band gap structural units comprises a parallel plate capacitor Cp,lParallel plate inductor Lp,lBridge microstrip capacitor CbrLocal inductance L of bridgebrAnd the gap coupling capacitance C between adjacent metal sheetsg(ii) a The lumped parameter of the small Koch fractal L-bridge electromagnetic band gap structure containing element is provided with a polygonal metal patch plate capacitor Cp,mlPolygonal patch board capacitor Lp,mlKoch fractal L bridge local inductance Lb,ml。
7. The miniaturized partially-fractal electromagnetic bandgap structure of high-speed power distribution network as claimed in claim 2, wherein: the small Koch fractal L bridge electromagnetic band gap structure uses a local topological network layout, and square small and deformed Koch fractal electromagnetic band gap structures are etched only at the positions of a noise source and a noise sensitive circuit, namely 1-1 position and 3-3 position of a power distribution network.
8. The self-adaptive design method of the miniaturized partially fractal electromagnetic band gap structure of the high-speed power distribution network according to claim 1, characterized in that based on the improved K-means generalized regression neural network and the genetic algorithm design, the genetic algorithm is utilized to simulate the selection in nature and the genetic process to continuously search the optimal solution in the region to be optimized, and the improved generalized regression neural network is used to replace the simulation software to model the designed initial electromagnetic band gap structure; after obtaining the simulation database, the genetic algorithm is used for optimization to obtain the optimal structural parameters to meet the design requirements, which includes the following steps:
81) obtaining sample data: selecting an initially optimized circuit structure and selecting subsequently optimized size parameters according to a circuit design basic principle, and determining a numerical range of the parameters to be optimized;
82) training the K mean generalized regression neural network: generating a sample in a parameter value range to be optimized, and training a K mean generalized regression neural network;
83) randomly generating 20 individuals according to a set coding rule within a parameter value range to be optimized to construct a first generation population, wherein the coding rule is as follows:
the length k of the chromosome is calculated using the following formula:
2k-1＜(U2-U1)×10n≤2k-1
wherein [ U ] is1，U2]Is the value range of the structural parameter, n represents the decimal place number;
the decimal value is then converted to a binary chromosome by the following equation:
84) then, decoding chromosomes in the population into decimal values, importing the decimal values into the trained improved generalized regression neural network, and outputting corresponding S parameters of individuals;
84) writing a fitness function according to design requirements, importing an S parameter of an individual into a fitness function equation, and calculating a fitness value corresponding to the individual;
85) in order to evaluate individual performance, the adaptation degree of individuals in the population to the environment is calculated, namely an adaptation function value, the adaptation degree value of the individuals is calculated through a set target function, after sequencing and comparison, the excellent degree of the individuals is determined, and then whether the individuals are used as a parent gene to be inherited to a next generation population is selected; the fitness function is as follows:
86) evaluating the individual performance according to the fitness value, and determining the optimal structural parameters when the fitness value reaches the highest value and meets the expected design performance; otherwise, generating a population by selecting a cross and variation method; and after the new generation population is generated, returning to the step 84) until an optimal solution meeting the design target is found.
