Aluminum Etch

Aluminum etch is a coupled surface-chemistry, ion-transport, heat-transfer, and residue-control process in which clearing the metal is only half the job: a successful recipe must also preserve the mask and underlayer, hold the intended profile across pattern density, and leave no chloride inventory capable of turning humid queue time into delayed pitting.

Overview

Why Aluminum Etch Modeling is Complex

Aluminum etching (typically using $\text{Cl}_2/\text{BCl}_3$ plasmas) involves multiple coupled physical and chemical phenomena:

Plasma generation and transport → determines species fluxes to wafer
Ion-surface interactions → physical and chemical mechanisms
Surface reactions → Langmuir-Hinshelwood kinetics
Feature-scale evolution → profile development inside trenches/vias
Redeposition and passivation → sidewall chemistry

Fundamental Reaction

The basic aluminum chlorination reaction:

Complications requiring sophisticated modeling:

Breaking through native $\text{Al}_2\text{O}_3$ layer (15-30 Å)
Maintaining profile anisotropy
Controlling selectivity to mask and underlayers
Managing Cu residues in Al-Cu alloys

Kinetic and Chemical Rate Modeling

General Etch Rate Formulation

A comprehensive etch rate model combines three primary mechanisms:

Parameter Definitions:

Surface Coverage Dynamics

The reactive site balance follows Langmuir-Hinshelwood kinetics:

Term-by-term breakdown:

Term 1: $k_{ads} \cdot \Gamma_{Cl} \cdot (1-\theta)$ — Adsorption rate (proportional to empty sites)
Term 2: $k_{des} \cdot \theta \cdot \exp(-E_d/k_B T)$ — Thermal desorption (Arrhenius)
Term 3: $Y_{react} \cdot \Gamma_{ion} \cdot \theta$ — Ion-induced reaction/removal

Steady-State Solution ($d\theta/dt = 0$):

Temperature Dependence

All rate constants follow Arrhenius behavior:

Typical activation energies for aluminum etching:

Ion-enhanced reactions: $E_a \approx 0.1 - 0.3 \text{ eV}$
Purely thermal processes: $E_a \approx 0.5 - 1.0 \text{ eV}$
Chlorine desorption: $E_d \approx 0.3 - 0.5 \text{ eV}$

Complete Etch Rate Expression

Combining all terms with explicit dependencies:

Ion-Surface Interaction Physics

Ion Energy Distribution Function (IEDF)

For RF-biased electrodes, the IEDF is approximately bimodal:

Key parameters:

$E_{dc} = e \cdot V_{dc}$ — DC self-bias energy
$E_{rf} = e \cdot V_{rf}$ — RF amplitude energy
Peak separation: $\Delta E = 2 E_{rf}$

Collisional effects:

In collisional sheaths, charge-exchange collisions broaden the distribution:

Ion Angular Distribution Function (IADF)

The angular spread is approximately Gaussian:

Angular spread calculation:

Typical values:

Ion temperature: $T_i \approx 0.05 - 0.5 \text{ eV}$
Sheath voltage: $V_{sheath} \approx 50 - 500 \text{ V}$
Angular spread: $\sigma_\theta \approx 2° - 5°$

Physical Sputtering Yield

Yamamura Formula (Angular Dependence)

Parameters for aluminum:

$f \approx 1.5 - 2.0$
$b \approx 0.1 - 0.3$ (depends on ion/target mass ratio)
Maximum yield typically at $\theta \approx 60° - 70°$

Sigmund Theory (Energy Dependence)

Where:

$S_n(E)$ = nuclear stopping power (Thomas-Fermi)
$U_s = 3.4 \text{ eV}$ (surface binding energy for Al)
$Q$ = dimensionless factor ($\approx 1$ for metals)
$\alpha$ = mass-dependent parameter
$M_1, M_2$ = projectile and target masses

Nuclear Stopping Power

With reduced energy:

Ion-Enhanced Etching Yield

The total etch yield combines mechanisms:

Synergistic enhancement factor:

For Al/Cl₂ systems, $\eta$ can exceed 10 under optimal conditions.

