Reactive Ion Beam Etching (RIBE)
- RIBE is a directional dry-etching technique that uses a collimated, accelerated beam of reactive ions, such as O₂⁺, to achieve precise anisotropic pattern transfer.
- The process leverages an external Kaufman–Robinson ion gun to independently control beam energy, ion current, and incident angle, optimizing etch rates and selectivity.
- Recent advancements in reactive ion beam angled etching (RIBAE) have enabled the fabrication of complex, non-planar photonic structures with high optical quality using tailored mask stacks.
Reactive Ion Beam Etching (RIBE) is a highly directional dry-etching technique, employing a collimated and accelerated beam of reactive ions—typically O₂⁺—to achieve anisotropic pattern transfer in bulk substrates. Unlike conventional inductively coupled plasma (ICP) etching, RIBE uses an external Kaufman–Robinson ion gun to generate a controlled ion beam with independently adjustable beam energy, ion current, and incident angle. Recent adaptations, especially in the form of reactive ion beam angled etching (RIBAE), have enabled the fabrication of complex non-planar geometries within materials such as diamond, supporting photonic devices at both visible and telecommunication wavelengths (Chia et al., 2022).
1. Apparatus and Process Fundamentals
RIBE systems are equipped with a collimated ion source external to the vacuum chamber. Ions are accelerated across a series of grids before impinging on the sample, which can be tilted and rotated to set the ion incidence angle with respect to the substrate normal. Operational parameters critically influencing the etch include beam voltage (), accelerator voltage, beam current (), ion species (notably O₂⁺ for diamond), and background pressure. In RIBAE, the substrate is typically tilted at a fixed angle (α = 45°) and rotated at ∼1 rpm to maintain symmetric, azimuthally uniform etching and undercut.
Two principal “recipes” for diamond RIBAE have been established:
- Recipe A (high-energy): = 200 V, = 100 mA, ≈ 3.8 nm/min
- Recipe B (low-energy): = 150 V, = 80 mA, ≈ 2.3 nm/min
Here, is the lateral diamond etch rate and is experimentally distinguished from the mask erosion rate . Etch selectivity is defined as 0 (Chia et al., 2022).
2. Mask Stack Engineering and Material Selection
Critical to the RIBE process is the selection and engineering of robust mask stacks able to withstand prolonged ion exposure without excessive erosion, lateral undercutting, or roughness propagation. Four mask schemes have been systematically investigated for diamond RIBAE:
| Mask Stack | Mask Thickness | Selectivity 1 | RMS Roughness |
|---|---|---|---|
| HSQ–Ti | 1 μm HSQ / 40 nm Ti | ≤ 5 | 2–3 nm |
| HSQ–Nb | 1 μm HSQ + 200 nm Nb | ≥ 20 | 5 nm (grainy) |
| PMMA/Nb | 800 nm PMMA → 400 nm Nb | ≥ 10 (locally) | 1–2 nm |
| HSQ–Al₂O₃ | 1 μm HSQ / 1 nm Al₂O₃ | ≈ 4.6 | ≤ 1 nm |
- HSQ–Ti (hydrogen silesquioxane with titanium): Compatible with visible-wavelength devices, but low selectivity (2) and substantial redeposition limit large-scale etching.
- HSQ–Nb (HSQ with niobium metal): Achieves very high selectivity (S ≥ 20), but grain-induced roughness in Nb can roughen photonic crystal holes.
- PMMA/Nb (template-inverted): Effective submicron resolution inherited from PMMA, but aspect-ratio and conformality constraints limit use in larger devices.
- HSQ–Al₂O₃ (HSQ with amorphous alumina by ALD): Exhibits minimal roughness (≤ 1 nm RMS) and balanced selectivity, giving optimal profiles for telecommunication-scale structures (Chia et al., 2022).
3. Etch Profile Formation and Device Geometry
RIBAE typically produces triangular cross-sections whose thickness (3) and width (4) are linked by the measured etch angle (5) through 6. Mask choice directly affects this angle as well as sidewall smoothness and undercut uniformity. For example, HSQ–Al₂O₃ masks yielded apex angles of 7 58° (with RMS roughness ≤ 1 nm), while metal-based masks (HSQ–Nb) introduced grain-induced scattering in feature sidewalls. During etching, lateral mask erosion and the time required to undercut determine final device suspension, with nanometer-scale sidewall deviations possible through precise stack optimization.
4. Process Optimization and Analytical Modeling
Optimization is achieved by sweeping ion beam voltage (150–200 V), ion current (80–100 mA), and mask thickness (0.5–1.5 μm) to maximize selectivity and minimize mask erosion:
- Increasing beam voltage 8 and current density 9 raises 0 but also increases 1, reducing selectivity.
- Lower 2 reduces 3 at the expense of a slower 4, leading to longer etch times and increased lateral mask erosion.
- Mask thickness should satisfy 5, where 6 is the lateral undercut time.
- Empirical modeling links apex-angle shift 7 to 8 and 9: 0, showing agreement within 1 across recipes (Chia et al., 2022).
5. Optical Performance and Limitations
Ring resonators and photonic crystal devices fabricated via RIBAE with optimized mask stacks exhibit high optical quality factors (2), with observed values ranging from 3 (HSQ–Al₂O₃) to 4 (HSQ–Nb) at telecommunication wavelengths. HSQ–Ti masks, while sufficient for visible-wavelength devices, fail at larger scales due to redeposition and lateral erosion, resulting in incomplete suspension or degraded optical performance. Grain structure in Nb mitigates 5 enhancements in photonic crystal holes despite overall higher selectivity.
6. Transferability to Other Materials and Outlook
The balancing principles of mask thickness, selectivity, aspect ratio tolerance, and smoothness established in diamond RIBAE apply directly to angled etching of other bulk dielectrics—such as SiC, GaN, and quartz—where thin-film platforms are unavailable. Mask stacks must be engineered so that: (1) 6 underetch time, (2) 7, (3) lithographic aspect-ratio tolerance is maintained, and (4) RMS roughness remains ≲ 1 nm.
Future improvements may include stress-free deposition of thick dielectrics (e.g., ALD-grown SiO₂ or TiO₂) and hybrid metal–oxide mask stacks to enable non-planar etch processes for even wider (>2 μm) suspended structures without sacrificing surface quality (Chia et al., 2022).