9. The adaptive design method of the miniaturized partially fractal electromagnetic bandgap structure of the high-speed power distribution network according to claim 8, wherein the training of the K-means generalized regression neural network comprises the following steps:
91) obtaining sample data: according to the circuit principle analysis, the internal length d of the polygonal patch board and the length L of the L-bridge short edge embedded in the polygonal patch board are selected2Distance L from the starting point of the long side of the L-bridge to the edge of the polygonal patch3Width w of L bridgel2Koch fractal iteration number i, operating frequency fcS of electromagnetic bandgap structure as input sample of model21Considered as output samples of the model;
92) sample pretreatment: normalizing original sample data to be in the range of [ -1,1] through a Sigmoid function, and then taking the sample as a training sample;
93) k-means clustering of training samples: randomly selecting K initial clustering centers from the training samples, calculating Euclidean distances between the rest sample data and the clustering centers, dividing the training samples into K groups by using a K mean value clustering method, then finding the clustering center closest to the target sample data, and distributing the sample data to the cluster corresponding to the clustering center; then, calculating the mean value of sample data in each cluster as a new cluster center, and calculating the error square sum of all clusters; finally, if the sum of the squares of the total errors is not changed, clustering is finished; otherwise, recalculating the Euclidean distances between the rest sample data and the clustering center and performing next iteration;
94) carrying out improved generalized recurrent neural network training:
respectively calculating the distance between each query point of a group of test data and the center of a K cluster, and selecting K training data as training samples in the cluster closest to the query point, wherein the group of samples are associated with the training of the improved generalized regression neural network; then, the root mean square error RMSE between the NN output and the actual S parameter is continuously reduced by adjusting the expansion factor;
if the RMSE is smaller than the expected value, the extension factor of the simulation precision is determined, and the accuracy and the reliability of the trained improved generalized recurrent neural network are correspondingly verified;
otherwise, continuously adjusting the diffusion factor to train a new improved generalized regression neural network model until the precision meets the preset condition;
and completing the establishment of a complete electromagnetic modeling database based on the electromagnetic band gap structure parameters and the transmission behaviors.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110584668.5A CN113312868A (en) | 2021-05-27 | 2021-05-27 | Miniaturized partially-fractal electromagnetic band gap structure of high-speed power distribution network and self-adaptive design method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110584668.5A CN113312868A (en) | 2021-05-27 | 2021-05-27 | Miniaturized partially-fractal electromagnetic band gap structure of high-speed power distribution network and self-adaptive design method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113312868A true CN113312868A (en) | 2021-08-27 |
Family
ID=77375396
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110584668.5A Pending CN113312868A (en) | 2021-05-27 | 2021-05-27 | Miniaturized partially-fractal electromagnetic band gap structure of high-speed power distribution network and self-adaptive design method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113312868A (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080091389A1 (en) * | 2006-09-13 | 2008-04-17 | Georgia Tech Research Corporation | Systems and methods for electromagnetic band gap structure synthesis |
| US20100085128A1 (en) * | 2008-10-08 | 2010-04-08 | Samsung Electro-Mechanics Co., Ltd. | Electro-magnetic bandgap structure |
| CN102694221A (en) * | 2012-05-31 | 2012-09-26 | 北京航空航天大学 | Miniature embedded S-shaped bridge plane electromagnetic bandgap structure and building method thereof |
| CN102957310A (en) * | 2012-10-26 | 2013-03-06 | 上海交通大学 | Power supply distribution network based on local electromagnetic band gap structure |
| CN206461068U (en) * | 2017-02-27 | 2017-09-01 | 西安电子科技大学 | Insert-type electromagnetic bandgap structure in one kind mixing |
| CN108011161A (en) * | 2017-11-23 | 2018-05-08 | 安徽大学 | Microwave filter based on electromagnetic band gap and complementary split ring resonator structure and design method thereof |
| CN108959779A (en) * | 2018-07-06 | 2018-12-07 | 西安电子科技大学 | Decoupling network design method based on power supply noise time domain analytical analysis |
| CN110147616A (en) * | 2019-05-22 | 2019-08-20 | 安徽大学 | A kind of Wide stop bands noise suppression arrangement and its intelligent optimized design method applied to power distribution network in system in package |
