CN105608257A - Method for generating large-scale optimal pinouts in BGA package based on genetic algorithm - Google Patents

Abstract

The present invention relates to a method for generating large-scale optimal pinouts in a BGA (Ball Grid Array) package based on a genetic algorithm. The method not only effectively solves a problem that a large number of I/O pins, P/G pins must be accommodated within a signal package and even a smaller volume since more feature functions are integrated into a signal subpackage, but also enables large-scale optimal pinouts to be automatically generated, so that the problem of signal integrity can be reduced.

Description

Extensive BGA based on genetic algorithm encapsulates optimum pin distribution generation method

Technical field

The present invention relates to extensive BGA encapsulation technology field, more specifically, relate to a kind of based on genetic algorithmExtensive BGA encapsulate optimum pin distribution generation method.

Background technology

Along with the continuous progress of electronic information technology and Product Process, the chip integration of chip improves constantly, I/ONumber of pins sharply increases, and power consumption also increases thereupon, also stricter to the requirement of integrated antenna package. For fullThe needs of foot product.

BGA encapsulation starts to be applied to produce. BGA is the english abbreviation of Ball-Gird-Array, sphericalEncapsulation, as chip encapsulation technology of new generation, his operation principle is to make battle array in the bottom of packaging body substrateRow soldered ball, as the I/O end and printed substrate (PCB) interconnection of circuit, mounts device phase with traditional pin shapeRatio, bga device has a lot of good characteristics, as more in I/O number, improved and mount yield rate, potentialThereby reduction cost, be conducive to heat radiation, due to the very short transmission path that shortens signal of soldered ball pin, subtractFew lead-in inductance, resistance, thus the whole performance etc. of circuit can be improved.

In recent years, along with continuous maturation and the progress of semiconductor silicon technology, the enriching constantly with the many of performance of productSample, causes the encapsulation of BGA product complicated all the more, and the number of pins of chip sharply increases, and is accompanied by BGA envelopeThe continuous difficulty that signal in dress, power pins distribute, tends to be accompanied by the problem of many signal integrities, asCrosstalk, track collapse electromagnetic interference, radiation etc. How could in BGA encapsulation, realize better pin dividesJoin, seem especially important.

In recent years, in production and chip are arranged, the method for a lot of pin Optimal Distribution is proposed successively,This research be mainly by genetic algorithm quoting in BGA layout realize optimum pin and distribute.

Genetic algorithm (GeneticAlgorithm) be one class use for reference living nature evolution laws (survival of the fittest, winningThe bad genetic mechanism of eliminating) develop and next randomization searching method. It is the J.Holland professor 1975 by the U.S.First year proposes, and its main feature is directly structure objects to be operated, and does not have differentiate and continuousRestriction; There is inherent disguise and better global optimizing ability; Adopt the optimization method of randomization, can be certainlyThe moving search volume that obtains and instruct optimization, adjusts the direction of search adaptively, does not need the rule of determining. LoseThese character of propagation algorithm, by people be widely used in Combinatorial Optimization, machine learning, signal processing, fromAdapt to the fields such as control and artificial life, it is modern about the key technology in intelligence computation.

Arranging in process at BGA pin, how to reduce as far as possible time loss and confirm global optimum, is mePrimary study direction. What mainly take is on the basis of genetic algorithm, to use static template to generate fast optimumPin.

Summary of the invention

The object of the invention is to overcome a large amount of pins arranges and causes complicated package design or even seriousThe phenomenon of problems of Signal Integrity, provides a kind of extensive BGA based on genetic algorithm to encapsulate optimum pin and dividesCloth generation method, this method not only solves effectively along with more power function is integrated in signal packing,A large amount of I/O pins, P/G pin must be contained in this problem in the volume that signal assemble is even less, andAnd can also automatically bear on a large scale optimum pin and distribute, to reduce the problem of signal integrity.