Plasma Modeling (Reactor Scale)

Species Continuity Equations

For each species $i$ (electrons, ions, neutrals):

Flux expressions:

Drift-diffusion: $\vec{\Gamma}_i = -D_i \nabla n_i + \mu_i n_i \vec{E}$
Full momentum: $\vec{\Gamma}_i = n_i \vec{v}_i$ with momentum equation

Source/sink terms:

Electron Energy Balance

Heat flux:

Power absorption (ICP):

Collisional losses:

Where $\varepsilon_j$ is the energy loss per collision event $j$.

Plasma Conductivity

Skin depth:

Electromagnetic Field Equations

Maxwell's equations (frequency domain):

Wave equation:

Sheath Physics

Child-Langmuir Law (Collisionless Sheath)

Where:

$J_{ion}$ = ion current density
$V_s$ = sheath voltage
$s$ = sheath thickness
$M$ = ion mass

Bohm Criterion

Ions must enter sheath with velocity:

Ion flux at sheath edge:

Sheath Thickness

Debye length:

Feature-Scale Profile Evolution

Level Set Method

The surface is represented implicitly by $\phi(\vec{r}, t) = 0$:

Normal velocity calculation:

Plus contributions from:

Neutral chemical etching
Redeposition
Surface diffusion

Hamilton-Jacobi Formulation

Hamiltonian for etch:

With $V_n$ dependent on:

Local surface normal: $\hat{n} = -\nabla\phi / |\nabla\phi|$
Local fluxes: $\Gamma(\vec{r})$
Local angles: $\theta = \arccos(\hat{n} \cdot \hat{z})$

Visibility and View Factors

Direct Flux

The flux reaching a point inside a feature depends on solid angle visibility:

Reflected/Reemitted Flux

For neutrals with sticking coefficient $s$:

This leads to coupled integral equations:

Kernel function:

Where $V(\vec{r}, \vec{r'})$ is the visibility function (1 if visible, 0 otherwise).

Aspect Ratio Dependent Etching (ARDE)

Empirical model:

Where:

$AR = \text{depth}/\text{width}$ (aspect ratio)
$AR_c$ = critical aspect ratio (process-dependent)
$n \approx 1 - 2$

Knudsen transport model:

Where:

$z$ = feature depth
$W$ = feature width
$\alpha$ = Clausing factor (depends on geometry and sticking)

Clausing factor for cylinder:

Aluminum-Specific Phenomena

Native Oxide Breakthrough

$\text{Al}_2\text{O}_3$ (15-30 Å native oxide) requires physical sputtering:

Why BCl₃ is critical:

Heavy $\text{BCl}_3^+$ ions provide efficient momentum transfer
BCl₃ scavenges oxygen chemically:

Breakthrough time:

Sidewall Passivation Dynamics

Anisotropic profiles require passivation of sidewalls:

Deposition sources:

$\text{AlCl}_x$ redeposition from etch products
Photoresist erosion products (C, H, O, N)
Intentional additives: $\text{N}_2 \rightarrow \text{AlN}$ formation

Why sidewalls are protected:

At grazing incidence ($\theta \approx 85° - 90°$):

Ion flux geometric factor: $\Gamma_{sidewall} = \Gamma_0 \cdot \cos(90° - \alpha) \approx \Gamma_0 \cdot \sin\alpha$
For $\alpha = 5°$: $\Gamma_{sidewall} \approx 0.09 \cdot \Gamma_0$

Sputtering yield at grazing incidence approaches zero
Net passivation accumulates → blocks lateral etching

Notching and Charging Effects

At dielectric interfaces, differential charging causes ion deflection:

Surface charge evolution:

Where:

$\sigma$ = surface charge density (C/cm²)
$J_{ion}$ = ion current (always positive)
$J_{electron}$ = electron current (depends on local potential)

Local electric field:

Laplace equation in feature:

Modified ion trajectory:

Result: Ions deflect toward charged surfaces → notching at feature bottom.

