Pattern Generator Using Genetic Algorithm Biology Essay
This paper presents a familial algorithm based trial vector compression technique. Familial algorithm has been used for optimising the trial vectors generated by a trial form generator. The digital circuits in VLSI system are increasing in complexness. In order to guarantee their dependability these circuits are tested utilizing trial vectors generated by Automatic Test Pattern Generator. The coevals of trial vectors with high mistake coverage is a expensive procedure for big circuits.
An efficient ATPG algorithm reduces the trial form coevals clip and cost, beside the high mistake coverage. The purpose of the algorithm is to cut down the figure of trial forms and therefore diminishing the trial application clip and to cut down the trial power. Experimental consequences showed that the figure of trial forms applied to prove any combinable circuit is reduced by about 50 % by utilizing the proposed algorithm. Hence the trial application clip is reduced without giving the mistake coverage.
Keywords- Automatic Test Pattern Generation ( ATPG ) , Genetic Algorithm ( GA ) , really big graduated table integrated ( VLSI ) , system-on-chip ( SOC ) , Linear Feedback Shift Register ( LFSR ) , circuit under trial ( CUT ) , Automatic Test Equipment ( ATE ) .IntroductionWith the progress of semiconducting material fabrication engineering, the demands of digital VLSI circuits which are composed of 1000000s of Gatess, have led to many challenges during proving. Furthermore, design complexness and the GHz scope of operating frequences make the testing of SOC designs a most demanding challenge. This is because the big french friess require a immense sum of trial informations and disperse a big sum of power during testing, which greatly increases the system cost.
There are many parametric quantities that should be improved in order to cut down the trial cost. That includes the trial power, trial length ( prove application clip ) , test mistake coverage, and trial hardware country operating expense. This work chiefly deals with cut downing the figure of trial forms ( test compression ) used to prove VLSI circuits, which reduces the trial application clip and therefore the trial power.
Test compression is the procedure of cut downing the length of a trial sequence ( or the entire length of a set of trial sequences ) for the circuit. Test compression is of import for cut downing the trial application clip and the volume of trial informations. Test compression processs can be classified into two classs.
Dynamic compression [ 6 ] process is integrated into the trial coevals process heuristics which aimed at cut downing the trial length. Therefore, they attempt to avoid the coevals of excess trial vectors. Inactive compression [ 6 ] processs perform test compression as a station processing measure, independent of the trial coevals procedure. Inactive compression has two advantages, 1. Unlike dynamic compression, inactive compression does non necessitate any alterations to the trial coevals process. 2. Since dynamic compression is based on heuristics and is non guaranteed to accomplish the minimal trial length, inactive compression is utile, even after dynamic compression is applied during trial coevals, to further cut down the length of the trial sequence.
In this work inactive trial compression technique is used. Familial algorithm is used for trial vector compression technique.testing of vlsi circuitsTesting of VLSI circuits is performed utilizing Automatic Test Equipment. ATE applies input stimulation to the circuit under trial and step the end product response. If the ATE observes a response different from the expected response, the CUT fails that trial.
Three constituents of the VLSI testing are input forms, CUT, and stored response as shown in Fig. 1. In VLSI circuits, the stimulations are called trial forms or trial vectors. Testing typically consists ofSet of trial stimulationCircuit Under Test ( CUT )Analyzing end product responsesIf wrong ( fail ) , CUT assumed to be faultyIf right ( base on balls ) , CUT assumed to be fault-freeATPG is a method used to happen an input ( or prove vectors ) sequence that, when applied to a digital circuit, enables examiner to separate between the right behaviour and the faulty circuit behaviour caused by defects. The efficiency of this method is measured by utilizing mistake coverage, computational clip and the length of trial set.Kanad Basu and Prabhat Mishra [ 4 ] had developed a bitmask choice technique for trial informations compaction in order to make maximal duplicate forms ; besides they developed a dictionary choice method which takes into history the bitmask based compaction. Subsets of primary input vectors of limited sizes were used [ 5 ] during proving for mark mistakes.
Chia-Yi Lin, Hsiu-Chuan Lin, and Hung-Ming Chen [ 2 ] proposed a technique for consecutive circuit proving based on the selective trial form compaction. This method can cut down considerable shift-in power by jumping the exchanging signal passing through long scan ironss. Irith Pomeranz and Sudhakar M. Reddy [ 1 ] had proposed inactive Test Data Volume Reduction algorithm in which the compacted trial vectors were obtained mistake imitating each mistake and taking the excess trial vectors which is a complex procedure for big circuits.Input Test StimuliCircuit under trialOutput Response AnalysisEnd productPass/ FailInput signalIFig 1 VLSI circuit provingFamilial algorithm proposed by Goldberg [ 13 ] is specially suited to work out big scale combination optimisation job.
