Semiconductor Manufacturing Etch Endpoint Process

Keywords: endpoint detection, etch endpoint, optical emission spectroscopy, OES, interferometry, endpoint monitoring, process control

Semiconductor Manufacturing Etch Endpoint Process

Overview

In semiconductor fabrication, etching selectively removes material from wafers to create circuit patterns. The endpoint detection problem is determining precisely when to stop etching.

The Core Challenge

Why Endpoint Detection Matters

• Under-etching: Leaves residual material → defects, shorts, incomplete patterns
• Over-etching: Damages underlying layers → profile degradation, reliability issues

At advanced nodes (3nm, 5nm), tolerances are measured in angstroms:

Primary Endpoint Detection Techniques

1. Optical Emission Spectroscopy (OES)

The most widely used technique for plasma (dry) etching.

Principle

During plasma etching, reactive species and etch byproducts emit characteristic photons. The emission intensity $I(\lambda)$ at wavelength $\lambda$ follows:

Where:

• $n_{\text{species}}$ = density of emitting species
• $\sigma_{\text{emission}}$ = emission cross-section
• $E_{\text{plasma}}$ = plasma excitation energy

Key Wavelengths for Common Etch Chemistries

Signal Processing

The endpoint is typically detected using derivative methods:

Endpoint trigger condition:

Advantages

• Non-contact, non-destructive measurement
• Real-time monitoring capability
• Works across entire wafer surface

Limitations

• Weak signals for very thin films ($d < 10$ nm)
• Pattern density affects signal intensity
• Requires optical access to plasma chamber

2. Laser Interferometry

Principle

A monochromatic laser beam reflects from the wafer surface. As etching progresses, film thickness changes alter the interference pattern.

The reflected intensity follows:

Where:

• $I_1, I_2$ = intensities from top surface and interface reflections
• $n$ = refractive index of the film
• $d$ = film thickness
• $\lambda$ = laser wavelength
• $\phi_0$ = initial phase offset

Fringe Analysis

Each complete oscillation (fringe) corresponds to:

Example calculation for SiO₂ with HeNe laser ($\lambda = 632.8$ nm):

Etch Rate Determination

Where $T_{\text{fringe}}$ is the period of one complete oscillation.

Advantages

• Quantitative thickness measurement
• Real-time etch rate monitoring
• High precision for transparent films

Limitations

• Requires optically transparent or semi-transparent films
• Pattern density complicates signal interpretation
• Multiple interfaces create complex interference

3. Residual Gas Analysis (Mass Spectrometry)

Principle

Analyze exhaust gas composition. Different materials produce different volatile byproducts:

Example Reactions

Silicon etching with fluorine:

Oxide etching with fluorine:

Aluminum etching with chlorine:

Mass-to-Charge Ratios

Advantages

• Works regardless of optical properties
• Chemically specific detection
• Can detect multiple transitions

Limitations

• Response time limited by gas transport: $\tau \approx 0.5-2$ s
• Requires differential pumping
• Sensitivity issues at low etch rates

4. RF Impedance Monitoring

Principle

Plasma impedance changes when material composition changes. The plasma can be modeled as:

Monitored Parameters

• Voltage: $V_{\text{RF}}$
• Current: $I_{\text{RF}}$
• Phase: $\phi = \arctan\left(\frac{X}{R}\right)$
• Impedance magnitude: $|Z| = \sqrt{R^2 + X^2}$

Advantages

• Uses existing RF infrastructure
• No additional optical access needed
• Sensitive to plasma chemistry changes

Limitations

• Subtle signal changes
• Affected by many process parameters
• Requires sophisticated signal processing

Advanced Considerations

Aspect Ratio Dependent Etching (ARDE)

High aspect ratio (HAR) features etch slower due to transport limitations:

Where:

• $AR = \frac{\text{depth}}{\text{width}}$ = aspect ratio
• $AR_c$ = characteristic aspect ratio (process-dependent)

Consequence: Dense arrays reach endpoint before isolated features.

Pattern Loading Effect

Local etch rate depends on pattern density $\rho$:

Where $K$ is the loading coefficient.

Selectivity

The selectivity $S$ between materials A and B:

Higher selectivity allows more overetch margin:

Practical Endpoint Strategy

Overetch Calculation

Total etch time:

Overetch percentage:

Typical values: 20-50% depending on uniformity and selectivity.

Statistical Process Control

Endpoint time follows a distribution:

Control limits:

Multi-Sensor Fusion

Modern systems combine multiple techniques:

Where weights $w_i$ are optimized by machine learning algorithms.

Sensor Contributions

Key Equations Summary

Interferometry

OES Endpoint Trigger

Selectivity

ARDE Model

Conclusion

Etch endpoint detection is critical for:

1. Yield: Complete clearing without damage 2. Uniformity: Consistent results across wafer 3. Reliability: Device performance and longevity

The combination of OES, interferometry, mass spectrometry, and RF monitoring—enhanced by machine learning—enables the precision required for sub-10nm semiconductor manufacturing.

Source: ChipFoundryServices — Search this topic — Ask CFSGPT