For realizing above goal of the invention, the technical scheme of employing is:

Extensive BGA based on genetic algorithm encapsulates an optimum pin distribution generation method, comprises following stepRapid:

S1. the BGA that is M*N for number of pins encapsulation, its pin distributes next by the matrix D of M*NRepresent, wherein represent respectively power supply signal pin, earth signal pin and spacing wave pin with 1,2,0;

S2. selected K encapsulation scheme, and represent that respectively its pin distributes, K envelope with the matrix of M*NDress forecast scheme configuration population Chrom, encapsulation scheme is as the individuality in population Chrom, this seasonal iterationst＝0；

S3. K individual population Chrom random alignment will be contained;

S4. the individual fitness value of population Chrom is evaluated to detection;

S5. the individuality in population Chrom is compared between two according to its fitness value, will be between two suitable in relativelyThe good individuality of response is put into the container of WinChrom, and in the container of WinChrom, total N is individual;

S6. make the individuality in WinChrom mate intersection, coupling intersect position by random number m1,M2, n1, n2 determine, if m1 is less than m2, the line position of the intersection region of a volume matrix arrives in m1+1Between m2, if m1 is greater than m2, the line position of the intersection region of a volume matrix is in 1 to m2 or m1+1Between K, n1, n2 are for determining the row of intersection region, the same m1 of its principle, m2. Can confirm thusJoin intersection region;

S7. judge whether power supply signal pin, earth signal number of pins that the individual relative after intersecting is answered have occurredChange, continue if not execution step S8;

S8. make the individuality in WinChrom make a variation, first calculate variation Probability p_m：

$p_m=\left\{\begin{array}{c}0.5,f_i<\stackrel{\&OverBar;}{f}\\ 0.5\&times;\frac{f_i-\stackrel{\&OverBar;}{f}}{f_{Max}-\stackrel{\&OverBar;}{f}}\\ 0.1,f_i=f_{Max}\end{array}\right.,f_{Max}>f_i>\stackrel{\&OverBar;}{f}$

Wherein f_Max、f_iFor adapting to most individual adaptive value, population mean P in population WinChrom respectivelyIndividual adaptive value should be worth, make a variation;

S8. in variation probability, judge whether individual power supply signal pin, the earth signal pin of variation equals to becomeTotal number of pins of different individuality, if by certain power supply signal pin and earth signal pin transposition, if not,Utilize rand function random number that produces between (0,1), if this number is greater than Pm, by individuality certainIndividual power supply signal pin and earth signal pin transposition, otherwise by certain power supply signal pin in individuality or groundSignal pins and spacing wave pin switch;

S9.N WinChrom carries out producing (K-N) individual WinChrom offspring after cross-matched, variation,By N WinChrom and (K-N) individual WinChrom offspring combination, form population NewChrom of new generation;

S10. calculate each individual adaptive value in population NewChrom, select the best individuality of adaptive value to beBestInd；

S11. make t=t+1, using population NewChrom as initial population repeated execution of steps S3～S10 untilt>k；

S12. preserve the optimum individual BestInd of each iteration in container B estInd_Temp, by relatively dividingAnalyse the adaptive value of each BestInd in BestInd_Temp and obtain optimal pin assignment.

Preferably, the adaptive value of described step S10 is defined as

f(x)＝w×M_sum(A)+D_sum(A)w∈[0，∞)；

Wherein w is non-negative weight coefficient, M_sum(A) be individual total inductance, D_sum(A) for individuality returns to roadPath quality.

Preferably, described weight coefficient w=0.75.

Preferably, individual total inductance M_sum(A) be expressed as follows:

$\begin{array}{c}{M}_{sum}=\underset{i=1}{\overset{G}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{G}{\mathrm{\&Sigma;}}}{a}_{ij}{M^{\&prime;}}_{p_{ij}}\\ =\underset{i=1}{\overset{G}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{G}{\mathrm{\&Sigma;}}}{a}_{ij}\ln \left(\frac{d_{Max}}{d_{ij}}\right)\end{array}$

a_ijFor-1,0,1, represent respectively pin i and pin, j is at direction, pin i=pin j, the pin of oppositionI is with pin j in these three kinds of different situations of identical direction, and G represents total number of pins, d_ijRepresent pin iAnd the distance between pin j, M '_pijRepresent the local mutual inductance of the pin of the capable j row of i, d_MaxFor fixed value,Represent the maximum of pin-pitch.