GA has been successfully applied in different countries of VLSI design, particularly in proving such as trial form coevals [ 5 ] .Familial Approach for Test Pattern GenerationTest coevals utilizing deterministic & A ; mistake oriented algorithm is extremely complex and clip devouring. New attacks are needed to cut down executing clip. GA was used for mistake simulation based trial coevals with conditional executing of trial vector under 2 stage strategy in [ 7 ] which is used for consecutive circuits.
Several attacks to prove coevals have been proposed. In [ 8 ] , [ 9 ] the fittingness rating and population marking is low cost and merely based on the mistake coverage of each trial vector. The disadvantage of the technique is that if a dropping mistake simulation is used, by experimentation after about 10 coevals, the generated vectors stop observing staying mistakes. This method has resulted in better concluding trial set, but it is really expensive.
A new operator is used in [ 8 ] , in which, after each coevals, the best vector in population is put on the concluding trial set and so rescored with a new decreasing fittingness.Familial Algorithm IntroductionThe most popular technique in evolutionary calculation is the familial algorithm. Holland proposed GA as a heuristic method based on “ Survival of the fittest ” . GA was discovered as a utile tool for hunt and optimisation job. Space of all executable solutions ( the set of solutions among which the coveted solution resides ) is called hunt infinite.
Each and every point in the hunt infinite represents one possible solution. Therefore each possible solution can be “ marked ” by its fittingness value, depending on the job definition. With Genetic Algorithm one looks for the best solution among a figure of possible solutions represented by one point in the hunt infinite.Basicss of Genetic AlgorithmGA starts by bring forthing an initial population of chromosomes. By and large, the initial population is generated indiscriminately. Then, the familial algorithm loops over an loop procedure to bring forth the population [ 11 ] . Each loop consists of the following four stairss:Choice: The first measure is choosing persons for reproduction.
This choice is based on the comparative fittingness of the persons so that best 1s are frequently chosen for reproduction.Reproduction: This is the 2nd measure where progeny ‘s are produced from the selected persons. For the following coevals new chromosomes are produced by either crossing over or mutant or both.Evaluation: In this measure the fittingness of the new persons ( chromosomes or offspring ‘s ) is evaluated.Replacement: In this last measure, persons from the old population are replaced by the new 1s.The algorithm is stopped when the population converges toward the optimum solution.
Familial Algorithm for Test Pattern GenerationIn this paper, familial algorithm is used for trial form coevals. The trial vectors or trial forms form the population or set of solution for GA. This algorithm is used to happen the optimum solution with high mistake coverage. A random population of n trial vectors or chromosome is foremost generated and fittingness of each person in the population is evaluated based their mistake coverage. By reiterating choice, crossing over, mutant and replacing, new population of chromosomes is created for following coevals. The compacted or optimum trial vectors generated by this algorithm had detected all the mistakes in the VLSI circuits. Results show that the trial set size has been reduced without any debasement in the mistake coverage for the benchmark circuits.
Flowchart for the familial algorithm used in this paper is shown in the fig 2.The stairss involved in the algorithm are1. Initial population: Random population of n chromosomes ( suited solutions for the job ) was created.
Population size for each coevals is chosen to be big plenty in order to guarantee equal diverseness. Here the persons of the population are the trial vectors for each circuit. The length of each chromosome or single is equal to the figure of primary inputs of the CUT. There is ever a trade off between the convergence rate and the hunt infinite. The population size additions with addition in the length of the trial vector. The tabular array 1 shows the approximative population size for different trial vectors.Table Ipopulation sizeTest Vector LengthPopulation size& lt ; 3448515StartGenerate initial population of trial vectorsMeasure the fittingness of each person by executing mistake simulation with mistake droppingOptimization standards met?YesBest Persons obtainedStopNoChoiceCrossing overMutantGenerate new populationFig 2 Flow chart of Genetic Algorithm2. Fitness Function Evaluation: In this measure the fittingness map for each chromosome ( test vector ) in the population is evaluated.
This is given by degree Fahrenheit ( ten ) . Fitness map provides a measuring for the quality of the trial vector. A trial vector that detects the maximal figure of mistakes is chosen as the best trial vector. When mistake dropping is included so a trial vector that detects maximal figure of new mistakes is considered to be the best trial vector.
Based on this fittingness map value the chromosomes ( test vectors ) were selected by the choice procedure for reproduction. The fittingness map is job particular. In this paper it is based on the mistake coverage. The fittingness map is given byF ( x ) = No of mistakes detected by trial vectorEntire figure of mistakesThe optimal solution is obtained for the maximal value of F ( x ) .3.
