Improving feature resolution and pore back effect in focused ion beam tomography of porous GaN thin films

Technical Summary: Improving Feature Resolution and Pore Back Effect in Focused Ion Beam Tomography of Porous GaN Thin Films

Problem Statement
Porous gallium nitride (GaN) is a critical mesoporous material for electronics and optoelectronics, yet its structural control is limited by a poor understanding of the electrochemical mechanisms governing pore formation. High-quality three-dimensional characterization is essential to bridge this gap. While Focused Ion Beam (FIB)-SEM tomography is a powerful tool for this purpose, its application to porous GaN is significantly hindered by two primary issues:

1. The Pore Back Effect: In porous materials, secondary and back-scattered electrons from the back wall of a pore can escape and be detected when imaging a cross-section. This creates unwanted contrast in frames where the pore back is not yet physically exposed, leading to a "smearing" of features along the milling direction. This distorts the reconstructed 3D volume and complicates quantitative analysis.
2. Voxel Anisotropy: Due to physical limitations in ion beam milling speeds and spot size, slice thickness (typically 5 nm) is often significantly larger than the in-plane pixel size (1 nm). This anisotropy compounds the resolution loss caused by the pore back effect, particularly along the milling direction.

Current literature lacks formal methods to quantify the severity of the pore back effect, and existing mitigation strategies (such as segmentation algorithms) often address the symptom rather than the root cause.

Methodology
The authors propose and evaluate a "plan-view" tomography methodology as an alternative to the conventional "cross-sectional" approach.

• Sample Preparation: Three porous GaN samples were fabricated via electrochemical etching (ECE) under varying conditions (voltage, electrolyte species, and cell configuration) to produce distinct pore morphologies: wide columnar pores (Sample 1), extremely narrow columnar pores with surface branching (Sample 2), and highly branched, tortuous pores (Sample 3).
• Conventional Approach (Cross-Sectional): Samples were mounted flat. The ion beam milled trenches perpendicular to the film surface, and the electron beam imaged the cross-section. The milling direction was parallel to the film surface.
• Proposed Approach (Plan-View): Samples were cleaved and mounted on angled stubs such that the ion beam milled parallel to the film surface (perpendicular to the original surface normal). The electron beam imaged the exposed surface of the porous layer as the ion beam sliced through it. This aligns the milling direction with the long axis of vertically etching pores.
• Evaluation Metrics: To quantify the pore back effect, the authors introduced two formalisms:
  1. Voxel Intensity Distribution: Analyzing histograms of voxel intensities. An ideal binary composite (GaN matrix vs. void) should show two distinct peaks. The pore back effect smears the low-intensity (void) peak into higher intensities, creating a continuous distribution.
  2. Feature Anisotropy: Extracting line profiles across pores. In the presence of the pore back effect, features appear asymmetric along the milling direction (darker at the "front" and brighter at the "back" due to shine-through), whereas profiles perpendicular to the milling direction remain symmetric.

Key Results
The study compared the two methodologies across the three distinct sample morphologies:

• Sample 1 (Wide Columnar Pores): The plan-view approach demonstrated a significant advantage. The reconstructed cross-sectional images showed high contrast and clear pore wall definition. In contrast, the cross-sectional tomography suffered from severe feature anisotropy and intensity smearing. Intensity histograms for the plan-view data showed a distinct low-intensity peak (void) and a high-intensity peak (GaN), whereas the cross-sectional data collapsed into a single continuous peak, indicating a severe pore back effect.
• Sample 2 (Narrow Columnar Pores): Both methods struggled due to the extremely fine feature size (pores imaged in only 2–5 slices). However, the plan-view approach still provided better contrast and a slightly broader intensity distribution toward the low-intensity range, suggesting partial mitigation of the pore back effect, though the resolution remained limited by the pore dimensions relative to the slice thickness.
• Sample 3 (Branched/Tortuous Pores): For pores lacking a clear dominant vertical axis, neither method held a decisive advantage. Both approaches suffered from the pore back effect, and intensity histograms for both showed similar, poorly resolved distributions. This suggests the plan-view method is most effective for pores with a strong vertical alignment.

Significance and Claims
The paper claims that rotating the tomography experiment to a plan-view geometry significantly mitigates the pore back effect in porous GaN thin films with vertically oriented pores, without increasing the total experimental duration.

• Quantification: The authors successfully introduced voxel intensity-based formalisms to quantify the extent of the pore back effect, moving beyond qualitative assessment.
• Resolution Improvement: By aligning the milling direction with the pore axis, the plan-view method reduces the distance electrons must travel to escape the pore, thereby reducing the "shine-through" artifact and improving feature resolution and segmentation accuracy.
• Limitations: The authors modestly note that this advantage is specific to anisotropic pore morphologies (vertical pores). For isotropic or highly branched pores (like Sample 3), the benefit is negligible, and the method does not overcome the fundamental resolution limits imposed by voxel anisotropy and pore size.
• Future Outlook: The paper suggests that while segmentation algorithms are improving, preventing the pore back effect through experimental geometry remains crucial for accurate material characterization. They propose that simulation workflows could be developed to optimize tomography orientation and beam parameters before conducting physical experiments.