【作者】 成丽丽；

【导师】 欧阳一鸣；

【作者基本信息】 合肥工业大学 ， 计算机应用技术， 2008， 硕士

【摘要】 SoC技术的出现,给超大规模集成电路VLSI的设计带来了一场革命,它通过复用大量具有自主知识产权(IP)的芯核,极大地缩短了开发周期,加快了产品上市速度,提高了芯片的产量。随着SoC技术的发展,芯片的集成度迅速提高,芯片测试的复杂性和测试数据量随之激增,而传统的ATE的存储空间、工作频率以及带宽却非常有限,这使得SoC测试面临着测试时间过长、测试难度和测试成本急剧增加等诸多方面问题。目前解决这类问题的一种行之有效的方法是运用测试数据压缩技术,该技术的运用能够达到简化测试实施过程、减少测试实施时间以及降低测试成本的目的。为了充分发挥测试集中无关位多且各测试向量所包含的无关位数量差异大的特点,本文提出了一种新颖的基于变长数据块相关性统计方法,以测试向量为单位,运用统计和数据块相关性的思想,先确定向量的参考数据块,再利用它与该向量中数据块的相关性进行压缩。各向量的参考数据块长度相互独立,且同时具有和该向量中数据块相关性频率最高以及用其对向量进行编码得到的位数最少的特点。该方案运用对不同测试向量分而治之的策略,突破了在整个测试集范围内操作所受的限制。同大部分现有方法相比,该方法更有效地压缩了测试集,且硬件解压结构也较简单,具有较好的适用性。本文随后提出了一种基于连续及非连续长度块编码方法,该方法从提高码字利用率出发,是一种新的固定长度编码方法。将测试集看成若干个连续‘0’序列和‘1’序列,将长度超过被选择的长度k的序列看成连续块,利用一个定长的k位二进制编码表示该块的长度信息;同时,创新性的对短序列进行了处理,即连续长度不足k位的序列,按一定的策略被划为非连续块,并且不对其进行编码。该策略有效地减小了编码中序列长度变化大带来的影响,避免了用长码字替换短序列的情况。同时本方案减小了使用前、后缀形式编码的复杂性,所以其编码及解码过程简单,具有简单的通讯协议。文中对ISCAS-89标准电路原测试集以及差分预处理后的测试集进行了实验,结果表明本方案能获得较高的压缩率,并通过设计的硬件解压结构能顺利还原成原始数据,是一种具有较强适用性的方法。更多还原

【Abstract】 The emergence of SoC technology brings a revolution to the design of the VLSI, through reuse a lot of independent intellectual property (IP) cores, chip design cycle is greatly shortted, the speed to market is accelerated, and the output increases correspondingly. Due to the development of SoC, the integration of the chip increase rapidly, the complexity of chip testing and test data volume also increase, while the storage capacity, frequency and the band of the traditional ATE are very limited, these make SoC test facing some problems, such as the time of test is too long, the difficulty and the costs of test increases rapidly and so on.Presently, test data compression technology is an effective solution to these problems. It can be used to simplify the implementation of the test, reduce the test implementation time and lower the test costs.In order to give full play to the features that the don’t-care bits in the test set are plentiful and the number of them in each test vectors are by large difference, this dissertation presents a new method based on statistic relativity of variable length data blocks, for each test vector, use the thought of statistics and relativity of data blocks, determine the referenced data block of it, then compress the vector by the relativity between its data blocks and the referenced data block. Besides, the length of referenced data block is independent from each other, at the same time, the referenced data block of each vector has the characteristics that it has the highest frequency of relativity and the least code. The proposed method uses the divide and rule tactic with different vectors to break the constraints in the entire test set. Compare with most of the existing method, it can compress the test set more efficiently, and the decompression architecture is relatively simple, so the applicability of this method is better than others.After that, a new fixed length test data compression method based on encoding the data blocks which are sequential or non-continuous is proposed, with intent to improve the efficiency of the code, view the test set as a number of ’0’ sequences and the ’ 1’ sequences, the sequences length over the selected block length k will be seen as sequential blocks, a binary code with fixed length is used to express the length information of the sequential block. On the other hand, the sequences which has lack of it-length is made to be non-continuous data blocks by some strategy, and not be encoded, so it can avoid the situation that short sequences replaced by long code. The rule proposed by the method reduces the complexity of encoding which makes use of prefix and tail code separately, so the process of it’s encoding and decoding is simple, and the protocol of communication is also simple. Experiments have been done in the original and precomputed difference test set of ISCAS-89 benchmark circuits, the result shows that this method can receive a higher compression ratio, the decompression architecture designed can decode the compressed test set to original test set accurately, so the method proposed has much more applicability.更多还原