E-Beam Inspection

E-beam inspection scans selected wafer regions with focused electrons to find physical or electrical anomalies that may be difficult to distinguish optically. It combines electron-image contrast with automated comparison and defect detection. Its value is not simply a sharp picture: the inspection must capture relevant defects at useful coverage, with controlled charging, dose, and nuisance rate.

Each pixel is a measured electron signal. A focused probe moves across the surface while detectors collect emitted electrons. Secondary electrons are particularly sensitive to near-surface structure and electrical potential; backscattered electrons provide other contrast, including material-dependent information. Detector choice, landing energy, current, focus, and dwell time all influence the result. The image is not a literal photograph of atoms, and one nominal resolution number cannot describe every layer or inspection recipe.

Physical inspection looks for unexpected structure or contrast. Registered die-to-die or cell-to-cell comparisons can reveal differences between nominally matching regions. Design-assisted inspection uses layout information to choose patterns of interest or interpret the results. A reported difference is a candidate: registration error, normal process variation, charging, and true pattern defects can all generate differences. A defect repeated identically in the reference can also escape a simple subtraction, so the reference strategy belongs in the coverage assessment.

Voltage contrast turns connectivity into an observable signal. Under electron exposure, a feature connected to an electrical reference can exchange charge through that path. An isolated feature can develop a different potential because its charge balance differs. That potential changes secondary-electron escape and collection, creating contrast even when two top surfaces look geometrically alike. An open contact or via may therefore be inferred from the behavior of an accessible feature above it.

There is no universal rule that a bright feature is good and a dark one is open. Contrast polarity and magnitude depend on landing conditions, detection settings, charging history, and the circuit. An intentionally floating structure is not a defect. Compare equivalent structures under a qualified recipe, and confirm the electrical interpretation rather than assigning every intensity change to an open or short.

A buried-defect signal is not unlimited depth vision. Voltage contrast senses the electrical consequences of a buried connection, not necessarily an image of the break itself. Backscattered-electron modes can reveal some subsurface material differences under suitable conditions, but their useful depth and sensitivity depend on the stack and beam settings. These mechanisms should not be conflated. A contrast anomaly may justify a targeted electrical test or cross-section rather than establish the root cause by itself.

Sampling sets a large part of the time cost. For a simple raster, pixel count is approximately inspected area divided by pixel pitch squared. Imaging time is roughly pixel count times dwell time, before stage, focus, alignment, and data-handling overheads. Halving pitch in both directions therefore gives four times as many samples over the same area. This is a sampling calculation, not a promise of twice the resolved detail.

Dwell time buys signal at a cost. Longer dwell generally collects more electrons and can improve counting statistics, but increases scan time and exposure. Beam current also matters; neither higher current nor longer dwell is free of image-quality or specimen constraints. Charge accumulation can change contrast during a scan or between repeat scans. Qualify landing energy, current, scan order, and any preconditioning together, and verify that inspection itself does not create unacceptable changes in the material.

Multibeam improves parallelism, not the definition of coverage. Multiple probes can acquire information concurrently, but effective throughput still depends on per-beam performance, usable field, data processing, alignment, and the inspection recipe. Beam count is system-specific. Production strategies may combine broad optical screening, targeted electron inspection, design-selected hot spots, and representative sampling. Neither “e-beam cannot be used in production” nor “multibeam sees everything” is a useful general rule.

Inspection and review have different jobs. Inspection searches a defined area for candidate defects. Review revisits selected coordinates, often supplied by optical inspection, to characterize morphology and classify what was found. A sharp review image proves something about that site; it does not demonstrate that the surrounding wafer was inspected. Similarly, a clean hot-spot sample supports only the sampled structures and the qualified defect types.

Report the coverage with the result. Preserve the layer, inspected area, selected patterns, pixel pitch, dwell, landing conditions, reference method, and detection criteria. Check capture on known relevant defects, repeatability, and nuisance classification. Relate voltage-contrast findings to the expected connectivity and supporting electrical evidence. The useful endpoint is a defensible process decision, not a universal resolution claim or a visually impressive micrograph.

Sources: ASML, HMI eScan 600 and HMI eScan 1000; Applied Materials, Analyze and New Imaging Technology Speeds Process Development; JEOL, Secondary Electron Image and Charging Phenomenon; Contact Holes in Vertical Electrode Structures Analyzed by Voltage Contrast-SEM and Conducting AFM; Trampert et al., Fixed Dwell-Time Budget in SEM. Diagrams are explanatory, not production specifications.