Precision Mechatronics for Overlay and Focus

Abstract

EUV exposure requires a reticle and wafer to scan in synchronized motion while the optical image remains aligned, focused, and dosed across a field. The stages accelerate rapidly, settle little, and repeat positions at nanometer scales despite structure, cables, cooling, vacuum, thermal drift, acoustic energy, floor motion, actuator force, sensor noise, and wafer variation. I organize the mechatronic system around metrology frames, magnetic or other contactless actuation, high-bandwidth feedback, feedforward, disturbance isolation, wafer mapping, reticle mapping, and per-exposure correction. Overlay is not stage repeatability alone. It includes mask writing and distortion, wafer process history, chucking, alignment marks, optics, thermal state, metrology bias, and intrafield imaging. High NA reduces depth of focus and introduces anamorphic fields and stitching, increasing vertical and cross-field demands. The correct output is the distribution of functional edge placement and focus at throughput, not the still-position accuracy of an unloaded stage.

Scan motion is part of image formation

In a step-and-scan system, a slit of the reticle is imaged while reticle and wafer move at velocities set by optical magnification. Position error during exposure becomes image blur or placement error. Velocity error changes dose and spatial mapping.

For reduction magnification $M$, ideal scan velocities satisfy

$$v_r=Mv_w$$

with sign set by optical geometry. High NA anamorphic imaging uses different magnification by axis, complicating field and stage relationships.

The synchronization error can be represented as

$$e_s(t)=x_r(t)/M-x_w(t).$$

Its spatial spectrum matters. Slow error shifts a field. Error near scan frequencies distorts within field. High-frequency vibration can blur or average depending on exposure integration.

The stage is a coupled dynamic plant

The wafer stage carries wafer, chuck, sensors, cables or contactless services, and actuators. The reticle stage carries mask, pellicle, chuck, and thermal state. Reaction forces move through frames and isolation. Vacuum removes aerodynamic damping and changes thermal paths.

A simplified plant is

$$M\ddot x+C\dot x+Kx=Fu+d,$$

where $u$ is commanded actuation and $d$ is disturbance. Real stages are multibody, nonlinear, flexible, and six-degree-of-freedom systems. Their modes can be excited by aggressive acceleration.

Control uses position metrology, feedforward from planned trajectories, feedback, observers, calibration maps, and iterative learning. Contactless magnetic actuation avoids friction and particle generation while introducing current, heat, force coupling, and magnetic considerations.

ReferenceMetrology frame and clocks

Defines stable coordinates and synchronized measurement time.

MoveReticle and wafer stages

Execute six-axis scan trajectories with controlled force and heat.

ObservePosition, focus, alignment, and environment

Measures state, wafer surface, marks, disturbance, and uncertainty.

CorrectFeedback, feedforward, and calibration

Rejects dynamic error and compensates repeatable signatures.

Overlay is the difference between histories

Overlay measures relative placement between layers. Each layer may have been exposed on another tool, chuck, reticle, date, temperature, and process stack. Wafer processing can expand, contract, warp, or distort the substrate.

I decompose overlay into:

Scale	Representative contributors	Correction scope
Lot and wafer	Thermal expansion, process stress, tool matching	Recipe and wafer model
Field	Translation, rotation, magnification, orthogonality	Alignment and stage correction
Intrafield	Optics, reticle distortion, slit, heating	Field and exposure correction
Local device	Pattern, mark asymmetry, process, stochastic edge	Computational and process design

Alignment marks are engineered for measurement and may respond differently from device patterns to deposition, CMP, etch, resist, and asymmetry. A precise mark measurement can be a biased estimate of device edge. Device-based metrology and electrical structures are needed to validate the mapping.

Overlay budgets must keep covariance. Wafer-scale stress can correlate marks. Tool matching can create systematic signatures. Averaging many marks does not remove common bias.

Focus follows a moving surface

The wafer is not perfectly flat. Thickness, bow, local topography, chuck particles, backside contamination, film stress, and temperature create height variation. The scanner maps and levels the surface, then controls vertical position and tilt during exposure.

High NA’s reduced depth of focus makes focus error a larger fraction of process window. The relevant plane may be inside a thin resist stack rather than the measured top surface. Optical focus sensors can depend on material stack and reflectivity.

Focus error budget includes measurement noise and bias, map sampling, stage vertical error, optical best-focus drift, wafer change between measurement and exposure, reticle contribution, and model residual. A per-wafer correction can address repeatable shape. Local unmeasured topography remains.

Measurement can disturb the object

Interferometers, encoders, alignment optics, focus sensors, temperature sensors, and accelerometers observe different parts of the system. Their coordinate transforms and timestamps matter. Laser wavelength varies with environment where not in vacuum. Structures expand. Electronics delay signals. Sensor mounting creates Abbe offsets, where angular error becomes apparent linear displacement away from the measurement line.

For offset $h$ and small angle $\theta$, Abbe error is

$$e_A\approx h\theta.$$

At $h=100\ \mathrm{mm}$ and $\theta=10\ \mathrm{nrad}$, error is one nanometer. Geometry that seemed mechanically small becomes lithographically large.

Metrology frames should isolate references from force and heat where possible. Calibration must track drift and uncertainty. Redundant sensors help detect disagreement, but common references can remain common causes.

MeasureWafer, marks, and stage state ModelFit geometry and process distortion ScanSynchronize reticle and wafer CorrectApply focus, dose, and placement control LearnUpdate from overlay and device edges

Throughput and precision share force and time

Higher acceleration reduces non-exposure time and raises reaction forces, vibration, heat, and settling demand. More alignment and leveling samples improve model observability and consume time. Slower scanning can reduce dynamic error and throughput.

The optimum is not maximum stage speed. It is annual good fields:

$$Q_{good}=Q_{nominal}A(1-P_{overlay}-P_{focus}-P_{other}),$$

where failure terms are not necessarily independent. A throughput increase that narrows focus margin can reduce good output.

Dual-stage architectures measure one wafer while another exposes, hiding some metrology overhead. This increases tool complexity and requires stage matching. Wafer exchange must preserve cleanliness and coordinate trace.

The fab is inside the control loop

Floor vibration, acoustic pressure, cooling-water fluctuation, temperature, humidity in relevant external paths, electromagnetic disturbance, utility pressure, neighboring tools, and building drift can influence precision. Facility specifications should derive from measured transfer functions to image placement, not broad conventional limits.

Predictive feedforward can use floor and facility sensors. Isolation attenuates selected frequencies and can amplify near resonances. The full tool, foundation, and building need a dynamic model validated after installation.

The mechatronic frontier is a scanner that maintains edge placement while moving at production speed, not a stage that holds still in a quiet laboratory. Its achievement lies in measuring enough of the physical world that nanometer errors become identifiable control states instead of unexplained wafer variation.