Silicon wafer surface roughness can influence both transistor performance and gate-insulator reliability, but a single average-height value may not tell the full story. For semiconductor process and equipment teams, the spatial scale of surface features matters: roughness with short spatial wavelengths can affect MOSFET electron mobility and several measures of gate-oxide reliability, while larger-scale variation may have different implications.
A study of n-channel MOSFETs and MOS capacitors fabricated on silicon wafers with varied surface profiles examined this relationship. Its results highlight why wafer characterization should consider roughness across multiple spatial scales, rather than relying only on conventional height-based metrics.
Why silicon wafer surface roughness matters
The interface between silicon and its gate dielectric is part of the electrical structure of a MOSFET. Surface topography at that interface can contribute to scattering that affects carrier mobility. It can also influence the local electric field in the gate insulating film, with consequences for breakdown and leakage behavior.
Roughness is commonly described using height metrics such as arithmetic mean roughness or root mean square roughness. These values summarize the magnitude of surface-height variation over a defined measurement area. However, two surfaces with similar height metrics can distribute their features differently across spatial scales. That distinction can matter when connecting wafer condition to device behavior.
Spatial frequency, or its equivalent spatial wavelength, describes how rapidly surface features repeat. A short wavelength corresponds to more closely spaced features; a long wavelength corresponds to broader variation. Power spectral density (PSD) provides a way to examine how roughness amplitude is distributed across these wavelengths.
Measuring roughness across different scales
The study compared wafers prepared with different polishing and cleaning conditions, including wafers flattened by annealing in an ultra-clean argon ambient. Atomic force microscopy (AFM) was used to examine small areas and shorter-wavelength features, while white-light interferometry captured larger areas and longer-wavelength variation. Using both methods was important because neither instrument covered the same range of surface scales.
The resulting PSD analysis revealed differences that a single roughness number could obscure. In particular, the as-polished wafer conditions differed in roughness below a spatial wavelength of about 300 nm. Argon annealing reduced roughness in this short-wavelength range. At longer wavelengths, the A-condition wafers showed greater roughness than the B-condition wafers for wavelengths of 30 μm or more, and the annealing did not change the long-range roughness detected by the interferometer.
This separation between short- and long-wavelength behavior provided a basis for comparing surface measurements with electrical results. The tested MOSFETs had an approximately 5 nm gate oxide, and the study also evaluated MOS capacitors for gate-insulator reliability.
Effects on MOSFET mobility and gate-oxide reliability
Electron mobility
Electron mobility was higher in the argon-anneal-flattened wafer samples than in samples without that treatment over the reported effective-electric-field range of 0.3–0.5 MV. Among the as-polished samples, the A condition also showed higher mobility than the B condition in that range. Because the B condition had greater roughness below approximately 300 nm, the comparison points to short-wavelength roughness as a factor associated with mobility degradation.
The difference was clearer at lower measurement temperatures. As phonon scattering decreases with temperature, the contribution from surface-roughness scattering becomes more apparent. The reported mobility differences occurred in the middle effective-field range rather than only at high field, indicating that the relationship may not follow a simple assumption that roughness effects emerge solely at high fields.
Breakdown, charge to breakdown and leakage
The study assessed gate-oxide reliability using breakdown electric field (Ebd), charge to breakdown (Qbd) and stress-induced leakage current (SILC). Across the tested samples, argon-anneal flattening was associated with improved results for these measures compared with wafers that were not annealed. The authors suggested that short-wavelength surface features may create local sites of electric-field concentration, contributing to weaker oxide reliability.
For Qbd, the flattened wafers showed substantially greater charge tolerance than the unannealed wafers. Among the as-polished conditions, the A condition also performed better than the B condition, despite having greater long-wavelength roughness. This comparison supports the importance of shorter-wavelength roughness in the measured Qbd behavior; the long-wavelength difference alone did not explain the result.
SILC after charge injection was lower in the anneal-flattened samples. Without flattening, the B-condition samples showed more SILC than the A-condition samples across the reported charge injections. The pattern was consistent with short-wavelength roughness contributing to local field concentration and leakage degradation.
Not every difference was fully explained by roughness. The researchers noted that wafer surface crystallinity, potentially influenced by polishing conditions, could also affect Qbd and SILC. The mechanisms behind differences between wafer conditions therefore require further investigation.
Practical implications for wafer process control
For process engineers and wafer suppliers, the findings reinforce three practical considerations:
- Characterize more than average height. Height metrics remain useful, but they do not describe how roughness is distributed across spatial wavelengths.
- Match metrology to the scale of interest. AFM and white-light interferometry provide complementary views of small-area and large-area surface features.
- Connect surface data with electrical results. Mobility, Ebd, Qbd and SILC measurements can help determine which aspects of surface condition matter for a given process and device structure.
The reported results concern specific wafer preparations and device structures, so they should not be treated as universal thresholds for every process. They do, however, show why polishing, cleaning and annealing conditions should be assessed in terms of both surface profile and device outcomes.
Conclusion
Silicon wafer surface roughness is not simply a question of how high or low surface features are. Their spatial distribution can be relevant to MOSFET electron mobility and gate-insulator reliability. In the studied samples, roughness at wavelengths below about 300 nm was associated with differences in mobility, breakdown, charge endurance and SILC, while larger-scale roughness did not explain the same trends. Combining scale-appropriate metrology with electrical characterization gives semiconductor teams a more useful basis for evaluating wafer surface conditions and process choices.