A method is achieved for making improved deep trench capacitors for DRAM circuits with reduced trench faceting at the wafer edge and improved pad silicon nitride (Si3N4) uniformity for increasing process yields. The method utilizes a thicker pad Si3N4 as part of a hard mask used to etch the deep trenches. Then, after forming the deep trench capacitors by a sequence of process steps a shallow trench isolation (STI) is formed. The method utilizes etching shallow trenches in the same thicker pad Si3N4 layer and into the silicon substrate. A second insulating layer is deposited and polished back (CMP) into the pad Si3N4 layer. A key feature is to use a second mask to protect the substrate center while partially etching back the thicker portion of pad Si3N4 layer at the substrate edge inherently resulting from the CMP. This minimizes the nonuniformity of the pad Si3N4 layer to provide a more reliable structure for further processing.