Error control coding for semiconductor memories

Abstract

All modern computers have memories built from VLSI RAM chips. Individually, these devices are highly reliable and any single chip may perform for decades before failing. However, when many of the chips are combined in a single memory, the time that at least one of them fails could decrease to mere few hours. The presence of the failed chips causes errors when binary data are stored in and read out from the memory. As a consequence the reliability of the computer memories degrade. These errors are classified into hard errors and soft errors. These can also be termed as permanent and temporary errors respectively. In some situations errors may show up as random errors, in which both 1-to-O errors and 0-to-l errors occur randomly in a memory word. In other situations the most likely errors are unidirectional errors in which 1-to-O errors or 0-to-l errors may occur but not both of them in one particular memory word. To achieve a high speed and highly reliable computer, we need large capacity memory. Unfortunately, with high density of semiconductor cells in memory, the error rate increases dramatically. Especially, the VLSI RAMs suffer from soft errors caused by alpha-particle radiation. Thus the reliability of computer could become unacceptable without error reducing schemes. In practice several schemes to reduce the effects of the memory errors were commonly used. But most of them are valid only for hard errors. As an efficient and economical method, error control coding can be used to overcome both hard and soft errors. Therefore it is becoming a widely used scheme in computer industry today. In this thesis, we discuss error control coding for semiconductor memories. The thesis consists of six chapters. Chapter one is an introduction to error detecting and correcting coding for computer memories. Firstly, semiconductor memories and their problems are discussed. Then some schemes for error reduction in computer memories are given and the advantages of using error control coding over other schemes are presented. In chapter two, after a brief review of memory organizations, memory cells and their physical constructions and principle of storing data are described. Then we analyze mechanisms of various errors occurring in semiconductor memories so that, for different errors different coding schemes could be selected. Chapter three is devoted to the fundamental coding theory. In this chapter background on encoding and decoding algorithms are presented. In chapter four, random error control codes are discussed. Among them error detecting codes, single* error correcting/double error detecting codes and multiple error correcting codes are analyzed. By using examples, the decoding implementations for parity codes, Hamming codes, modified Hamming codes and majority logic codes are demonstrated. Also in this chapter it was shown that by combining error control coding and other schemes, the reliability of the memory can be improved by many orders. For unidirectional errors, we introduced unordered codes in chapter five. Two types of the unordered codes are discussed. They are systematic and nonsystematic unordered codes. Both of them are very powerful for unidirectional error detection. As an example of optimal nonsystematic unordered code, an efficient balanced code are analyzed. Then as an example of systematic unordered codes Berger codes are analyzed. Considering the fact that in practice random errors still may occur in unidirectional error memories, some recently developed t-random error correcting/all unidirectional error detecting codes are introduced. Illustrative examples are also included to facilitate the explanation. Chapter six is the conclusions of the thesis. The whole thesis is oriented to the applications of error control coding for semiconductor memories. Most of the codes discussed in the thesis are widely used in practice. Through the thesis we attempt to provide a review of coding in computer memories and emphasize the advantage of coding. It is obvious that with the requirement of higher speed and higher capacity semiconductor memories, error control coding will play even more important role in the future.