Aluminum Metal Etch Mathematical Modeling

Keywords: aluminum etch,al metal etch,aluminum metal etch modeling,al etch modeling,aluminum chlorine etch,alcl3,metal etch plasma,aluminum plasma etch,bcl3 etch

Aluminum Metal Etch Mathematical Modeling

1. Overview

1.1 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

1.2 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

2. Kinetic and Chemical Rate Modeling

2.1 General Etch Rate Formulation

A comprehensive etch rate model combines three primary mechanisms:

Parameter Definitions:

2.2 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 ($d\theta/dt = 0$):

2.3 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: $E_d \approx 0.3 - 0.5 \text{ eV}$

2.4 Complete Etch Rate Expression

Combining all terms with explicit dependencies:

3. Ion-Surface Interaction Physics

3.1 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:

3.2 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: $\sigma_\theta \approx 2° - 5°$

3.3 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:

3.4 Ion-Enhanced Etching Yield

The total etch yield combines mechanisms:

Synergistic enhancement factor:

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

4. Plasma Modeling (Reactor Scale)

4.1 Species Continuity Equations

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

Flux expressions:

• Drift-diffusion: $\vec{\Gamma}_i = -D_i

abla n_i + \mu_i n_i \vec{E}$

• Full momentum: $\vec{\Gamma}_i = n_i \vec{v}_i$ with momentum equation

Source/sink terms:

4.2 Electron Energy Balance

Heat flux:

Power absorption (ICP):

Collisional losses:

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

4.3 Plasma Conductivity

Skin depth:

4.4 Electromagnetic Field Equations

Maxwell's equations (frequency domain):

Wave equation:

4.5 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:

5. Feature-Scale Profile Evolution

5.1 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

5.2 Hamilton-Jacobi Formulation

Hamiltonian for etch:

With $V_n$ dependent on:

• Local surface normal: $\hat{n} = -

abla\phi / | abla\phi|$

• Local fluxes: $\Gamma(\vec{r})$
• Local angles: $\theta = \arccos(\hat{n} \cdot \hat{z})$

5.3 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).

5.4 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:

6. Aluminum-Specific Phenomena

6.1 Native Oxide Breakthrough

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

Why BCl₃ is critical:

1. Heavy $\text{BCl}_3^+$ ions provide efficient momentum transfer 2. BCl₃ scavenges oxygen chemically:

Breakthrough time:

6.2 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

6.3 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

6.4 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

7. Numerical Methods

7.1 Level Set Discretization

Upwind Finite Differences

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

One-sided derivatives:

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

Reinitialization

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

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

7.2 Monte Carlo Feature-Scale Simulation

Algorithm:

1. INITIALIZE surface mesh 2. 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 3. 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

7.3 Coupled Multi-Scale Modeling

Coupling strategy:

7.4 Plasma Solver Discretization

Finite element for Poisson's equation:

Weak form:

Finite volume for transport:

8. Process Window and Optimization

8.1 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:

8.2 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)

8.3 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)

8.4 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:

9. Equations:

abla\phi| = 0$ |

abla \cdot \vec{\Gamma}_i = S_i - L_i$ |

10. Modern Developments

10.1 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:

10.2 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

10.3 Digital Twins

Components:

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

Update equation:

10.4 Uncertainty Quantification

Bayesian calibration:

Propagation through models:

Monte Carlo uncertainty:

Physical Constants

Process Conditions

Source: ChipFoundryServices — Search this topic — Ask CFSGPT