Pattern Transfer, Etch, and Yield

Abstract

An EUV resist image is temporary. Device value appears only when the edge is transferred through an underlayer or hard mask into silicon, dielectric, metal, or another functional material. High NA pushes resist films thinner, reducing collapse and focus burden while leaving less material to survive plasma exposure. Pattern transfer must preserve line width, placement, profile, roughness, and rare-feature survival while controlling selectivity, loading, charging, microtrenching, footing, residue, mask erosion, and chamber drift. I organize the process as sequential state conversion with metrology after resist, hard mask, target etch, and cleanup. Etch can smooth selected roughness frequencies and amplify others. Mean critical dimension can remain on target while local necking or profile error destroys electrical yield. I therefore close the loop with electrical test structures and defect inspection, not resist images alone. The best lithography process is the one that creates the most functional device edges per unit exposure and process cost.

Resist is a sacrificial information carrier

The resist contour encodes the intended boundary but cannot usually serve as the final device material. Plasma etch uses reactive radicals, ions, photons, and surface chemistry to remove unprotected material and preserve protected regions.

For target depth $H$ and resist thickness $t_r$, minimum selectivity screen is

$$S_{min}=\frac{H}{t_r-t_{margin}}.$$

If a 20-nanometer resist must transfer a 60-nanometer film while retaining 5 nanometers of margin, required effective selectivity exceeds 4 before local nonuniformity and profile effects. High NA may push resist thinner, making direct transfer harder.

An underlayer or hard mask can convert the resist image into a more etch-resistant material. That adds deposition, lithographic interaction, etch, strip, defect, and alignment considerations. Co-optimization matters more than one material’s nominal selectivity.

Plasma etch is spatially selective surface chemistry

Reactive species adsorb and react at surfaces. Directional ions provide anisotropy, activate reactions, remove passivation, and can damage material. Sidewall passivation suppresses lateral etch. Pressure, gas composition, source power, bias power, temperature, flow, chamber wall, loading, and feature aspect ratio shape the result.

Etch mechanism	Desired contribution	Failure mode
Chemical reaction	Material selectivity and removal	Isotropic loss or residue
Ion bombardment	Directionality and activation	Damage, charging, mask erosion
Passivation	Sidewall protection	Excess footing, taper, or blocked feature
Volatile product formation	Clean removal	Condensation or redeposition
Thermal control	Stable reaction and profile	Wafer or local CD nonuniformity

Pattern density changes local reactant and product transport. Dense and isolated features can etch at different rates. Aspect-ratio-dependent etch means deeper or narrower features evolve differently. Chamber history changes surface recombination and species balance.

Edge motion has a spectrum

Let resist edge be $x_r(y)$ and final edge be $x_f(y)$. A simple transfer representation is

$$x_f(y)=x_r(y)+b(y)+n_e(y),$$

where $b$ is systematic etch bias and $n_e$ is added or transformed variation. In frequency domain, etch can act as transfer function $H_e(f)$:

$$PSD_f(f)=|H_e(f)|^2PSD_r(f)+PSD_{added}(f).$$

Some high-frequency resist roughness may smooth. Mask erosion, microloading, material grains, and plasma variability can add other scales. An rms number before and after etch cannot reveal which wavelengths changed.

NIST roughness work emphasizes spectral analysis and metrology noise. I would track the same edge where possible, using fiducials and three-dimensional methods to connect top contour, sidewall, and bottom contour. The electrically relevant edge may lie at a particular depth.

ResistMeasure developed contour MaskConvert to transfer layer EtchShape functional material CleanRemove residue and sacrificial films TestInspect contour and electrical function

Profile matters as much as top-down CD

A top-down image can report correct width while sidewalls taper, bow, roughen, notch, or carry residue. Contact and via resistance depend on bottom opening and subsequent fill. Gate performance depends on channel-edge geometry. Interconnect reliability depends on line cross-section and defects.

Profile metrology may use cross-section electron microscopy, scatterometry, atomic-force methods, tomography, or destructive analysis. Production control needs fast proxies calibrated to slower references. Model uncertainty must include material optical constants and assumed shape.

Charging can distort ion trajectories in insulating features. Local electric fields change profile or damage underlying layers. Pulsed plasmas, frequency choice, conductive paths, and recipe design can mitigate it. High aspect ratio and novel materials make these effects more important.

Residue after etch or strip can block deposition, raise resistance, contaminate interfaces, or seed defects. Cleaning must remove residue without moving critical edges or attacking fragile materials.

Stochastic resist failures can be rescued or amplified

A marginal resist bridge may clear during hard-mask etch. A slight neck can break. An undersized contact can close. Etch is therefore not a passive copier.

Process design can intentionally bias resist and etch so the final distribution centers correctly. This only works with a joint stochastic model and large-area evidence. Correcting average bias while worsening failure tails is unacceptable.

ImageResist contour and failures

Captures lithographic mean, spectrum, profile, and rare-event state.

TransferHard-mask and target etch

Applies material-selective bias, filtering, and added variation.

ObserveContour, profile, residue, and defects

Measures the functional geometry with known detection limits.

ValueElectrical performance and yield

Identifies which physical variations actually fail the product.

Chamber state is part of the recipe

Etch tools condition during operation. Wall coatings, deposits, cleans, part age, gas delivery, pumps, matching networks, temperature, and sensors change plasma. A recipe name does not uniquely define the process.

The chamber needs a state model based on wafer count, material exposure, cleaning, optical emission or other plasma diagnostics, pressure, power, match, temperature, and reference wafers. Preventive maintenance should restore a qualified state, not simply replace parts.

Matching multiple chambers increases capacity and resilience. Chamber-specific correction can compensate stable bias, but large correction may hide degradation. The control limit should preserve process window and ensure a wafer can move among approved chambers without surprise.

Yield is multiplicative and contextual

For defect density $D_0$ and critical area $A_c$, a simple Poisson yield model is

$$Y=e^{-D_0A_c}.$$

Real yield includes clustered defects, parametric distributions, redundancy, systematic signatures, and many process steps. Still, the exponential form explains why small defect-density changes matter on large die.

Patterning metrics should connect to device criticality. A line-edge fluctuation can alter transistor current. A via failure can be masked by redundancy or open an essential path. A small bridge can short power. Design-aware inspection and electrical test prioritize the right edges.

Optimize the integrated module

Imec emphasizes resist, underlayer, hard mask, etch, metrology, and mask co-optimization for High NA because no thin resist solves transfer alone. I would qualify complete stacks across source, focus, dose, film, chamber, wafer, and time.

The pattern transfer frontier is not the highest etch selectivity or smoothest isolated line. It is a stable conversion from an aerial image into functional material whose edge distribution, defect probability, electrical consequence, and maintenance cost are all understood. Lithography finishes when the device works.