This paper proposes a new environment for high-level VLSI design specification validation using "Algorithmic Debugging" and evaluates its benefits on three significant examples (a protocol processor, an 8-bit CPU, and a Prolog processor). A design is specified at a high-level using the structured analysis (SA) method, which is useful for analyzing and understanding the functionality to be realized. The specification written in SA is transformed into a logic programming language and is simulated in it. The errors (which terminate with an incorrect output in the simulation) included in the three large examples are efficiently located by answering junt a few queries from the algorithmic debugger. The number of interactions between the designer and the debugger is reduced by a factor of ten to a hundred compared to conventional simulation based validation methodologies. The correct SA specification can be automatically translated into a Register Transfer Level (RTL) specification suitable for logic synthesis. In this environment, a designer is freed from the tedious task of debugging a RTL specification, and can concentrate on the design itself. This environment promises to be an important step towards efficient high-level VLSI design specification validation.

Existing algorithmic debugging methods which can locate faults under the guidance of a system have a number of shortcomings. For example, some cannot be applied to imperative languages with side effects; some can locate a faulty function but cannot locate a faulty statement; and some cannot detect faults related to missing statements. This paper presents an algorithmic critical slice-based fault-locating method for imperative languages. Program faults are first classified into two categories: wrong-value faults and missing-assignment faults. The critical slice with respect to a variable-value error is a set of statements such that (1) a wrong-value fault contained in any instruction in the critical slice may have caused that variable-value error, and (2) a wrong-value fault contained in any instruction outside the critical slice could never have caused that variable-value error. The paper also classifies errors found during program testing into three categories: wrong-output errors, missing-output errors, and infinite-loop errors with no output. It finally shows that it is possible to algorithmically locate any fault, including missing statements, for each type of error.