Diffraction-Based Overlay
Diffraction-based overlay (DBO) is an optical metrology technique that measures how precisely two patterned layers line up with each other by analyzing the light diffracted off specially-designed overlay targets, rather than by directly imaging the two layers and comparing their positions visually. It achieves sub-nanometer overlay accuracy — a level of precision direct imaging struggles to match — by exploiting a clever asymmetry in how diffracted light behaves when two layers aren't perfectly aligned.
DBO targets are deliberately built from two overlapping gratings with known, intentional offsets, specifically to turn alignment error into a measurable optical signal. Rather than measuring two layers' absolute positions and comparing them, a DBO target pairs two grating structures — one on each layer — pre-programmed with small, deliberate offsets in opposite directions. When light is diffracted off this combined structure, the +1st and -1st diffraction orders (the light bent symmetrically to either side of the straight-through beam) carry information about how the two gratings actually sit relative to each other, encoded in exactly how those two diffraction orders compare to one another.
Perfect overlay produces a symmetric diffraction pattern, and any overlay error breaks that symmetry in a way that's directly measurable. When the two layers are perfectly aligned, the +1st and -1st diffraction orders come out with equal intensity, because the target's built-in geometric symmetry is preserved. Any real overlay error between the layers disturbs that symmetry, making one diffraction order measurably brighter than the other — and the size of that intensity imbalance is directly proportional to the actual overlay error. This relationship is what lets a purely optical intensity measurement stand in for a direct geometric alignment measurement, and it's precise enough to resolve overlay errors well under a nanometer.
The intensity-versus-overlay relationship follows a predictable sinusoidal curve, called a swing curve, which has to be calibrated before raw intensity measurements mean anything quantitatively. Because the underlying physics is a diffraction phenomenon, the measured intensity doesn't simply increase linearly with overlay error — it follows a sinusoidal relationship that depends on the specific target design and the optical wavelength being used. Mapping out this swing curve on known reference samples, with precisely known programmed overlay offsets, establishes the calibration needed to convert a raw measured intensity imbalance into an actual overlay-error number for unknown samples measured afterward.
Micro-DBO targets solve DBO's biggest practical limitation — target size — by shrinking the measurement footprint small enough to fit inside a die rather than needing dedicated scribe-line space. Traditional overlay targets needed to be placed in the scribe line (the narrow, non-functional space between dies on a wafer), consuming area that couldn't otherwise be used and limiting how many overlay measurements could be taken across a wafer. Micro-DBO shrinks the target down to roughly ten microns on a side, with multiple measurement pads packed into that small footprint to capture both X and Y overlay simultaneously — small enough to be placed directly within the die itself, dramatically increasing how many overlay measurements a fab can take without sacrificing usable die area to accommodate them.
Tool-induced shift (TIS) is DBO's main systematic error source, and correcting for it properly is essential to actually achieving the technique's claimed sub-nanometer accuracy. The optical measurement system itself — lens aberrations, illumination asymmetries, slight misalignments in the metrology tool — can introduce its own small, systematic bias into the measured diffraction asymmetry, indistinguishable from real overlay error unless it's specifically accounted for. Standard practice measures the same target both normally and rotated 180 degrees; genuine overlay error reverses sign under this rotation while tool-induced shift does not, letting the two be separated mathematically. Getting TIS correction right is treated as a core, non-optional part of implementing DBO correctly, not an optional refinement — without it, DBO's real-world accuracy falls well short of the sub-nanometer performance the technique is capable of under properly corrected conditions.