Shallow trench isolation (STI) requires a high quality oxide with superior fill capability provided by High Density Plasma (HDP) oxide. Unfortunately, the HDP deposition process can create large within die topographies that are difficult to polish directly using conventional silica slurries. As a result, etch back integration schemes have traditionally been incorporated for STI polish. A more revolutionary approach is the use of Fixed Abrasive (FA) CMP [1]. FA CMP allows direct STI polish with good planarization/process stability, eliminating the need for prior etch back. The planarization efficiency is strongly dependent on the shape of the pad composites that hold the CeO2 mineral. Fixed abrasive pads with pyramid and pole shapes are available. In this work, three different fixed abrasive pads supplied by 3M corporation were evaluated for STI CMP polish performance using the Obsidian 8200C web format CMP tool. Basic polish characteristics such as planarity (dependence on sub-pad/pattern density), selectivity to topography, oxide dishing and nitride erosion are presented. The FA pads discussed here have been classified as “slow”, “medium” or “fast” depending on blanket oxide removal rate.

The slow rate pad had a very high selectivity to topography and very low dishing of the down area oxide. The removal rate of blanket oxide was less than 100 Å/min. The pad was best suited to the polish of isolation trench structures with small, controlled overfill (> 200 Å) across the wafer. A large process window was demonstrated. The removal rate of the “medium” pad also decreased significantly at the onset of planarization with a blanket oxide removal rate of ca. 200 Å/min. Unlike the slow rate pad, the medium rate pad did not provide a suitable overpolish process window required for a manufacturable STI process. It is believed this pad would be a good choice for BPSG polish. In contrast to the slow and medium rate pads, the blanket oxide removal rate of the fast pad was ca. 2000 Å/min with no self-stopping capability at the onset of planarization. The removal rate was extremely center fast, such that it could not be compensated by adjustment of tool parameters. Use of a modified process developed within the DRAM development alliance (DDA) at East Fishkill (IBM/ Infineon) enabled the fast pad to polish deep STI structures that would otherwise be impossible using the slow or medium rate pads.