Preferably, described individual return path quality representation is as follows:

$D_{sum}=\mathrm{var}\left(D_{\min }\right)-\underset{i=1}{\overset{G}{\mathrm{\&Sigma;}}}\frac{d_j}{G}$

Wherein D_minThe vector that in representing matrix, the least member of every row forms, G represents pin total number, d_jLeast member in j row in representing matrix D, var (D_min) expression D_minMean square deviation.

Preferably, in described step S12, by fitting of each BestInd in comparative analysis BestInd_TempShould be worth that to obtain the process of optimal pin assignment specific as follows:

Whether the number that S121. judges BestInd in BestInd_Temp is greater than 1 is less than 8, if carry out stepRapid S123, otherwise execution step S122;

S122. export the individuality of adaptive value maximum;

S123. judge in BestInd_Temp whether have identical individuality, if export the individual of adaptive value maximumBody, otherwise execution step S121.

Compared with prior art, the invention has the beneficial effects as follows:

(1) greatly improve the efficiency that pin distributes, improved signal integrity;

(2) human eye can directly carry out visual assessment to final result;

(3) shorten time of whole extensive BGA package design, can extensively be useful in modern extensive high propertyIn energy system in package.

Brief description of the drawings

Fig. 1 is the flow chart of optimum pin distribution generation method.

Fig. 2 is the interconversion relation judgement figure of signal of telecommunication pin, earth signal pin, spacing wave pin.

Fig. 3 obtains optimal pin by the adaptive value of each BestInd in comparative analysis BestInd_Temp to divideThe process schematic diagram of joining.

Fig. 4 uses matrix notation pin to divide other exemplary plot.

Fig. 5 is the schematic diagram of the algorithm of tournament selection after improvement.

Fig. 6 is definite exemplary plot of intersection region.

Detailed description of the invention

Accompanying drawing, only for exemplary illustration, can not be interpreted as the restriction to this patent;

Below in conjunction with drawings and Examples, the present invention is further elaborated.

Embodiment 1

As shown in Figure 1, optimum pin distribution generation method is specifically implemented as follows:

1. initialize population

The BGA encapsulation that a given number of pins is M*N size, we can be by a M*N for this pin clothMatrix notation, can transmit by proposing a kind of matrix coder mode the rule that in BGA encapsulation, pin distributes here,Respectively with 1,2,0 replace power supply,, other signal pins arrange rule as shown in Figure 4. Pass throughUse the mode of matrix coder, first we are than being easier to obtain in matrix the distance of pair of pins arbitrarily, therebyFacilitate our calculating target function; Secondly be easier to realize for the operation ratio of gene pin.

In addition, between each individuality of initial population, should maintain a certain distance, and define equal length withA certain constant is that in two character strings of base, corresponding different quantity is broad sense Hamming distance between the two, this requirementThe bright sea of broad sense distance between the individuality of selected initial population must be greater than the value of certain setting. Adopt in this wayCan ensure has obvious difference between random each individuality producing, and makes them can be evenly distributed on solution spaceIn, increase the possibility of obtaining globally optimal solution. Simultaneously the size of initial population is also by the whole genetic algorithm of impactEfficiency, scale is larger, can obtain globally optimal solution, but meanwhile aspect time loss, can pay moreLarge cost.