Mitigation strategies:

Pulsed plasmas (allow electron neutralization)
Low-frequency bias (time for charge equilibration)
Conductive underlayers

Copper Residue Formation (Al-Cu Alloys)

Al-Cu alloys (0.5-4% Cu) leave Cu residues because Cu chlorides are less volatile:

Volatility comparison:

Residue accumulation rate:

Where:

$x_{Cu}$ = Cu atomic fraction in alloy
At low temperature: $ER_{Cu} \ll x_{Cu} \cdot ER_{Al}$

Solutions:

Elevated substrate temperature ($>$150°C)
Increased BCl₃ fraction
Post-etch treatments

Numerical Methods

Level Set Discretization

Upwind Finite Differences

Using Hamilton-Jacobi ENO (Essentially Non-Oscillatory) schemes:

One-sided derivatives:

Godunov flux for $H = V_n |\nabla\phi|$:

Reinitialization

Maintain $|\nabla\phi| = 1$ using:

Iterate in pseudo-time $\tau$ until convergence.

Monte Carlo Feature-Scale Simulation

Algorithm:

INITIALIZE surface mesh
FOR each time step:
a. FOR i = 1 to N_particles:
Sample particle from IEDF, IADF
Launch from plasma boundary
TRACE trajectory until surface hit
APPLY reaction probability:
Etch (remove cell) with probability P_etch
Reflect with probability P_reflect
Deposit with probability P_deposit
b. UPDATE surface mesh
c. CHECK for convergence
OUTPUT final profile

Variance reduction techniques:

Importance sampling: Weight particles toward features of interest
Particle splitting: Increase statistics in critical regions
Russian roulette: Terminate low-weight particles probabilistically

Coupled Multi-Scale Modeling

Coupling strategy:

Plasma Solver Discretization

Finite element for Poisson's equation:

Weak form:

Finite volume for transport:

Process Window and Optimization

Response Surface Modeling

Quadratic response surface:

Key process variables ($x_i$):

Pressure (mTorr)
RF source power (W)
RF bias power (W)
Cl₂ flow (sccm)
BCl₃ flow (sccm)
Temperature (°C)

Matrix formulation:

Least squares solution:

Multi-Objective Optimization

Desirability function approach:

Individual desirabilities:

Optimization problem:

Subject to:

$85° < \text{sidewall angle} < 90°$
$\text{Selectivity}_{Al:resist} > 3:1$
$\text{Selectivity}_{Al:TiN} > 10:1$
$\text{Uniformity} < 3\%$ (1σ)

Virtual Metrology

Prediction model:

Input features:

Recipe: Power, pressure, flows, time
OES: Emission line intensities (e.g., Al 396nm, Cl 837nm)
Chamber: Impedance, temperature, previous wafer history

Machine learning approaches:

Neural networks (for complex nonlinear relationships)
Gaussian processes (with uncertainty quantification)
Partial least squares (for high-dimensional, correlated inputs)

Run-to-Run Control

EWMA (Exponentially Weighted Moving Average) controller:

Where:

$\Lambda$ = diagonal weighting matrix (0 < λ < 1)
$G$ = process gain matrix ($\partial y / \partial x$)

Drift compensation:

Equations:

Modern Developments

Machine Learning Integration

Applications:

Yield prediction: Neural networks trained on MD simulation data
Surrogate models: Replace expensive PDE solvers for real-time optimization
Process control: Reinforcement learning for adaptive recipes

Example: Gaussian Process for Etch Rate:

With squared exponential kernel:

Atomistic-Continuum Bridging

ReaxFF molecular dynamics:

Reactive force fields for Al-Cl-O systems
Calculate fundamental yields and sticking coefficients
Feed into continuum models

DFT calculations:

