Interferometry speeds surface-roughness analysis

Surface roughness is an important parameter for gauging semiconductor material quality, but it is rarely measured quantitatively. This is because the technique often used for quantitative analysis - atomic force microscopy (AFM) - is poorly suited to production environments if we discount the expensive fully automated systems that are only practical in large silicon device lines. So instead, assessments are usually carried out with either qualitative techniques such as Nomarski interference microscopy, or inferred from optical scatterometry methods.

However, there is another technique that can routinely provide quantitative roughness measurements and surface images - white-light interferometry (see "White-light interferometry" for a detailed explanation). This method is well established and has been used for many years to obtain contactless measurement of step height or similar parameters (see figure 1). However, recent advances in instrumentation software have improved the height resolution of today s white-light interferometers to a level comparable with that of an AFM.

For example, the Taylor Hobson CCI 3000A interferometer we have employed at IQE can determine the surface position with a vertical resolution of 0.01 nm by using a patented correlation algorithm to locate the coherence peak and phase position of an interference pattern produced by a white light source. This coherence correlation interferometry (CCI) technique is sufficient to measure roughness levels on many semiconductor samples, since these materials have a roughness of typically 0.1-1 nm. Lateral resolution is still significantly inferior to an AFM and is diffraction-limited to 0.35 μm, which defines the maximum lateral magnification. However, for practical production measurements of semiconductor samples this limitation is not significant, while it is an advantage to be able to measure larger areas than those imaged by an AFM.

We assessed our Taylor Hobson interferometer by mounting it on a passive anti-vibration table and using it to image GaAs and InP-based substrates and epiwafers. All these measurements used a "×50" objective lens with a working distance of 3.4 mm. This magnification provides the maximum lateral resolution over an imaged area of 280 μm when using a 1024 × 1024 pixel CCD array. The instrument can also be used with lower magnification lenses to image areas up to 7 mm × 7 mm, but this reduces the lateral resolution.

Plan-view images of both a part-polished and a fully polished GaAs semiconductor substrate are shown in figure 2. They reveal the residual damage left on the part-polished GaAs substrate by the sawing process that creates individual substrates from the bulk single crystal. This part-polished substrate has a root mean square (RMS) roughness of 0.36 nm, and distinctive, pronounced sharp features associated with saw damage. The fully polished production substrate (see figure 2b) has a much lower RMS surface roughness of 0.16 nm. This image is indicative of most high-quality conventional semiconductor surfaces - a blotchy background with a few sharp features.

The method used for acquiring these images is straightforward, and the skills required are similar to those needed for operating a normal optical microscope. Setup time is short and the images in figure 2 took just 90 seconds to acquire. In comparison, an AFM image on a smaller area of a similar sample would have required approximately 15 minutes of measurement time and a similar setup time. It would have also involved contact between the sample and probe, and required a skilled and experienced operator to ensure a trustworthy result.

High-quality epitaxial layers, particularly if they are thin, produce only small changes in surface characteristics compared with the underlying platform when grown on production-quality substrates. For example, the surface roughness of an InP/InGaAs/InP PIN structure with a total thickness of 5 μm is very similar to that of its substrate (see figure 3a). The low-roughness surfaces are produced by maintaining growth conditions within required, carefully chosen limits. However, any significant variation in the growth conditions, or a switch to inferior substrates, can impair the surface quality and increase its roughness (see figure 3b).

White-light interferometry can also reveal the details in the surface of structures incorporating material with a higher strain. For example, the technique can expose the orthogonal features in relaxed lattice-mismatched long-wavelength (2.2 μm) InGaAs PIN structures grown on InP substrates.

Direct quantitative comparisons between the surface-roughness measurements produced by AFM and white-light interferometry is difficult. Each technique has its own advantages and disadvantages, and produces its best results with different-sized lateral areas. For example, the white-light interferometry measurements presented here have been taken over 280 × 280 μm areas - 30 times larger than the standard maximum area for AFM measurements (50 × 50 μm). In addition, each method has a different lateral resolution, data analysis and filtering techniques, making rigorous comparisons very complex.

Despite these drawbacks, we found good agreement between AFM and white-light measurements using the CCI technique for a range of semiconductor samples with a roughness between 0.1 and 1.5 nm (see figure 4). Because CCI measurements, unlike those made by AFM, are completely contactless, repeated measurements can also be made to establish reproducibility. These measurements have been very stable with time, showing a ±0.01 nm deviation on an average roughness of 0.15 nm that befits a production-worthy technique.

This stability, allied to a height resolution that is comparable to that of an AFM, but with a faster, non-contact approach, makes the technique well suited to a cleanroom production environment.

This work is part of a project entitled "Semiconductor and Optical Layer Analysis and Definition using Interference Microscopy (SOLADIM)", which is funded by the UK Department of Trade and Industry s Micro- and Nano-Technologies Applied Research Program. The other partners in the project are Applied Multilayers, Taylor Hobson, the Centre for Precision Technologies at the University of Huddersfield and the National Physical Laboratory.