Damaged-layer

A damaged layer between 1 to 10 nm thick according to the material and the polishing conditions is generated by the CMP process. This layer would seem to have to be removed as it presents poorly defined physical properties, for example, in terms of contamination, internal stress, insulating characteristics, and the like. Nevertheless the detrimental effects of this layer still have to be clearly demonstrated. 

The damaged layer left by the CMP on the insulator depends to a large extent to the polished material and CMP conditions. It can practically reach... 

The damaged layer can be determined by measuring its chemical resistance by a succession of short dips in diluted etching mixtures. When the etching rate reaches a constant value (corresponding to the bulk material) the damaged layer is removed. Figure 23 shows that both 0.1% HF and hot ammonia lead to the same thickness. 

In the case of the scrubberless approach, the damaged layer is in practice generally removed during the underetch particle removal process (ammonia... 

Fig. 23. Example of damaged layer estimation in PECVD TEOS oxide as measured by ellipsometry using 0.1% HF and [0.25, 6] NH4OH 70°C.

When using a scrubber without any etching step (HF), an additional etching bath can be implemented in order to remove the damaged layer. 

We must ensure that the etching steps performed during the cleaning processes are sufficient to remove the damaged layer as well. 

The simple scrubber process is very efficient to eliminate slurries but does not remove the metallic contamination or the damaged layer. The simplest additional process is to use an HF-based step which removes both of them. The use of an HF-compatible scrubber saves an additional wet bench with a dryer and a wafer transfer. The chemistries used must avoid loading effects. 

On a wet bench, a conventional SCI with megasonics to remove the particles followed by an HF dip to remove both the damaged layer and metallic contamination gives satisfactory results. 

The efficient scrubberless alternative consists in using hot diluted ammonia in a specific bath with very high megasonic power. In this case, the backside surface must be protected with an oxide or nitride layer to prevent a severe silicon roughening effect from occurring. Then an HF-HCl dip enables the metallic contamination and damaged layer to be removed. HCl turns the respective zeta potentials into favorable conditions that limit the particle redeposition. 

The EBSD technique measures the top few monolayers of a sample to determine its crystal structure. However, most polishing techniques leave a damaged layer on the samples. Even 1 or 0.25 pm diamond polish is not sufficient. For EBSD work, a final polish with colloidal silica (grain size 0.05 pm) is often required to remove the damaged surface layers. 

It should be noted that dielectric and optical properties of the near-the-surface layer of a semiconductor, which vary in a certain manner under the action of electric field, depend also on the physicochemical conditions of the experiment and on the prehistory of the semiconductor sample. For example, Gavrilenko et al (1976) and Bondarenko et al. (1975) observed a strong effect of such surface treatment as ion bombardment and mechanical polishing on electroreflection spectra. The damaged layer, which arises in the electrode due to such treatments, has quite different electrooptic characteristics in comparison with the same semiconductor of a perfect crystalline structure (see also Tyagai and Snitko, 1980). 

---