System-on-Chip Test Scheduling and Test Infrastructure Design

Sammanfattning: There are several challenges that have to be considered in order to reduce the cost of System-on-Chip (SoC) testing, such as test application time, chip area overhead due to hardware introduced to enhance the testing, and the price of the test equipment. In this thesis the test application time and the test infrastructure hardware overhead of multiple-core SoCs are considered and two different problems are addressed. First, a technique that makes use of the existing bus structure on the chip for transporting test data is proposed. Additional buffers are inserted at each core to allow test application to the cores and test data transportation over the bus to be performed asynchronously. The non-synchronization of test data transportation and test application makes it possible to perform concurrent testing of cores while test data is transported in a sequence. A test controller is introduced, which is responsible for the invocation of test transportations on the bus. The hardware cost, introduced by the buffers and test controller, is minimized under a designer-specified test time constraint. This problem has been solved optimally by using a Constraint Logic Programming formulation, and a tabu search based heuristic has also been implemented to generate quickly near-optimal solutions. Second, a technique to broadcast tests to several cores is proposed, and the possibility to use overlapping test vectors from different tests in a SoC is explored. The overlapping tests serve as alternatives to the original, dedicated, tests for the individual cores and, if selected, they are broadcasted to the cores so that several cores are tested concurrently. This technique allows the existing bus structure to be reused; however, dedicated test buses can also be introduced in order to reduce the test time. Our objective is to minimize the test application time while a designer-specified hardware constraint is satisfied. Again Constraint Logic Programming has been used to solve the problem optimally. Experiments using benchmark designs have been carried out to demonstrate the usefulness and efficiency of the proposed techniques.