2. select

By algorithm of tournament selection method to the selection of reducing by half of initial population individuality. Again with traditional algorithm of tournament selectionAlgorithm has difference, and traditional algorithm of tournament selection easily reduces the diversity of early stage population, strangles other may have entirelyThe gene of office's optimal solution, causes genetic algorithm to restrain in advance; May, the later stage, certain be connect very much againThe individuality of nearly globally optimal solution, because of crossover and mutation, on the contrary away from optimal solution, causes evolutionary rate very slow,Even cannot converge to globally optimal solution. The algorithm of tournament selection of improvement version as shown in Figure 5, specific implementation process asUnder:

(1) M individual population Chrom random alignment will be contained;

(2) individuality is compared between two, being 1 as sequence number compares with 2, and sequence number is 3,4Compare, each group relevance grade best individuality is put into the container (N altogether) of WinChrom,If last individuality of odd number of individual is directly stored in WinChrom.

(3) N WinChrom carried out to mutual crossover and mutation computing, produce (M-N) individual WinChromOffspring.

(4) by N WinChrom and (M-N) individual WinChrom offspring combination, form a new generation and plantGroup NewChrom.

3. intersect

Intersection is the built-in function that encoder matrix is carried out, and it must standard keep three principles 1. between generations simultaneouslyMust ensure that high heredity is in the time of interlace operation; 2. interlace operation at random existence always change whole GAAbility of searching optimum; 3. interlace operation should not destroy chromosomal validity. Specifically be implemented as follows:

(1) produce at random m₁、m₂、n₁、n₂If, m₁Be less than m₂, the line position of intersection region is in m₁+1To m₂Between; If m₁Be greater than m₂, the line position of intersection region is in 1 to m₂Or m₁Between+1 to M.For the row n of intersection region₁With n₂In like manner, as shown in Figure 6.

(2) carry out legitimacy matching judgment, the individual A power supply ground number of pins corresponding with B after intersectionThere is change, in order to maintain individual legitimacy, cross one another individual guarantee will with original power supply drawPin number is identical. Can be divided into nine kinds of situation discussion, reach pin style specifications.

(3) intersection offspring and prechiasmal population at individual are combined, form new population, can deriveThe individual while making new advances also can ensure that already present optimal base is because retaining, and can not lose because of interlace operation excellentElegant individual.

4. variation

Variation is the process of an accelerating ated test, in the last optimizing process of genetic algorithm, can keep populationDiversity and prevent the Premature Convergence of population.

In general genetic manipulation, fixing variation probability P m has had a strong impact on convergence of algorithm performance, ifPm is too small, is just not easy to produce new individuality, causes Evolution of Population slow; If Pm is excessive, that willDestroy the heritability that interlace operation brings, caused search to become pure random search algorithm. Here adoptWith a kind of self adaptation adjustment based on the average relevance grade of ideal adaptation degree and colony variation probability:

$p_m=\left\{\begin{array}{c}0.5,f_i<\stackrel{\&OverBar;}{f}\\ 0.5\&times;\frac{f_i-\stackrel{\&OverBar;}{f}}{f_{Max}-\stackrel{\&OverBar;}{f}}\\ 0.1,f_i=f_{Max}\end{array}\right.,f_{Max}>f_i>\stackrel{\&OverBar;}{f}$

Wherein f_Max、f_iBe respectively and in a Chorm of colony, adapt to individual adaptive value most, plant group meanAdaptive value, the individual adaptive value that makes a variation, can find out from above formula, adaptive value is than little being endowed of mean valueHigh variation probability, produces new individuality for suddenling change, and explores the farther solution in space far away; Adaptive value is than averageThe individuality that value is high has less variation probability, outstanding gene is retained and intersect the next generation,Variation, the strategy of the outstanding gene of this reservation is more conducive to explore local optimum. Concrete mutation process is as follows:

In variation probability, judge whether power supply ground number of pins equals total number of pins, if certain power supply is drawnPin and ground pin transposition, if not, utilize rand function random number that produces between (0,1),If this number is greater than 0.7, by certain power supply signal pin in individuality and earth signal pin transposition, otherwise willCertain power supply signal pin or earth signal pin and spacing wave pin switch in individuality; Specific implementation methodAs shown in Figure 2.

