Random Access Memory Testing: Theory and Practice. The Gains of Fault Modeling

摘要

It is shown that functional testing of RAMs can be improved with the aid of fault models. Fault coverage increases when algorithms are used based on a well-defined fault model. A considerable reduction of the test time is possible, since the optimized patterns require substantially fewer memory accesses. The fault coverage of the stuck-at model turns out to be too limited if the memories contain hardly detectable coupling failures, especially caused by faults in the decoder or read/write logic. Ad-hoc procedures like GALPAT, Shifted Diagonals, show a decreased performance if a fault does not suit their specific detection sensitivity. Addressing order does not influence the properties of the patterns for permanent faults. Although the Marching 0's/1's procedure displays excellent performance for nearly good devices, a drop occurs for memories with coupling faults due to the limitations of the fault model. Design errors and processing shortcomings result frequently in time-dependent faults.