Overlay Accuracy in Maskless Lithography: How Encoder Feedback Bounds Layer-to-Layer Registration

When the encoder becomes the mask

A projection lithography tool transfers registration from a physical mask; a maskless tool transfers it from the motion system. The layout file contains ideal coordinates, but the position those coordinates actually land at is the position reported by the stage encoders at the moment of exposure. In that sense the encoder is not merely a servo component but the pattern coordinate system itself. Any error it reports, the writer believes, and the resist records. This is why overlay conversations that treat the encoder as a solved commodity so often end with unexplained placement scatter.

The consequence for procurement is structural: overlay claims must decompose, and the encoder line in the decomposition must carry a number. A single figure for overlay accuracy hides whether the budget is spent on alignment sensing, stage geometry, thermal behaviour or feedback error. Suppliers who allocate the budget are inviting verification; suppliers who do not are asking to be trusted. Since multi-layer processes multiply whichever weakness exists, the allocation matters more in direct write than in any mask-based flow.

Decomposing the overlay budget into encoder terms

A workable budget separates three families of terms. Placement error describes how accurately a feature lands relative to the machine coordinate system within a single field, dominated by scale accuracy, interpolation error and the writer optics. Field matching describes how consistently consecutive fields land relative to each other, dominated by stitching behaviour, reference repeatability and stage dynamics. Layer-to-layer overlay describes how the whole written pattern registers against the previous layer, adding the alignment sensor, substrate handling and process contributions on top.

Encoder feedback appears in all three, but with different faces. In placement it is scale error and interpolation noise. In field matching it is datum repeatability and velocity-dependent feedback lag during settle. In layer-to-layer overlay it is thermal drift and reference strategy, because the second exposure must rediscover the coordinate system the first one used. Asking for the encoder contribution to each family separately is the single most revealing question a buyer can pose, and the hesitation or fluency of the answer is itself evidence.

How each encoder error becomes a visible defect

The mapping from feedback physics to resist image is worth knowing because it makes diagnosis faster. Scale line error repeats at fixed stage coordinates, so the same distortion signature appears in every layer written at the same field, a fingerprint error mapping can compensate. Interpolation noise is random and speed dependent, so it survives into line-edge roughness and vanishes in slow tests. Abbé error multiplies small angular and straightness errors by the distance between the encoder axis and the focal plane, which is why placement degrades toward substrate edges and why tall chucks change results.

Thermal drift is the slowest and most misleading term: a scale warming under motor and encoder electronics heat expands, so features written late in a run sit at slightly stretched coordinates relative to early features. Between layers, the drift becomes an overlay scaling mismatch that process engineers chase through dose and bake. Each of these has a countermeasure, compensation tables, dynamic verification, Abbé arm minimisation, scale material selection and warm-up discipline, but none can be applied to a term that was never measured.

• Repeating field distortion: suspect scale error, request mapping compensation
• Speed-dependent edge roughness: measure interpolation error at writing velocity
• Edge-of-substrate growth: shorten Abbé arms and verify stage yaw
• First-to-last-die shift: control scale temperature gradients and warm-up

On-the-fly writing and the synchronization burden

Stop-and-expose writing hides the hardest feedback problem; writing on-the-fly exposes it. When the writer fires while the stage moves, the exposure coordinates are a fusion of encoder position sampled at high rate, writing-clock timing and galvanometer or pixel clock deflection. Encoder data latency, interpolation filter delay and the synchronization scheme between encoder clock and writing clock all become placement errors with the sign of a phase lag: features lead or trail their nominal position by the stage travel during the delay.

Qualifying this regime needs dynamic evidence, not datasheets. Ask for interpolated position error measured in the installed configuration at the actual writing velocity, the latency figure of the encoder data path, and how the writing clock is slaved to position rather than to an independent timebase. A tool that writes only on-the-fly should demonstrate stitching and placement results at speed; a tool that claims both modes should demonstrate both, because the fast mode is usually the one that stresses the encoder chain hardest.

Measuring overlay so the budget can be closed

Overlay claims are only as good as the metrology behind them, so the acceptance plan deserves as much attention as the specification. A meaningful plan measures registration across the full substrate area, not only the centre field; measures field-to-field stitching along both axes; and, for multi-layer work, measures actual layer-to-layer overlay using the process engineer's own mark scheme after development. The measurement tool and its uncertainty should be named, because a budget of a hundred nanometres cannot be verified by a microscope reticle.

Structure the results against the allocated budget rather than against hope. Placement residuals that repeat every layer point at machine geometry or scale error; residuals that step between layers point at reference or thermal terms; residuals that correlate with substrate radius point at Abbé or orthogonality. With the allocation in hand, a failing result becomes a diagnosis rather than a dispute, and the corrective action, whether compensation, environmental control or encoder replacement, follows from evidence instead of argument.

Specification clauses that protect the overlay budget

Five clauses do most of the work. First, the allocated overlay budget with the encoder contribution stated for placement, field matching and layer-to-layer overlay. Second, dynamic encoder evidence: interpolation error and jitter at writing velocity in the installed configuration. Third, geometry declarations: Abbé offsets per axis, scale material and expansion behaviour, and axis orthogonality. Fourth, reference strategy with measured repeatability across power cycles and substrate exchanges. Fifth, the acceptance overlay metrology plan, defined before order and executed at factory and site acceptance.

The clauses also shape daily operation. Warm-up routines, substrate and chuck temperature stabilisation, and scheduled verification of encoder amplitude and error become budget-maintenance tasks rather than optional hygiene. Tools that treat these as discretionary drift into a state where the original overlay figure is unreproducible, and the process engineer inherits a moving coordinate system. The specification, written once, is what keeps the coordinate system still for the life of the tool.