AFM images of deoxidized samples AR-0, HCl-0, and PAW-0. Typical images of samples after thermal desorption of the surface oxide with no subsequent buffer layer growth for wafers which were used (a) as received, (b) after HCl treatment, and (c) after PAW-HCl treatment. The scale for each image is 12 nm.
(Color online) Roughness of deoxidized and 100 nm thick buffer samples. The axis has been expanded for small tile edge lengths. All three deoxidized samples are considerably rougher than any of the grown buffer layer samples. The roughness also increases to its plateau value very quickly in the deoxidized samples, indicating that the characteristic length scale of the roughness is shorter than in the grown buffer samples. The error bars indicate the standard deviation of the roughness values at each tile size.
AFM images of samples after growth of a 100 nm buffer layer. Typical images of samples after thermal desorption of the surface oxide and 100 nm buffer layer growth for wafers which were used (a) as received (AR-100), (b) after HCl treatment (HCl-100), and (c) after PAW-HCl treatment (PAW-100). The scale for each image is 2 nm.
AFM images of samples after growth of a 500 nm buffer layer with varying pre- and postgrowth conditions. Typical images of samples after thermal desorption of the surface oxide and 500 nm buffer layer growth for wafers (a) as received (AR-500), and (b) after PAW-HCl treatment (PAW-500). The wafers shown in (c)–(e) were all used as received. A 15 min anneal at the growth temperature of (anneal 1) produced an extremely smooth surface (c). A longer 30 min anneal at (anneal 2) produced a surface (d) that was patchier and less smooth than the high-temperature anneal. A pulsed start of the growth with no anneal resulted in (e) a mounded surface. Note that the scale for (a)–(c) is 1.4 nm but increases to 2.0 nm for (d) and (e).
(Color online) Roughness comparison of samples with 500 nm growth on as-received and PAW treated samples. To check sample-to-sample variation, two samples of 500 nm thickness were grown for both the as-received and the PAW treated wafers. The rms roughness varies by 15%–20% for both the repeated AR-500 and PAW-500 samples. The samples grown on as-received wafers are as smooth as the samples grown on wafers pretreated with the PAW-HCl treatment before growth, so no sample improvement results from pretreatment.
(Color online) Roughness comparison of different growth conditions for 500 nm growth on as-received samples. The sample AR-500 AN1, which was annealed at the growth temperature of for 15 min after material deposition was complete, has the smoothest surface. Sample AR-500 PS, grown using a pulsed start, is considerably rougher than all other samples due to the large mounds that developed during growth. The measured surface roughness varies significantly depending on whether the top or the edge of a mound is being measured; this natural spread in the values results in the larger error bars for this sample.
(Color online) Bearing analysis of different growth conditions for 500 nm buffer layers. The bearing percentage is the percentage of all pixels in the analyzed images that fall within the given height range. Compared to the standard continuous start sample, the pulsed start sample is rough and mounded; this broader range of heights is seen in the bearing analysis as a wider, shorter peak in the distribution of pixel heights. The bearing analysis for the extremely flat annealed sample shows a sharp peak with nearly all the imaged surface within a 1.0 nm range. The pixel height histogram results were fitted to a Gaussian distribution, which is shown as a solid line for each sample.
Summary of the growth conditions for buffer layer samples. All samples were grown at and with a continuous start unless a pulsed start is noted.
Comparison of rms and average roughness for 500 nm buffer layer samples. The rms roughness and average roughness were calculated only from full scans.
Gaussian widths, amplitudes, and the peak amplitude to width ratio calculated from the bearing analysis for the images of each of the samples. The uncertainty of the Gaussian fit for each sample was 1% of the reported value or less in each case, which is smaller than the sample-to-sample variation. Samples AR-500 and PAW-500 were both repeated, so two values are reported for each. Note that, because of the deep pits present on samples AR-0 and PAW-0, the distributions of pixel heights for those samples were skewed negatively and were not well fitted by the Gaussian distribution. Thus the reported width values for these samples (marked below by asterisks) are not easily comparable to the width values for the other samples.
Article metrics loading...
Full text loading...