Research On Generation Algorithm Of Base Vector In Split Compression

In the 1990 s,a more complex chip contained hundreds of thousands to millions of transistors,but now a chip can contain tens of billions of transistors.The increasing integration and complexity of chips directly increase the probability of producing failed chips.In order to ensure that the chip’s failure coverage rate meets the standard,a large amount of test data is required to detect it.The huge test data increases the hardware cost and test time.By compressing the test data,the test time can be greatly reduced and the hardware cost can be saved.This article on the basis of split compression technology,how to generate The basic vector has been researched,and the main work has been done in the following two aspects:(1)A pre-populated strategy is proposed to process the test set.This method eliminates the extraneous bits contained by pre-filling the test set.Filling methods can be divided into two types: direct filling and strategy filling.Direct filling refers to directly filling the irrelevant bits in the original test set as codeword 0 or codeword 1.Strategy filling is to compress the irrelevant bits according to help to improve the specific coding Fill in a ratewise manner.After filling the test set,this paper selects the required base vector based on the principle of the largest distance between vectors.The experimental results show that by using the pre-filled compression method,the compression rate of RLHuff encoding can reach 74.32%,which is an increase of 11.75% compared to the direct encoding of the test set.Increased by 2.47%.(2)A method of generating basis vectors based on kmeans++ clustering algorithm combined with test set is proposed.This method first selects the initial clustering center of the test set based on the principle of maximum distance,then calculates the Euclidean distance between each column vector in the test set and the clustering center,and re-columns the column vectors based on the principle of minimum distance Go to different clusters and calculate new cluster centers,and iterate in this way.When the column vectors in the cluster basically no longer change,the resulting cluster center is the base vector sought in this paper.The experimental results show that the average compression rate of RL-Huff coding can be increased to 76.30% by using the kmeans++ clustering algorithm combined with the test set to generate basis vectors.Compared with the direct compression of the test set,the average compression rate is improved.13.73%,the average compression ratio is increased by 4.45%compared with Hada code conversion.At the same time,I also combined the kmeans++clustering algorithm with the bit flip algorithm to further improve the compression rate.The main idea of this method is to invert part of the determined bits in the original test set into irrelevant bits without affecting the fault coverage,and then XOR the test set with the main component set to obtain the residual component set,thereby improving compression rate.The experimental results show that the average compression rate can be increased by 7%on the basis of kmeans++ by using kmeans++ clustering algorithm combined with bit flip algorithm.