| CN110191572A (en) * | 2019-05-31 | 2019-08-30 | 杭州电子科技大学 | A kind of electromagnetic bandgap structure realized ultra wide band and inhibit simultaneous switching noise |
| CN110225649A (en) * | 2019-05-17 | 2019-09-10 | 杭州电子科技大学 | It is a kind of for inhibiting the electromagnet bandgap structure of simultaneous switching noise |
-
2021
- 2021-05-27 CN CN202110584668.5A patent/CN113312868A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080091389A1 (en) * | 2006-09-13 | 2008-04-17 | Georgia Tech Research Corporation | Systems and methods for electromagnetic band gap structure synthesis |
| US20100085128A1 (en) * | 2008-10-08 | 2010-04-08 | Samsung Electro-Mechanics Co., Ltd. | Electro-magnetic bandgap structure |
| CN102694221A (en) * | 2012-05-31 | 2012-09-26 | 北京航空航天大学 | Miniature embedded S-shaped bridge plane electromagnetic bandgap structure and building method thereof |
| CN102957310A (en) * | 2012-10-26 | 2013-03-06 | 上海交通大学 | Power supply distribution network based on local electromagnetic band gap structure |
| CN206461068U (en) * | 2017-02-27 | 2017-09-01 | 西安电子科技大学 | Insert-type electromagnetic bandgap structure in one kind mixing |
| CN108011161A (en) * | 2017-11-23 | 2018-05-08 | 安徽大学 | Microwave filter based on electromagnetic band gap and complementary split ring resonator structure and design method thereof |
| CN108959779A (en) * | 2018-07-06 | 2018-12-07 | 西安电子科技大学 | Decoupling network design method based on power supply noise time domain analytical analysis |
| CN110225649A (en) * | 2019-05-17 | 2019-09-10 | 杭州电子科技大学 | It is a kind of for inhibiting the electromagnet bandgap structure of simultaneous switching noise |
| CN110147616A (en) * | 2019-05-22 | 2019-08-20 | 安徽大学 | A kind of Wide stop bands noise suppression arrangement and its intelligent optimized design method applied to power distribution network in system in package |
| CN110191572A (en) * | 2019-05-31 | 2019-08-30 | 杭州电子科技大学 | A kind of electromagnetic bandgap structure realized ultra wide band and inhibit simultaneous switching noise |
Non-Patent Citations (6)
| Title |
|---|
| HAO-RAN ZHU ET AL.: "A Novel Vertical Wire-Bonding Compensation Structure Adaptively Modeled and Optimized With GRNN and GA Methods for System in Package", 《IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY》 * |
| 张森文等: "《科学与人文（分形与混沌理论在多学科中的应用）》", 30 September 2020 * |
| 李勇等: "《复杂情感分析方法及其应用》", 30 April 2020 * |
| 杨红波等: "基于改进聚类分析的神经网络集成方法应用研究", 《计算机工程与科学》 * |
| 杨维明等: "基于分形技术的阶跃阻抗微带低通滤波器设计", 《电波科学学报》 * |
| 黄小龙等: "一种抑制同步开关噪声的新颖电磁带隙结构", 《电子科技》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7459638B2 (en) | Absorbing boundary for a multi-layer circuit board structure | |
| US6571184B2 (en) | System and method for determining the decoupling capacitors for power distribution systems with a frequency-dependent target impedance | |
| CN114386610B (en) | Quantum chip and construction method and construction device thereof | |
| CN111539178A (en) | Chip layout design method and system based on neural network and manufacturing method | |
| CN106844924A (en) | Method based on Response Surface Method and genetic algorithm optimization PCB microstrip line constructions | |
| CN110728034B (en) | Antenna rapid multi-target modeling method using multistage cooperative machine learning | |
| CN107832526A (en) | A kind of method of optimization BGA solder joint return losses | |
| CN100428245C (en) | Process for designing high frequency circuits in multiple domains | |
| Soliman et al. | Neural networks-method of moments (NN-MoM) for the efficient filling of the coupling matrix | |
| CN101694675A (en) | Circuit simulation method and system | |
| US20050113952A1 (en) | Design method for antenna and antenna using the same | |
| CN113312868A (en) | Miniaturized partially-fractal electromagnetic band gap structure of high-speed power distribution network and self-adaptive design method thereof | |
| CN103874323A (en) | Coplane electromagnetic band gap plate based on vertical cascade connection and manufacturing method thereof | |
| Guney et al. | Artificial neural networks for calculating the input resistance of circular microstrip antennas | |
| CN112290955A (en) | Grid node coding method and system based on integrated circuit impedance network extraction | |
| CN109948195B (en) | Impedance modeling method for PCB RFID tag bent dipole antenna | |
| KR20210032113A (en) | Transition structure between microstrip and hollow substrate integrated waveguide | |
| US10461387B2 (en) | Systems and methods for improved chip device performance | |
| CN102033976B (en) | Compact electromagnetic band gap structure for avoiding high-speed circuit noise | |
| Shang et al. | One method for via equivalent circuit extraction based on structural segmentation | |
| Chen et al. | Planar EBG structure for broadband suppression of simultaneous switching noise | |
| US11765822B1 (en) | Printed circuit boards with meshed conductive structures | |
| Pantoli et al. | Calibration Strategy for Tunable Devices Based on Artificial Neural Network Modelling | |
| Shankar et al. | Study of Resonant Microstrip Antennas on Artificial Neural Networks | |
| Nguyen et al. | Sparse Wire Grid 3D Printed Patch Antenna |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |