Cad Tools For Bist Insertion And Analysis

Abstract

This paper presents the analysis and solutions for problems relating to two scenarios in a design incorporating LFSR based BIST. The first design scenario is a BIST structure that has repeated test patterns applied to its LFSR based signature analyzer. The effect of applying repeated patterns on error masking is investigated. Also, guidelines on checking if a given LFSR polynomial is suitable for use in this design scenario are provided. The second design scenario involves connecting a k-output circuit to a r-stage MISR (k < r). The effect of different selections on error masking, and the selection for least error masking is described. Finally, tools for use in BIST design that utilize designer specified limits on error masking are described. 1. Introduct ion Signature analysis is now a well accepted technique in industry for compacting the Circuit Under Test's (CUT) output responses in Built-In Self Test (BIST) designs [l]. It employs a Linear Feedback Shift Register (LFSR) which is a shift register with a feedback network made up of linear elements [2,3]. The serial or parallel output bit stream from the CUT is fed to an LFSR connected at the CUT output. The LFSR contents at the end of vector compaction is a signature, hence the name signature analysis. An LFSR used as a signature analyzer maps the CUT output vector space into a signature. Since the mapping is not one-to-one, it is possible for the output patterns in a faulty CUT to produce the same signature as that in the good CUT. This phenomenon is called error masking and is of two types - aliasing and error cancellation. Considerable research has been done in the area of aliasing for LFSR based signature analysis [4-121. This paper addresses two practical design issues in the implementation of LFSR based BIST - i) effect on aliasing and error cancellation of applying repeated test patterns to an LFSR and ii) effect of tap selection to a MISR on aliasing and error cancellation. The paper gives a quick review of the standard calculations for error masking. These calculations assume the inputs are applied with an independent bit stream. However, in certain cases, the bits in the input stream may not be independent which restricts the number of possible patterns that get applied to the LFSR based signature analyzer. Since not all input streams are possible at the LFSR input, a CAD tool is needed to determine whether the signatures resulting from possible streams are distributeduniformly over all possible signatures. The length of the LFSR is chosen to meet a minimum acceptable level of aliasing. When the CUT has fewer outputs than the number of stages in the MISR, the tap point selection will affect both aliasing and error cancellation. This paper will show the error calculations for optimum tap selection.