We present a hierarchical technique HCLIP to generate near-optimum layouts of CMOS cells in the two-dimensional (2-D) style. HCLIP is based on integer-linear programming and extends our previously published CLIP technique to much larger cells and to 2-D cell-arrays. HCLIP partitions the circuit into clusters, generates minimum-width 1-D placements (chain covers) for each cluster, and then selects one cover for each cluster such that the overall 2-D cell width and height is minimized. In doing so, HCLIP explores all diffusion sharing between transistor chains belonging to the selected covers. For width minimization, HCLIP yields 2-D layouts that have minimum width with respect to the given set of covers. For both width and height minimization, since HCLIP is approximate and can overestimate cell height, we analyze the theoretical worst-case approximation. Experimental results demonstrate that HCLIP still yields near-optimal layouts in most cases.