Planar 3-DOF UVW (XXY) Alignment Stage Forward & Inverse Kinematics Derivation
In semiconductor photolithography, OLED display lamination, MLCC stacking, and high-density SMT screen printing, substrates must be aligned to fiducial targets with sub-micron ($\pm 0.5\ \mu\text{m}$) translational accuracy and arc-second ($\pm 0.001^\circ$) rotational repeatability. Traditional stacked $X$-$Y$-$\theta$ stages suffer from tall vertical profiles, cumulative Abbe errors, and heavy moving cable inertia.
A Planar 3-DOF Parallel UVW Stage (also called an XXY Stage) replaces the stacked rotary motor with three coplanar linear ball-screw or voice-coil actuators ($U$, $V$, and $W$) coupled to the top plate via crossed-roller bearing pivot joints. By driving the three linear axes synchronously, the stage generates pure translation in $X$, pure translation in $Y$, or planar rotation $\theta$ around any arbitrary programmable center of rotation.
1. Geometric Coordinate Frame & Actuator Layout
Consider a standard orthogonal **XXY / UVW configuration** with stage radius $R$ (or half-span $L = R/\sqrt{2}$) measured from the mechanical origin $O(0, 0)$ to the three kinematic coupling pivots $P_U, P_V, P_W$:
| Actuator Axis | Linear Drive Direction | Nominal Pivot Position $(x_{i0}, y_{i0})$ | Angular Orientation $\phi_i$ |
|---|---|---|---|
| $U$ Axis (X1) | Parallel to $+X$ Axis ($0^\circ$) | $(-L,\ +L)$ or $(0,\ +R)$ | Drives top plate horizontally along $X$ at upper pivot. |
| $V$ Axis (X2) | Parallel to $+X$ Axis ($0^\circ$) | $(-L,\ -L)$ or $(0,\ -R)$ | Drives top plate horizontally along $X$ at lower pivot. |
| $W$ Axis (Y1) | Parallel to $+Y$ Axis ($90^\circ$) | $(+L,\ -L)$ or $(+R,\ 0)$ | Drives top plate vertically along $Y$ at lateral pivot. |
2. Closed-Form Inverse Kinematics: $(X, Y, \theta) \longrightarrow (U, V, W)$
Given a desired planar pose command $(X_c, Y_c, \theta_c)$ of the stage center (where $\theta_c$ is expressed in radians), each pivot point $P_i = (x_{i0}, y_{i0})$ attached to the rigid top plate undergoes a 2D Euclidean rigid-body transformation:
$$\begin{bmatrix} x_i' \\ y_i' \end{bmatrix} = \begin{bmatrix} \cos\theta_c & -\sin\theta_c \\ \sin\theta_c & \cos\theta_c \end{bmatrix} \begin{bmatrix} x_{i0} \\ y_{i0} \end{bmatrix} + \begin{bmatrix} X_c \\ Y_c \end{bmatrix}$$
Projecting the displaced pivot coordinates $(x_i' - x_{i0},\ y_i' - y_{i0})$ onto each actuator's constrained linear guide axis yields the exact nonlinear Inverse Kinematics equations:
// Exact Nonlinear Inverse Kinematics for Orthogonal XXY (U, V along X; W along Y)
// Pivot coordinates relative to stage center:
// P_U = (x_U, y_U), P_V = (x_V, y_V), P_W = (x_W, y_W)
const cosT = Math.cos(thetaRad);
const sinT = Math.sin(thetaRad);
const U = Xc + x_U * (cosT - 1.0) - y_U * sinT;
const V = Xc + x_V * (cosT - 1.0) - y_V * sinT;
const W = Yc + x_W * sinT + y_W * (cosT - 1.0);
Small-Angle Linearized Approximation (For $|\theta| < 2^\circ$)
When $\theta_c$ is very small ($\cos\theta_c \approx 1$ and $\sin\theta_c \approx \theta_c$), the Jacobian matrix simplifies to the classic linear differential kinematics used in PLC ladder blocks:
- $U \approx X_c - y_U \cdot \theta_c = X_c - R \cdot \theta_c$
- $V \approx X_c - y_V \cdot \theta_c = X_c + R \cdot \theta_c$
- $W \approx Y_c + x_W \cdot \theta_c = Y_c + R \cdot \theta_c$
Warning: While the linear approximation is convenient, at $\theta_c = 3^\circ$ on a stage with $R = 150\text{ mm}$, ignoring the $(1 - \cos\theta_c)$ foreshortening term introduces a systematic geometric error of $R(1 - \cos 3^\circ) = 150 \times 0.00137 = \mathbf{205.6\ \mu\text{m}}$! Always evaluate the exact trigonometric $\sin\theta$ and $\cos\theta$ equations in production controllers.
