The importance of reporting both composite and maze yield for process split yield learning

Electrical composite yield for a given macro is calculated from many smaller macros that are called mazes. Maze yield can also be calculated. For defects that have a random distribution the composite and the maze yield always provide the same trend for process splits. Specifically the composite yield can be expressed by p
^k
where p is the maze yield and k is the number of mazes. This is no longer true when systematic and random defects co-exist in the dataset. The reading of the composite yield alone can not provide sufficient information of the process impact. Here we provide five real-world case-studies where a complete picture of the electrical fail mechanism is obtained only by analyzing both the maze and the composite yield.