Adsorption energies: $E_{ads} = E_{surface+adsorbate} - E_{surface} - E_{adsorbate}$
Activation barriers via NEB (Nudged Elastic Band)
Electronic structure effects on reactivity

Digital Twins

Components:

Real-time sensor data ingestion
Physics-based + ML hybrid models
Predictive maintenance algorithms
Virtual process development

Update equation:

Uncertainty Quantification

Bayesian calibration:

Propagation through models:

Monte Carlo uncertainty:

Physical Constants

Process Conditions

The useful mental model begins with a sequence of gates, not a single etch rate. A chlorine-bearing plasma must first penetrate or transform native aluminum oxide, then chlorinate exposed metal, then remove the resulting aluminum chloride before it accumulates or redeposits. Directional ions must keep the feature bottom reactive without destroying mask selectivity or charging-sensitive structures. This sequence explains why a recipe can show a high blanket rate yet stop on patterned wafers, why the first seconds differ from steady state, and why more bias may clear residue while worsening faceting. A compact balance is $R_{Al}=N_s\Gamma_iY_{Al}(E_i,\theta_{Cl},\theta_O)+N_sk_{chem}(T)\theta_{Cl}$, but every term changes after oxide breakthrough. Treat breakthrough time, steady metal rate, and overetch response as separate observables.

Chlorine provides the principal chemical path to removable aluminum chlorides. A stoichiometric bookkeeping reaction is $2Al+3Cl_2\rightarrow2AlCl_3$, although the surface proceeds through adsorbed Cl, partially chlorinated AlCl$_x$, defects, and ion-stimulated events. Kummel and co-workers’ molecular-beam and first-principles work on Cl$_2$/Al(111) showed that mobile chlorine can agglomerate and that chloride formation and desorption are strongly exothermic rather than a quiet equilibrium process. This matters diagnostically: radical delivery, local coverage, and energy-assisted product release are coupled. A higher optical chlorine signal does not prove that more useful chlorine reaches a trench floor, and a rate increase after a bias change does not prove pure sputtering.

Boron trichloride is most valuable when the surface is not yet clean aluminum. Native Al$_2$O$_3$ and oxygen-bearing chamber or mask surfaces consume chlorine chemistry differently from metal. BCl$_3$ is commonly used because boron-containing fragments act as oxygen getters and promote oxide breakthrough, while the mixture still supplies chlorine for metal removal. That does not make BCl$_3$ a universal rate accelerator. Raising its fraction can dilute Cl$_2$, alter ion composition and electron kinetics, and increase boron-oxygen residue. Compare breakthrough delay and post-breakthrough slope separately. If BCl$_3$ shortens the delay but reduces the later slope, it is doing useful oxide work while limiting steady chlorination; an average endpoint time hides that trade.

Aluminum chloride volatility is necessary, but chamber transport decides whether it is sufficient. AlCl$_3$ is far more removable than AlF$_3$, a central reason chlorine chemistry is favored over fluorine chemistry for subtractive aluminum patterning. Yet volatile does not mean instantly absent. Product partial pressure, surface and wall temperature, conductance, residence time, and cold spots determine whether chloride leaves, condenses, or returns. A chamber residence estimate is $\tau_r=V/S_{eff}$, while a surface Damköhler-like ratio compares reaction with evacuation. High conversion and long residence can produce rapid etching and substantial chamber memory together. Diagnose pressure-dependent residue with wall and exhaust temperatures in view, not gas ratio alone.

Ion bombardment creates anisotropy by renewing the bottom surface faster than the sidewall. Positive ions cross the sheath with angular and energy distributions governed by bias waveform, pressure, collisions, plasma potential, and local charging. Their job is not merely to knock out aluminum atoms. They can break bonds, remove oxide and inhibitor, enhance chlorination, and stimulate desorption of AlCl$_x$. The sidewall receives fewer near-normal ions and can remain protected. Compare $R_{Cl+i}$ with $R_{Cl}+R_i$; a positive difference indicates ion-enhanced chemistry. This prevents the common error of labeling all bias dependence as physical sputtering when the sputter yield at the applied energy cannot explain the observed rate.

