This research describes an approach for path generation using an observability metric for delay test. K Longest Path Per Gate (KLPG) tests are generated for sequential circuits. A transition launched from a scan flip-flop (SFF) is captured into another SFF during at-speed clock cycles, that is, clock cycles at the rated design speed. The generated path is a 'longest path' suitable for delay test. The path generation algorithm then utilizes observability of the fan-out gates in the consecutive, lower-speed clock cycles, known as coda cycles, to generate paths ending at a SFF, to capture the transition from the at-speed cycles. For a given clocking scheme defined by the number of coda cycles, if the final flip-flop is not scan-enabled, the path generation algorithm attempts to generate a different path that ends at a SFF, located in a different branch of the circuit fan-out, indicated by lower observability. The paths generated over multiple cycles are sequentially justified using Boolean satisfiability. The observability metric optimizes the path generation in the coda cycles by always attempting to grow the path through the branch with the best observability and never generating a path that ends at a non-scan flip-flop. The algorithm has been developed in C++. The experiments have been performed on an Intel Core i7 machine with 64GB RAM. Various ISCAS benchmark circuits have been used with various KLPG configurations for code evaluation. Multiple configurations have been used for the experiments. The combinations of the values of K [1, 2, 3, 4, 5] and number of coda cycles [1, 2, 3] have been used to characterize the implementation. A sublinear rise is run time has been observed with increasing K values. The total number of tested paths rise with K and falls with number of coda cycles, due to the increasing number of constraints on the path, particularly due to the fixed inputs.
This research describes an approach for path generation using an observability metric for delay test. K Longest Path Per Gate (KLPG) tests are generated for sequential circuits. A transition launched from a scan flip-flop (SFF) is captured into another SFF during at-speed clock cycles, that is, clock cycles at the rated design speed. The generated path is a 'longest path' suitable for delay test. The path generation algorithm then utilizes observability of the fan-out gates in the consecutive, lower-speed clock cycles, known as coda cycles, to generate paths ending at a SFF, to capture the transition from the at-speed cycles. For a given clocking scheme defined by the number of coda cycles, if the final flip-flop is not scan-enabled, the path generation algorithm attempts to generate a different path that ends at a SFF, located in a different branch of the circuit fan-out, indicated by lower observability. The paths generated over multiple cycles are sequentially justified using Boolean satisfiability. The observability metric optimizes the path generation in the coda cycles by always attempting to grow the path through the branch with the best observability and never generating a path that ends at a non-scan flip-flop.
The algorithm has been developed in C++. The experiments have been performed on an Intel Core i7 machine with 64GB RAM. Various ISCAS benchmark circuits have been used with various KLPG configurations for code evaluation. Multiple configurations have been used for the experiments. The combinations of the values of K [1, 2, 3, 4, 5] and number of coda cycles [1, 2, 3] have been used to characterize the implementation. A sublinear rise is run time has been observed with increasing K values. The total number of tested paths rise with K and falls with number of coda cycles, due to the increasing number of constraints on the path, particularly due to the fixed inputs.