Wafer bow and warp describe the unconstrained three-dimensional shape of a semiconductor wafer, while wafer-curvature film-stress measurement uses a change in that shape to infer the average stress added by a film. These quantities affect focus and leveling, chucking, robot handling, bonding, CMP contact, thermal uniformity, and package assembly. They are easy to confuse with thickness variation or local surface flatness, so a defensible measurement begins by defining the surface, reference plane, support condition, edge exclusion, orientation, and temperature.

Bow, warp, thickness variation, and flatness are different measurands. The median surface lies halfway between corresponding front and back surfaces, so it represents wafer shape without directly including thickness variation. Under a specified standard, bow is a signed center displacement of that median surface relative to a defined reference plane, whereas warp is a peak-to-valley range of median-surface deviation. Total thickness variation is the maximum minus minimum local thickness. Front-surface flatness and site flatness instead depend on a surface reference and often a constrained or chucked condition. Values from different definitions are not interchangeable.

Support condition can change the shape being measured. A free-wafer result aims to remove chuck force, clamping, and support deformation, but gravity and support reactions remain important for thin or low-stiffness substrates. Three-point support, vertical orientation, edge support, semicontinuous support, and two-sided scanning can yield different apparent shapes unless the method corrects their mechanical influence. SEMI MF1390 specifies automated noncontact measurement of bow and warp on an unconstrained median surface and examines both external surfaces, distinguishing the result from a front-surface height map on a vacuum chuck.

Curvature change, not absolute bow alone, supports film-stress inference. For a uniform thin film on a much thicker isotropic substrate under small-deflection, equibiaxial conditions, the Stoney relation can be written

where $t_s$ and $t_f$ are substrate and film thickness, $E_s$ and $v_s$ are substrate Young’s modulus and Poisson ratio in the isotropic approximation, $M_s$ is substrate biaxial modulus, and $\Delta\kappa=\kappa_{after}-\kappa_{before}$. Sign depends on the curvature and stress convention. Crystalline silicon requires an orientation-appropriate biaxial modulus, and anisotropic or direction-dependent curvature should be measured along documented wafer axes rather than collapsed into one scalar.

A simple sag-to-curvature conversion is valid only for an assumed shape. For a spherical arc with aperture radius $a$ and center sag $b$, curvature is

Real wafers can be cylindrical, saddle-shaped, edge-rolled, or spatially nonuniform, so one bow number need not determine curvature. Polynomial or Zernike-like detrending can summarize shape but may remove physically meaningful modes. Two-dimensional curvature fields or principal curvatures preserve more information for anisotropic films, patterned wafers, bonded stacks, and stress gradients.

Thermal mismatch makes temperature part of the stress definition. A constrained-film approximation illustrates the effect,

where $M_f$ is an appropriate film biaxial modulus and $\alpha_s$, $\alpha_f$ are substrate and film expansion coefficients. The actual response can include plasticity, creep, cure shrinkage, phase change, cracking, delamination, or temperature-dependent moduli. Room-temperature curvature before and after deposition gives residual stress at that state; an in-situ temperature scan separates reversible thermoelastic curvature from irreversible process evolution only when thermal gradients and chuck interaction are controlled.

st=>start: Define bow, warp, TTV, flatness, curvature, or film stress measurand
state=>operation: Specify wafer side, diameter, thickness, notch orientation, edge exclusion, and temperature
support=>operation: Select free-wafer support and gravity correction or documented constrained state
cal=>operation: Calibrate height sensors, stage, reference artifact, drift, and front-back registration
scan=>operation: Acquire both surfaces or validated median-surface map with repeated orientations
quality=>condition: Coverage, support repeatability, edge behavior, and sensor agreement acceptable?
repair=>operation: Correct support, vibration, contamination, alignment, drift, or missing data
shape=>operation: Compute median surface, reference plane, bow, warp, and curvature without hidden filtering
stress=>condition: Is film stress requested and Stoney regime valid?
model=>operation: Use before-after curvature, film thickness, orientation modulus, and sign convention
advanced=>operation: Use plate or laminate model for thick, anisotropic, patterned, or multilayer stacks
unc=>operation: Propagate height, support, gravity, thickness, modulus, fit, temperature, and model uncertainty
out=>end: Report maps, definitions, support state, metrics, stress model, and uncertainty
st->state->support->cal->scan->quality
quality(yes)->shape->stress
quality(no)->repair->support
stress(yes)->model->unc->out
stress(no)->unc
model->advanced
advanced->unc

Spatial maps reveal mechanisms hidden by one global number. Radially symmetric curvature can indicate uniform film stress; cylindrical curvature can reflect anisotropy or scan-direction process history; saddle modes can arise from crystalline anisotropy, patterned stress, or support; edge roll-off can dominate warp while leaving center bow modest. Comparing maps before and after deposition, anneal, backside grind, temporary bonding, debond, or CMP helps localize the process step that adds a mode. Map registration to notch coordinates is essential when connecting shape to tool azimuth or layout.

Thin, bonded, and patterned wafers often exceed the classical plate assumptions. As substrate thickness falls, gravitational sag and geometric nonlinearity increase strongly, and small support forces can dominate the result. Bonded stacks introduce multiple neutral axes, asymmetric moduli, bonding-layer viscoelasticity, voids, and temperature history. Patterned films create locally varying force and bending moment rather than a uniform blanket stress. Modified Stoney, multilayer laminate, finite-element, or full-field inverse models may be required, with independent thickness and material-property constraints.

The uncertainty budget must follow the complete shape-processing chain. Height-sensor linearity, front/back registration, stage runout, vibration, refractive-index correction, backside roughness, wafer temperature, contamination, missing edge data, support repeatability, gravity compensation, reference-plane removal, spatial filtering, curvature fitting, substrate thickness, film thickness, and biaxial modulus all contribute. Because Stoney stress scales with $t_s^2/t_f$, substrate-thickness uncertainty is doubled in relative form and thin-film-thickness uncertainty can dominate. Repeated remounts reveal support sensitivity that repeated scans without remounting cannot.

Process limits should match the downstream constrained state. Free-wafer bow and warp determine whether robots, aligners, deposition tools, and bonders can acquire and flatten a wafer, but lithography sees residual topography after chucking. A wafer with large free shape may flatten acceptably; another with modest global bow may retain local high-spatial-frequency error. Qualification should combine free-shape metrics with relevant chuck or bonding simulation, site flatness, edge geometry, and handling trials rather than relying on one universal warpage threshold.

A trustworthy wafer-shape result states which surface was measured, how the wafer was supported, how the reference plane and edge were treated, and whether film stress came from a valid before–after curvature model. That is the median-surface-support-and-curvature-change lens.