The mask stack is an active chemical participant rather than a passive ruler. Photoresist, hard mask, antireflection coating, and cap layers change the local carbon, hydrogen, oxygen, and nitrogen inventory. Resist erosion can supply inhibitor while creating faceting and microtrenching; a TiN cap can generate particles or leave a refractory fence if the metal step ignores cap opening. Selectivity has at least three meanings: thickness selectivity, profile selectivity, and defect selectivity. A recipe that preserves nominal resist thickness but rounds the mask edge can transfer a wider aluminum line. Record top critical dimension, bottom critical dimension, sidewall angle, and remaining mask independently.

Alloying elements often become the last material standing. Production aluminum commonly contains Cu and may contain Si, so rapid Al removal can enrich the surface in less volatile components. Cu-rich islands, intermetallics, or oxidized inclusions can become micromasks that seed grass and residue. Marx, Ma, and Chen reported BCl$_3$–Cl$_2$–N$_2$ ECR etching of Al–1%Si–0.5%Cu with rates above $1\,\mu m/min$, across-wafer uniformity near $\pm4\%$, and photoresist selectivity from roughly 2 to 3.8 under their conditions. These figures demonstrate a capable regime, not a portable recipe. The transferable lesson is that alloy composition, additive chemistry, source power, bias, pressure, and geometry form a coupled system.

Pressure changes chemistry, directionality, and residence time at once. Lower pressure usually lengthens ion mean free path and narrows angular spread, but may reduce radical density or alter dissociation. Higher pressure can raise chemical utilization while broadening ion angles and increasing wall-mediated recycling. The scaling $\lambda\propto T/(p\sigma)$ captures only one part. Effective pumping speed and plasma impedance can also move, so a pressure sweep is not a clean single-factor test. Log matching settings, self-bias, source current, throttle position, and endpoint behavior at every point. Interpret profiles through the delivered ion and neutral distributions rather than the pressure setpoint alone.

Source power and bias power should be separated experimentally whenever the reactor permits it. In an inductively coupled plasma, source power primarily changes electron heating, dissociation, and ion flux, whereas substrate bias primarily changes sheath voltage and ion energy. The separation is imperfect because density changes sheath impedance and bias affects plasma balance. Still, a two-dimensional source-by-bias matrix is much more informative than increasing generic power. A flux-limited regime responds strongly to source power; an activation-limited regime responds strongly to bias; a transport-limited regime may barely respond until pressure, temperature, or conductance changes. Include center and edge blanket coupons with dense and isolated structures.

Wafer temperature controls more than a tabulated vapor pressure. Temperature changes adsorption residence, chloride desorption, inhibitor stability, resist behavior, backside heat transfer, and condensation nearby. The surface temperature may differ from chuck setpoint because plasma heating, helium pressure, wafer bow, contact, and pattern-dependent heat generation intervene. A term $k_d=\nu\exp(-E_d/k_BT_s)$ can be extremely sensitive to $T_s$, so a few degrees of drift may masquerade as seasoning or flow sensitivity. Verify backside helium integrity and calibrated wafer temperature before assigning a rate drift to chlorine chemistry. Compare temperature maps with residue and clear-time maps.

Pattern loading is a reactant-accounting problem before it is a geometry problem. A dense aluminum field consumes chlorine and emits AlCl$_3$ over more local area than an isolated line. If replenishment or evacuation is finite, dense regions clear slowly and may retain more residue. The local neutral balance resembles $\nabla\cdot(D\nabla C)-\vec{u}\cdot\nabla C-k_sa_sC=0$, where exposed area density $a_s$ changes with layout and time. Compare open-field fraction, local perimeter, feature depth, and distance from large metal blocks. Density split structures placed at several radii separate chamber-scale depletion from microloading.

