In order to improve the performance of the comparison, we will take into account the specific properties of the algorithm used :
- It is possible to divide the work in sub-tasks as we saw that all the cells on a diagonal can be evaluated simultaneously.
- The evaluation of a cell is the same for all of them.
- To evaluate the value of a cell, we only require the value of three neighbour cells.
- The evaluation is constrained by the computation and not by the I/O throughput.
As the result of these observations, a systolic architecture is the most adapted.