5. preferred

To wait sequence of operations population and initial population to be afterwards combined into number of individuals with variation through intersectingThe new population that order is double, and calculate whole individual adaptive values based on two object functions in new population, countA best individuality is BestInd, and definition adaptive value is:

f(x)＝w×M_sum(A)+D_sum(A)w∈[0，∞)；

Wherein w is a non-negative weight coefficient, M_sum(A) be individual total inductance, D_sum(A) for individuality returnsLoop path quality, is a very important step for suitable weight coefficient is set, if w is too little, and whole something lostPropagation algorithm will more be paid close attention to return path quality, and ignore the mutual inductance between each pin, if weight coefficient w tooGreatly, algorithm can more be paid attention to the inductance between each pin, thereby ignores return path quality. Through a series of realityTest comparison comparative analysis, w=0.75 can reasonable balance both sides relation, and adaptive value can apply to variousThe power supply ground pin assignment ratio of various kinds.

Iterative cycles is carried out GA, if population is without any evolution in the circulation in continuous 80 generations, retains outstandingIndividual BestInd, in container B estInd_Temp, then resets to population comprehensively, by comparative analysisIn BestInd_Temp, the adaptive value of each BestInd obtains optimal pin assignment, specific implementation method asShown in Fig. 3.

Obviously, the above embodiment of the present invention is only for example of the present invention is clearly described, and is notIt is the restriction to embodiments of the present invention. For those of ordinary skill in the field, in above-mentioned explanationBasis on can also make other changes in different forms. Here without also cannot be to all enforcementMode gives exhaustive. All any amendments of doing within the spirit and principles in the present invention, be equal to replace and improveDeng, within all should being included in the protection domain of the claims in the present invention.

Claims (6)

1. the extensive BGA based on genetic algorithm encapsulates an optimum pin distribution generation method, and its feature existsIn: comprise the following steps:

S1. the BGA that is M*N for number of pins encapsulation, its pin distributes next by the matrix D of M*NRepresent, wherein represent respectively power supply signal pin, earth signal pin and spacing wave pin with 1,2,0;

S2. selected K encapsulation scheme, and represent that respectively its pin distributes, K envelope with the matrix of M*NDress forecast scheme configuration population Chrom, encapsulation scheme is as the individuality in population Chrom, this seasonal iterationst＝0；

S3. K individual population Chrom random alignment will be contained;

S4. the individual fitness value of population Chrom is evaluated to detection;

S5. the individuality in population Chrom is compared between two according to its fitness value, will be between two suitable in relativelyThe good individuality of response is put into the container of WinChrom, and in the container of WinChrom, total N is individual;

S6. make the individuality in WinChrom mate intersection, coupling intersect position by random number m1,M2, n1, n2 determine, if m1 is less than m2, the line position of the intersection region of a volume matrix arrives in m1+1Between m2, if m1 is greater than m2, the line position of the intersection region of a volume matrix is in 1 to m2 or m1+1Between K, n1, n2 are for determining the row of intersection region, the same m1 of its principle, m2. Can confirm thusJoin intersection region;

S7. judge whether power supply signal pin, earth signal number of pins that the individual relative after intersecting is answered have occurredChange, continue if not execution step S8;

S8. make the individuality in WinChrom make a variation, first calculate variation Probability p_m：

$p_m=\left\{\begin{array}{cc}0.5,& f_i<\stackrel{\&OverBar;}{f}\\ 0.5\&times;\frac{f_i-\stackrel{\&OverBar;}{f}}{f_{Max}-\stackrel{\&OverBar;}{f}},& f_{Max}>f_i>\stackrel{\&OverBar;}{f}\\ 0.1,& f_i=f_{Max}\end{array}\right.$

WhereinFor adapting to most individual adaptive value, population mean P in population WinChrom respectivelyIndividual adaptive value should be worth, make a variation;

