The Research Of SoC Test Data Compression Based On Multiple Features Sequence Coding

The technique of design System-on-a-Chip, which is based on IP cores reusing technology, greatly predigests the design of chip and shorts the time for products to come into the market, but with the increasing number of IP cores integrated in a chip, the volume of test data that testing the chips also have been increasing rapidly, leading to test of chip is more and more difficult. Meanwhile, due to the expensive of automatic test equipment, the cost of testing takes a very high weight in the cost of SoC manufacture, so the test of SoC has been drawing more and more attention. To address the problem of the large volumn of test data, a direct and effective solution is to compress the test set.This dissertation, on the basis of the analysis of several test data encoding scheme, proposed two novel encoding schemes. The main work is as follows:(1) Introducting the basic knowledge of test technology and the problems of SoC testing, summarizing the research achievements on test data compression technology at home and abroad. These achievements can be divided into the built-in self-test and build-out self-test. By way of illustration, their principles are analyzed and their fetures are presented.(2) The first scheme is encoding multi-feature sequences. FDR codes only encode run-length of zero, so it is not effective for encoding run of one. Proposed scheme encode four distinct sequences. In this scheme, the test sets are seen as data flows consisted of these four distinct sequences, in this way, it can reduces the occurrences of encoding short sequence, improving the efficiency of codes.(3) The second scheme is one of Huffman coding scheme. Due to the variable length input coding scheme only statistic the occurrence of run-length of zero, this scheme provide a novel set base on the alternative run-length to statistic, and encoding each lengths according to their occurrences.The experimental results show that these two schemes are both efficient on test data compression. And these two schemes are both encoding the pre-computed test set directly without difference operation, so their decompression architecture does not require CSR.