Creation of New Population: New population is created by reiterating the undermentioned stairss choice, crossing over and mutant.Choice: In choice procedure two parent chromosomes from the population is selected for bring forthing offspring ‘s. Choice of parent chromosomes is based on their fittingness value, better fittingness, the bigger opportunity to acquire selected. The selected chromosomes were used in the crossing over procedure.There are different types of choice procedure. In this paper Tournament Selection strategy is used.Tournament Choice: Tournament selects each parent by taking persons at random, the figure of which we can stipulate by Tournament size, and so taking the best single out of that set to be a parent.
The best person from the tourney is the 1 with the highest fittingness, which is the victor. Tournament competitions and the victor are so inserted into the coupling pool. The tournament competition is repeated until the coupling pool for bring forthing new progeny is filled. The copulating pool comprising of the tourney victor has higher mean population fittingness. This method is more efficient and leads to an optimum solution.Crossing over: Crossing over is the procedure of sing two parents and bring forthing offspring ‘s from them. With a cross chance, cross over of the parents selected by the choice procedure is performed to bring forth new offspring ( kids ) .
If no crossing over was performed, offspring is the exact transcript of parents. After the choice procedure, the population consists of better persons. The three stairss of the crossing over operator are:I. The reproduction operator selects at random a brace of two single strings for the coupling.two.
A cross site is selected at random along the twine length.three. Finally, the place values are swapped between the two strings following the cross site.In this paper individual point crossing over is performed which converges much faster. In individual point crossing over two coupling chromosomes are cut one time at matching points and the subdivisions after the cuts is exchanged between them to organize new offspring ‘s.
Here, a cross-site or crossover point is selected indiscriminately along the length of the parent chromosomes and spots next to the cross-sites are exchanged. Single point crossing over is shown in the Fig 3.Fig 3 Single point crossing overMutant: After crossing over, mutant is performed on the progeny ‘s that were obtained from the crossing over. Mutant prevents the algorithm from making the local lower limit. Mutation is viewed as a background operator to keep familial diverseness in the population.
It introduces new familial constructions in the population by randomly modifying some of its edifice blocks. The mutant varies for different sorts of representation. For binary representation, mutant inverts the spots from ‘0 ‘ to ‘1 ‘ and ‘1 ‘ to ‘0 ‘ with a little mutant chance. The mutant is shown in fig 4.Fig 4 MutantOptimization Standards: The status for the algorithm to halt is called the optimisation or halting standards. In this paper the fillet standard is the entire figure of mistakes detected, which is the amount of the mistakes detected by each trial vector. Fault simulation is performed for each mistake considered and so each trial vector is evaluated based on the mistakes detected.Consequences and treatmentThe combinable circuits were taken as the CUT.
These circuits were simulated in Matlab for mistake coverage. The figure of mistakes detected by each trial vector to the entire figure of mistakes injected in the circuit gives the mistake coverage for each trial vector. 20 mistakes were introduced to the ISCAS’85 combinable benchmark circuit c17. Since it is a 5 input circuit, the entire figure of trial vectors was 32. The Genetic algorithm was applied and found that out of 32 trial vectors merely 5 trial vectors were plenty to observe all the injected mistakes. For the 1 spot adder which is 3 input circuit, has 8 trial vectors. 22 mistakes were introduced to this circuit.
GA was applied and found that merely 4 trial vectors were plenty to observe all the mistakes.1. Experimental consequence shows that the trial form size has been reduced up to 58 % in comparing with other methods inactive TDVR algorithm as shown in table II.Fig 5 Circuit under Test benchmark Circuit – c17TABLE IIResult trial set compressionCutNumber Of Inputs And Patterns AppliedMistake Model Considered And Number of Faults IntroducedNumber Of Test vectors for 100 % Fault coverage% Reduction Of Number Of Test VectorBy TDVRBy GAC175 input32 trial vectorsStuck at mistake theoretical account20 mistakes12558 %1 spot adder3 input8 trial vectorsStuck at mistake theoretical account22 mistakes6433 %DecisionFamilial algorithm has been used for trial form coevals which has reduced the trial set size without any debasement in the mistake coverage. This algorithm has been applied to assorted combinable benchmark circuits like c17, 1 spot full adder and found that the per centum decrease in the trial form was found to be about 50 % when compared to the other trial compression technique like TDVR. The compacted trial set has achieved the same mistake coverage as the original trial set.
This decrease in the trial vectors in bend reduces the trial application clip and therefore the trial power. In future GA can be applied for the consecutive circuits in order to obtain the decreased trial set.