S8. in variation probability, judge whether individual power supply signal pin, the earth signal pin of variation equals to becomeTotal number of pins of different individuality, if by certain power supply signal pin and earth signal pin transposition, if not,Utilize rand function random number that produces between (0,1), if this number is greater than Pm, by individuality certainIndividual power supply signal pin and earth signal pin transposition, otherwise by certain power supply signal pin in individuality or groundSignal pins and spacing wave pin switch;

S9.N WinChrom carries out producing (K-N) individual WinChrom offspring after cross-matched, variation,By N WinChrom and (K-N) individual WinChrom offspring combination, form population NewChrom of new generation;

S10. calculate each individual adaptive value in population NewChrom, select the best individuality of adaptive value to beBestInd；

S11. make t=t+1, using population NewChrom as initial population repeated execution of steps S3～S10 untilt>k；

S12. preserve the optimum individual BestInd of each iteration in container B estInd_Temp, by relatively dividingAnalyse the adaptive value of each BestInd in BestInd_Temp and obtain optimal pin assignment.

2. it is raw that the extensive BGA based on genetic algorithm according to claim 1 encapsulates optimum pin distributionOne-tenth method, is characterized in that: the adaptive value of described step S10 is defined as

f(x)＝w×M_sum(A)+D_sum(A)w∈[0，∞)；

Wherein w is non-negative weight coefficient, M_sum(A) be individual total inductance, D_sum(A) for individuality returns to roadPath quality.

3. it is raw that the extensive BGA based on genetic algorithm according to claim 2 encapsulates optimum pin distributionOne-tenth method, is characterized in that: described weight coefficient w=0.75.

4. it is raw that the extensive BGA based on genetic algorithm according to claim 2 encapsulates optimum pin distributionOne-tenth method, is characterized in that: individual total inductance M_sum(A) be expressed as follows:

$\begin{array}{c}{M}_{sum}=\underset{i=1}{\overset{G}{\&Sigma;}}\underset{j=1}{\overset{G}{\&Sigma;}}{a}_{ij}{M^{\&prime;}}_{p_{ij}}\\ =\underset{i=1}{\overset{G}{\&Sigma;}}\underset{j=1}{\overset{G}{\&Sigma;}}{a}_{ij}\ln \left(\frac{d_{Max}}{d_{ij}}\right)\end{array}$

a_ijFor-1,0,1, represent respectively pin i and pin j direction, pin i=pin j, the pin in oppositionI is with pin j in these three kinds of different situations of identical direction, and G represents total number of pins, d_ijRepresent pin iAnd the distance between pin j, M '_pijRepresent the local mutual inductance of the pin of the capable j row of i, d_MaxFor fixed value,Represent the maximum of pin-pitch.

5. it is raw that the extensive BGA based on genetic algorithm according to claim 2 encapsulates optimum pin distributionOne-tenth method, is characterized in that: described individual return path quality representation is as follows:

$D_{sum}=\mathrm{var}\left(D_{\min }\right)-\underset{i=1}{\overset{G}{\&Sigma;}}\frac{d_j}{G}$

Wherein D_minThe vector that in representing matrix, the least member of every row forms, G represents pin total number, d_jLeast member in j row in representing matrix D, var (D_min) expression D_minMean square deviation.

6. it is raw that the extensive BGA based on genetic algorithm according to claim 1 encapsulates optimum pin distributionOne-tenth method, is characterized in that: in described step S12, by each in comparative analysis BestInd_TempIt is specific as follows that the adaptive value of BestInd obtains the process of optimal pin assignment:

Whether the number that S121. judges BestInd in BestInd_Temp is greater than 1 is less than 8, if carry out stepRapid S123, otherwise execution step S122;

S122. export the individuality of adaptive value maximum;

S123. judge in BestInd_Temp whether have identical individuality, if export the individual of adaptive value maximumBody, otherwise execution step S121.