3. Closed-Form Forward Kinematics: $(U, V, W) \longrightarrow (X, Y, \theta)$
To reconstruct the actual stage pose $(X, Y, \theta)$ from live linear encoder feedback $(U, V, W)$ when the two $X$-axis actuators $U$ and $V$ are separated by vertical span $\Delta Y = y_U - y_V = 2R$:
- Rotation Angle $\theta$: Since both $U$ and $V$ move along $X$ with identical $x_0$ offsets ($x_U = x_V$), subtracting $V$ from $U$ isolates the sine of the stage rotation angle: $$\sin\theta = \frac{V - U}{y_U - y_V} = \frac{V - U}{2R} \quad \Longrightarrow \quad \theta = \arcsin\!\left(\frac{V - U}{2R}\right)$$
- Horizontal Translation $X$: Once $\theta$ and $\cos\theta = \sqrt{1 - \sin^2\theta}$ are known, substitute back into the symmetric midpoint of $U$ and $V$: $$X = \frac{U + V}{2} - x_U(\cos\theta - 1)$$
- Vertical Translation $Y$: Substitute $\sin\theta$ and $\cos\theta$ into the $W$-axis constraint equation: $$Y = W - x_W\sin\theta - y_W(\cos\theta - 1)$$
4. Rotation Around an Arbitrary Offset Pivot $(P_x, P_y)$
If a vision camera detects that a glass panel needs to rotate by $\Delta\theta$ around a corner fiducial mark located at $(P_x, P_y)$ rather than the stage mechanical center $(0, 0)$, the required stage-center translation $(X_c, Y_c)$ that keeps $(P_x, P_y)$ stationary is: $$X_c = P_x(1 - \cos\Delta\theta) + P_y\sin\Delta\theta, \qquad Y_c = P_y(1 - \cos\Delta\theta) - P_x\sin\Delta\theta$$
Frequently Asked Questions (FAQ)
Q. Why does a UVW stage have a smaller maximum rotation angle (typically +/- 3 to 5 degrees) than a rotary table?
Because rotation is generated by differential linear travel (V - U) across fixed crossed-roller bearings and angular contact pivots, mechanical stroke limits on the linear ball screws constrain theta to +/- 3 to 8 degrees.
Q. What happens if both U and V actuators move by +5.0 mm while W remains at 0.0 mm?
Since V - U = 0, sin(theta) = 0 (zero rotation), so the stage executes a pure +5.0 mm linear translation along the X axis.
Q. Why must the exact cos(theta) - 1 term be included instead of the linear R * theta approximation?
Even at a 2-degree rotation on a 100 mm radius stage, the cosine foreshortening term R*(1 - cos(theta)) equals 61 microns—far exceeding a 1-micron semiconductor alignment tolerance.
Q. How can a UVW stage rotate a wafer around a corner camera mark without translating that mark?
By coupling a compensating translation Xc = Px*(1 - cos(dTheta)) + Py*sin(dTheta) and Yc = Py*(1 - cos(dTheta)) - Px*sin(dTheta) into the inverse kinematics command.