Top-Down SEM
Top-down SEM uses a scanning electron microscope to examine a patterned surface in plan view, approximately along the surface normal. It is particularly useful for lateral dimensions: line and space widths, hole shapes, pitch, and local pattern defects. In a critical-dimension SEM (CD-SEM), a controlled imaging and analysis recipe turns the signal into dimensional measurements.
The image is not a direct height map. Brightness depends on electron emission and collection, material, surface geometry, and imaging conditions. A top-down image may contain information influenced by three-dimensional shape, but it does not uniquely reveal every sidewall or buried interface.
Start with the measurement direction. For a line pattern, linewidth is measured across the line; pitch is the repeat distance between corresponding positions on adjacent lines. These are not the same quantity. A change in linewidth can occur while pitch remains unchanged, and irregular edges make the measured width depend on where along the line it is evaluated.
The measurement recipe must specify the target, scan direction, analysis region, and definition of the reported CD. A line CD, a space CD, and a contact-hole diameter require different geometry. For a noncircular hole, a single diameter can hide shape information; the definition should state whether it represents a particular direction, an average, or a fitted contour.
An image edge is an inference from a signal. An SEM raster scans a focused electron beam and records detector response versus position. Feature edges often produce strong contrast, but the brightest pixel is not automatically the geometric boundary. The apparent edge depends on the interaction between the sample and beam, detector collection, and the edge-location algorithm.
Threshold, derivative, fitted-profile, and model-based methods can assign different positions to the same image. That does not make every method equally valid for every target. The method needs calibration and validation for the intended measurand, including how changes in focus or feature shape affect bias. Comparing tools requires compatible measurement definitions, not just matching the displayed magnification.
Repeatability, resolution, and accuracy are different. Repeatability describes variation when measurements are repeated under specified conditions. Spatial resolution describes the ability to distinguish image detail. Accuracy also involves bias relative to the intended reference value. A tool can repeat a biased linewidth very consistently, and subpixel edge fitting does not by itself establish subpixel accuracy.
One visible width does not establish a full profile. A line's top width, bottom width, height, and sidewall form are separate properties. A simple plan-view CD cannot uniquely determine them all. Additional viewing angles, validated signal models, cross sections, or complementary metrology may be needed, depending on the structure and uncertainty required.
This limitation should not be overstated as “top-down SEM contains no depth information.” Shape affects the signal, and specialized methods can extract additional parameters under assumptions. The important question is whether the available data and model constrain the desired quantity sufficiently, not whether an image looks three-dimensional.
Roughness needs more than one cross-line measurement. Line-edge roughness describes edge-position variation along a line. Line-width roughness describes variation of the separation between its two edges. These differ because the two edges can move together or independently. Measurement length, sampling interval, filtering, and noise treatment affect the reported statistic and must accompany meaningful comparisons.
Image noise can appear as edge roughness. Focus, scan stability, drift, and edge detection affect the result, so a noisier image is not evidence that the physical line became rougher. Averaging or filtering can reduce apparent noise but can also remove real spatial variation. The analysis should state the spatial range and corrections used.
The beam can influence the specimen. Charging, contamination, and irradiation-induced changes can affect contrast or dimensions. Resist and other sensitive materials may change during repeated imaging. Dose, scan strategy, landing conditions, and the analysis recipe must therefore be qualified for the target; “non-sectioning” does not mean perfectly non-perturbing.
In production, CD measurements and defect review typically examine selected locations. A useful sampling plan connects those sites to the process question: local pattern fidelity, across-wafer variation, or review of previously detected defects. A clean image at one site does not establish that the whole wafer is defect-free.
The key idea: top-down SEM is powerful lateral metrology when the viewing geometry, signal interpretation, measurement definition, and uncertainty are controlled. Treat the reported CD as a qualified measurement, not a ruler laid across an unquestionable picture.
Sources: Hitachi High-Tech CD-SEM documentation; NIST studies of linewidth bias, focus variation, and edge-roughness measurement. Diagrams are explanatory schematics, not measured SEM images or instrument specifications.