Aspect-ratio-dependent etching combines neutral shadowing, ion angular loss, and charging. As a feature deepens, fewer neutrals reach the bottom without wall collision, and off-axis ions strike sidewalls or masks. Isolated conductors can develop potentials that deflect ions. A blanket rate cannot predict trench completion. Normalize clear time by actual metal thickness, then plot residual against aspect ratio and opening width. If depth matters at fixed width, transport is implicated; if width matters before depth develops, charging or mask-top scattering deserves attention. Profile simulators help only after their angular distributions and wall probabilities are constrained by measurement.

Microtrenching is a trajectory signature rather than merely excess overetch. Enhanced removal near a sidewall foot can result from ion reflection from sloped mask surfaces, electric-field focusing, or reduced inhibitor at the corner. More overetch reveals the symptom but may not create the cause. Compare both corners, feature orientation, mask slope, and wafer position. A symmetric foot trench suggests angular or reflection physics; asymmetry can point to tilted incidence, placement, or mask asymmetry. Reducing bias may help, but changing pressure, mask shape, or pulsed bias can address the trajectory cause with less penalty to center clearing.

Undercut means lateral chemical attack outran sidewall protection. High chlorine activity, elevated temperature, weak inhibitor, long neutral exposure after bottom clear, or mask loss can widen the profile. Timing distinguishes the cause. Undercut present early in interrupted cross sections indicates inadequate protection during main etch; undercut appearing during overetch indicates excessive chemical exposure after clear. Measure sidewall position at several normalized depths and times. Additives such as N$_2$ or carbon-bearing species may strengthen inhibition, but they can also raise residue and reduce open-area rate.

Tapered or stopped profiles often mean the bottom is insufficiently activated. Causes include a broad or low-energy ion distribution, charging, excessive inhibitor, oxide inclusions, low temperature, or product accumulation. Increasing bias is one test, not an automatic fix. If a small bias increase produces a large bottom-rate response with little blanket response, activation is likely controlling. If source power or Cl$_2$ fraction matters more, radical starvation is plausible. If chuck or wall temperature dominates, product removal or film balance deserves priority. A designed perturbation matrix identifies these sensitivities with fewer wafers than sequential tweaking.

Endpoint should identify a physical transition rather than merely satisfy a timer. Optical emission may track consumption or release of chlorine species, interferometry can track thickness or reflectance, and electrical signals may respond as exposed area changes. Every signal has transport delay, background drift, and density dependence. Use the derivative and trace shape, not only an absolute threshold. Correlate the trace feature with physical clear verified by cross section or sheet resistance and with the needed overetch margin. When metal area is small, global emission may be insensitive and a statistically bounded timed component may remain necessary.

Overetch is an insurance policy with a measurable premium. It covers incoming thickness variation, within-wafer nonuniformity, endpoint delay, and loading, but spends selectivity and increases sidewall and underlayer exposure. If $t_c$ is the slowest credible clear and $t_e$ nominal endpoint, the base fractional margin is $(t_c-t_e)/t_e$ plus detection and control uncertainty. Build a distribution from thickness maps, clear maps, and endpoint latency, then test its tail. Excess margin drives undercut, microtrenching, mask loss, substrate damage, and chloride retention even while opens improve.

Post-etch corrosion begins with retained chlorine and becomes visible after exposure. Hygroscopic aluminum chloride residues can react with moisture, creating acidic local chemistry that attacks Al and produces pits, halos, or electrical drift. Damage may be absent immediately and emerge after a humid queue, wet transfer, or package exposure. This delay is why corrosion is often assigned to the wet clean alone. Split by queue time and humidity, including controlled-atmosphere transfer where possible. XPS or ion chromatography can connect residual Cl with damage, while optical and SEM inspections establish morphology.

