Research On Test Data Partition Compression Of System-on-Chip

The technique of SoC brings a revolution to the design of the VLSI. Since the chip design cycle is greatly shortted, the output increases correspondingly. Due to the SoC the integrity and the velocity of the chip increase rapidly, on the other hand, the band of the ATE is limited, such we have to spend much time on the test of the SoC, the test cost increases largely. So a great chanllenge is presented in the Soc test.Presently, test resource partitioning for SoC offers a promising solution to these problems. There are mainly three kinds of the test resource partition technique: test set compaction, built-in self test and test data compression. Test data compression is a major research in this thesis.Due to the unbalanced distribution of the deterministic bits in the test set, the effectiveness of the compatible compression is impaired seriously. Firstly, a compression algorithm of grouped standard vector difference for test set based on multiple scan chains is proposed. The test set is formatted according to the multiple scan chains, and then it is divided into several groups by the same compatible relations of test vector inner pattens. And the consistent compressions are implemented in earch group. Next, the results are rearranged and the differences vectors are abtained by standard vector difference method. Finally, the differences vectors were compressed using the distance-marking run-length code. The proposed method provides better compression than FDR codes in all cases. At the same time, the number of CSR needed for test vectors difference is much smaller than that of FDR.After that, a dictionary-based test data partition compression algorithm is presented. In this method, the test vectors are partitioned into several fixed-length blocks and the blocks corresponding to the same position in each vector are grouped together, each group is frequency-directed encoded using the dictionary-based algorithm for compression. The index of the dictionary is composed of two parts: group number and inner group index, and the group numbers are generated automaticly by the group counter within decompression architecture, the average length of the dictionary index is shorten effectively. The compression ratio obtained from this method outperforms the FDR code and selective Huffman code, and furthermore, the decompression architecture of this method is simpler than others.Finally, in order to reduce the date volume of dictionary, a dictionary-based method of test data partition statistic coding is proposed. At first, the test vectors are partitioned into several data blocks with fixed-length, and then several successive blocks with the same position in each vector are grouped, finally each block is encoded using the dictionary-based statistic coding algorithm for compression. This method is an improvement to the method of dictionary-based test data partition compression. The compression ratio obtained from this method is higher than that of FDR. The data volume of dictionary is obviously reduced and the overhead of decompression architecture is decreased significantly.