Image-Based Overlay
Image-based overlay (IBO) measures layer-to-layer registration the way the name suggests — directly imaging a mark, usually a box-in-box or bar-in-bar target, through a high-magnification microscope and extracting the actual offset between the current layer's feature and the reference layer's feature from that image. It's the oldest, most established overlay technique in the industry, backed by decades of accumulated calibration and characterization data, and it remains the reference standard newer techniques like scatterometry overlay still get validated against.
A box-in-box or bar-in-bar target makes misalignment something you can literally see, which is exactly why IBO remains so intuitively trustworthy even as faster alternatives have emerged. An outer box or frame printed from the reference layer, and an inner box printed from the current layer, are deliberately designed to sit perfectly centered within each other when overlay is zero — any real registration error between the two layers shows up directly as the inner feature sitting off-center relative to the outer one. Measuring that offset directly, as a simple geometric displacement between two imaged features, is conceptually about as straightforward as overlay measurement gets.
Image processing algorithms determine each target element's precise center, and the overlay value is simply the difference between those two measured centers — but getting that center determination accurate enough to matter requires serious computational care. Locating a feature's true center from a captured microscope image, to the sub-nanometer precision advanced-node overlay budgets demand, means correcting for everything from illumination non-uniformity to subtle asymmetries in how the target itself was actually patterned — a nontrivial image-processing problem even though the underlying geometric concept (measure two centers, subtract) sounds simple.
Multi-wavelength measurement lets IBO adapt to whatever specific layer stack it's measuring, because a single fixed wavelength doesn't deliver equally clean signal across every possible combination of underlying films and materials. Different layer stacks interact differently with light at different wavelengths — some combinations produce excellent contrast and signal quality at one wavelength while barely registering at another. Measuring at multiple wavelengths and selecting whichever one delivers the cleanest, most reliable signal for that specific layer combination is standard practice, letting IBO maintain measurement quality across the wide variety of layer stacks a real fab actually needs to measure overlay on.
Tool-induced shift (TIS) is IBO's primary systematic error source, and correcting for it is treated as a mandatory, built-in part of the measurement process rather than an optional refinement. The measurement optics themselves — lens aberrations, illumination asymmetries — can introduce a small, systematic bias into the measured offset, indistinguishable from genuine overlay error unless specifically accounted for. Standard practice measures the same target twice, once normally and once with the wafer rotated 180 degrees: genuine overlay error reverses sign under that rotation, while tool-induced bias doesn't, letting the two be mathematically separated and the tool bias subtracted out.
Traditional IBO targets' real-estate cost — typically 20 to 30 microns across — is the single biggest practical limitation that opened the door for scatterometry-based alternatives, even though IBO's underlying measurement remains the trusted calibration reference. That target size consumes genuinely meaningful scribe-line area, limiting how many overlay measurement sites a wafer can practically accommodate without sacrificing space that could otherwise be used for other scribe-line structures. This is exactly the limitation that drove development of scatterometry overlay's dramatically smaller micro-DBO targets — but IBO's decades of accumulated calibration data and well-understood error characteristics are precisely why it remains the reference technique every newer overlay measurement approach still has to validate itself against before earning production trust.