A constraint solver utilizes a modified relaxation process to generate multiple different stimulus stream arrays that comply with multi-dimensional (e.g., 2D or 3D) constraints. First, an array is generated including rows and columns of randomly generated test vector values. During a first revision phase, the array is modified to comply with first-dimension constraints (e.g., selected test vector values are changed in non-compliant rows until every row complies with all row constraints). A second revision phase is then performed in multiple cycles, where each cycle includes identifying a current element having a greatest impact on non-compliance of the array on second-dimension (e.g., column and/or diagonal) constraints, and revising the current element's test vector value in a way that both minimizes the non-compliance, and also maintains compliance of the array with the first-dimension constraints. The second revision phase repeats until the array converges on a solution that complies with all multi-dimensional constraints.