A post-etch treatment must remove or immobilize chloride without sacrificing the stack. Options include an in-situ conversion or clean, controlled dry handling, prompt solvent and aqueous cleans, and compatible inhibitors. Fluorine-containing treatments can replace or passivate chlorine, but nonvolatile AlF$_3$ and attack on other materials must be considered. Oxygen cleans remove organics while changing oxide state. Specify maximum air break, queue environment, clean sequence, and dry protocol as part of the etch recipe. An etcher-qualified wafer that corrodes in the queue is not an etch success.

Chamber seasoning is a boundary condition on every wafer. Wall films absorb and release chlorine, oxygen, boron, carbon, and aluminum products; their state changes after cleans, idle periods, dummy wafers, and product mixes. Walls influence radical recombination, particles, and condensation. Track wafer number since clean, cumulative exposed aluminum, idle time, and prior chemistry. A first-wafer effect that relaxes with metal wafers suggests equilibration. Drift following wall temperature suggests condensation or desorption. Define seasoning by stable trace shape and rate, not only a fixed dummy count.

Across-wafer nonuniformity should be decomposed into supply, energy, and temperature maps. Center-fast behavior can reflect radical or ion density; edge-fast behavior can reflect sheath geometry, edge-ring condition, pumping, or temperature. A metric $(R_{max}-R_{min})/(2R_{mean})$ is useful for control but insufficient for diagnosis. Compare rate, clear time, residual, angle, mask loss, and residue at the same sites. Rotate wafers or use hardware splits to distinguish wafer-fixed from chamber-fixed signatures. If a defect follows chamber orientation, suspect injection, pumping, coil, or electrode asymmetry before changing global gas ratio.

The edge ring and focus ring shape the plasma-to-wafer transition. Their height, erosion, material, coating, and thermal contact affect local sheath shape and ion incidence. A worn ring can create edge microtrenching or CD shift without much center-rate change. Track ring life by cumulative plasma time and chemistry, and measure height or erosion rather than relying on maintenance interval. After replacement, reach the specified seasoning state before comparison. Ring signatures often correlate weakly with endpoint yet strongly with radial sidewall and mask-edge morphology.

Plasma diagnostics become useful when tied to a specific causal question. Optical emission can show relative changes in excited Cl, BCl, Al, or other emitters, but intensity depends on electron energy as well as density. Mass spectrometry reveals exhaust products and transients but is filtered by conductance and walls. VI probes reveal delivered electrical conditions, not surface ion energy directly. Langmuir or ion-flux measurements can help in development chambers but perturb some plasmas. Select the diagnostic whose transfer function addresses the hypothesis and validate it against wafer observables.

Feature-scale simulation needs calibrated surface probabilities rather than decorative complexity. Monte Carlo profile models require ion energy-angle distributions, neutral flux, sticking, reflection, reaction probability, sputter yield, and passivation kinetics. Reactor models provide boundary fluxes that inherit uncertainty from plasma chemistry and walls. A three-dimensional rarefied-flow study of Cl$_2$, BCl$_3$, and AlCl$_3$ in a commercial etcher matched measured profiles with a simplified reaction and assumed probability near 0.25. That shows transport-reaction coupling can predict, not that 0.25 is universal. Calibrate multiple geometries so compensating parameters cannot fit one profile accidentally.

A reduced model is often more diagnostic than a maximum-detail model. Start with balances that can be constrained: chlorine flux, ion flux and characteristic energy, exposed area, product conductance, and a few surface states. Use reaction-to-transport and ion-to-neutral ratios. Add mechanisms only when residuals show a systematic signature. If blanket rate fits but density loading does not, add neutral depletion or product inhibition. If the average profile fits but corner trenches do not, add angular reflection. If fresh-chamber data fits but wafer sequence does not, add wall state. Complexity should enter in response to falsified predictions.

A practical experiment starts by classifying the failure in space and time. Determine whether the issue is global, radial, azimuthal, layout-local, feature-local, first-wafer, progressive, or delayed after etch. Identify whether it appears during oxide breakthrough, main removal, overetch, strip, wet clean, queue, or reliability stress. This sharply reduces the hypothesis set. A radial sidewall defect after ring aging differs from density-correlated residue; immediate grass differs from pits after humid storage. Preserve representative wafers before cleaning whenever safe because a clean can erase evidence separating formation from revelation.

~~~svg

~~~

~~~svg

~~~

~~~svg

~~~

~~~svg

~~~

~~~svg

~~~

~~~svg

~~~

A screening matrix should perturb mechanisms rather than merely recipe names. Split BCl$_3$/Cl$_2$ ratio to separate oxide conditioning from chlorine supply, source power for reactive and ion flux, bias for activation, pressure for angular transport and residence, and temperature for desorption and inhibition. Add chamber-state and queue blocks where drift or corrosion is suspected. Randomize or bracket runs so history does not alias with a factor. Collect endpoint and hardware variables automatically, then use SEM and surface analysis on conditions that discriminate hypotheses. The goal is a model predicting which defect moves and why, not only a smooth response surface.

Control limits should surround mechanisms that precede wafer failure. Leading indicators include breakthrough duration, main-step endpoint slope, delivered impedance, pressure response, wall temperature, backside helium leak rate, and seasoning state. Lagging indicators include residual thickness, CD, sidewall angle, corrosion count, and electrical yield. Limits require stable definitions and gauge capability. A tight limit on ambiguous optical intensity can create alarms without protection, while a physically correlated derivative may be useful. Link each limit to an action naming the suspected subsystem and verification measurement.

Material compatibility defines the safe edge of the process window. Aluminum may sit above Ti, TiN, W, dielectric, or sensitive junctions and below resist, oxide, nitride, or antireflection coatings. Chlorine plasma, bias, ultraviolet radiation, ash, and wet cleans act on them all. Measure underlayer loss after realistic overetch, not nominal clear alone. Check galvanic couples during wet processing and charging damage on product-like antennas. For MEMS, gaps can trap residue; for bond pads, surface state affects bonding; for power metal, local pitting drives current crowding.

Literature values are anchors for mechanism rather than drop-in setpoints. The Marx–Ma–Chen ECR study establishes that high-rate vertical Al-alloy etching with useful uniformity and resist selectivity is achievable in BCl$_3$–Cl$_2$–N$_2$. ASTM work reports strong Cl$_2$ concentration dependence and additive effects on anisotropy. Directed Cl$_2$ plus ion-beam experiments reported through NASA further isolate the benefit of combining reactive flux with directional energy. Reactor geometry differs from a production ICP, so transfer mechanistic trends and experimental structure, then re-establish the window on the actual stack.

Qualification must include the tails of manufacturing variation. Challenge high and low metal thickness, maximum open area, minimum opening, dense and isolated patterns, center and edge, fresh and seasoned chambers, and credible endpoint delay. Include the longest permitted queue and controlled humidity when corrosion is possible. Report confidence intervals and sample locations, not only averages. A nominal window excluding these tails merely postpones discovery. A mechanism-supported window can justify smaller overetch margins and reduce mask loss without sacrificing clear probability.

The strongest closure test predicts a new condition before it is run. After choosing a cause, predict the sign and approximate magnitude of a response outside calibration: how a denser layout changes clear time, how a fresh chamber changes breakthrough, or how shorter air exposure changes pits. Run that condition with predefined criteria. A model that only explains completed experiments may be overfit; one that predicts a new geometry or chamber state is useful. Failed predictions identify missing wall, transport, charging, or material-state physics.

Aluminum etch succeeds only when surface state and integration state agree. The plasma must break oxide, deliver reactive chlorine, provide directional activation, evacuate chloride products, and protect sidewalls. The module must then remove or stabilize residue before moisture creates corrosion while preserving the mask, underlayer, dimensions, and electrical reliability. Rate, endpoint, profile, residue, and queue response are one evidence set. Read aluminum etch through a coupled reaction-transport-and-integration lens rather than a single-rate recipe lens.