Future of Plasma Etching for Microelectronics — Key Trends and Roadmap

By NineScrolls Engineering · 2025-09-08 · 18 min read · Nanotechnology

Target Readers: Process engineers, equipment evaluators, and R&D leaders in semiconductor fabrication, MEMS, photonics, and advanced packaging who need to plan etch capability roadmaps for sub-3 nm nodes and next-generation device architectures.

1) Why the Future of Plasma Etching Matters

Plasma etching has been a cornerstone of semiconductor manufacturing for over four decades, enabling the pattern transfer that defines every transistor, interconnect, and via on a chip. As the industry pushes beyond the 3 nm process node, the demands on etch technology are intensifying in ways that existing continuous-mode plasma etching processes struggle to meet.

Three converging forces are driving this transformation:

• Dimensional scaling: Gate-all-around (GAA) nanosheet transistors, backside power delivery networks, and 3D NAND stacks exceeding 200 layers demand etch precision at the atomic level — tolerances of a few angstroms across features just a few nanometers wide.
• Heterogeneous integration: Chiplet-based architectures, hybrid bonding, and through-silicon vias (TSVs) introduce a growing diversity of material stacks — from conventional silicon and SiO₂ to III-V compound semiconductors (GaN, InGaAs), 2D materials (MoS₂, WSe₂), and ultralow-k dielectrics — each with unique etch chemistries and damage sensitivities.
• Yield and cost pressure: At advanced nodes, a single etch excursion can scrap wafers worth tens of thousands of dollars. Real-time process control, tighter uniformity specifications (< 1% within-wafer CD variation), and predictive maintenance are no longer optional — they are economic imperatives.

The sections that follow examine the key technology directions that address these challenges, provide quantitative benchmarks where available, compare new approaches with conventional methods, and offer practical adoption guidance for labs and fabs evaluating their etch roadmaps.

2) Atomic Layer Etching (ALE)

2.1 How ALE Works

Atomic layer etching is a cyclic, self-limiting process that removes material one atomic layer at a time. Each ALE cycle consists of two half-reactions separated by purge steps:

1. Surface modification: A reactive species (e.g., Cl₂ plasma for silicon, or fluorocarbon radicals for oxides) adsorbs on the top atomic layer, forming a modified surface layer with altered bond strength. The reaction is self-limiting — once the surface is fully saturated, additional exposure produces no further modification.
2. Purge: Excess reactants and byproducts are evacuated from the chamber.
3. Removal: A low-energy ion beam (typically Ar⁺ at 10–50 eV) or thermal activation selectively removes only the modified layer. Because the underlying unmodified material has higher bond energy, it remains intact — this is the second self-limiting step.
4. Purge: Removal byproducts are evacuated before the next cycle begins.

Figure 1: The ALE Cycle — Four-step self-limiting process consisting of surface modification (reactive adsorption), purge, energetic removal of the modified layer, and a final purge. Each cycle removes a precisely controlled amount of material, typically 0.5–2 Å per cycle

2.2 Thermal ALE vs Plasma-Enhanced ALE

Two main variants of ALE have emerged, each suited to different applications:

• Thermal ALE: Uses thermally driven ligand-exchange reactions for removal. Operates without ion bombardment, making it inherently isotropic and damage-free. Ideal for selective lateral etching of sacrificial layers (e.g., SiGe selective to Si in GAA nanosheet release) and conformal etching inside high-AR features. Typical etch-per-cycle (EPC): 0.5–1.5 Å/cycle at 200–300 °C.
• Plasma-enhanced (directional) ALE: Uses low-energy ions for the removal step, providing anisotropic etching. Essential for patterning operations requiring vertical sidewalls — gate spacer etching, fin recess, and contact hole definition. Typical EPC: 1–2 Å/cycle with ion energies of 10–50 eV. The key advantage over continuous RIE is the decoupling of chemical modification from physical removal, which eliminates the selectivity-versus-anisotropy tradeoff that limits conventional etch.

2.3 Applications and Performance Benchmarks

ALE is already in production for several critical etch steps at advanced foundries:

• Gate spacer recess: SiN ALE achieves < 0.3 nm depth uniformity across 300 mm wafers, compared to 0.8–1.2 nm with conventional RIE — a 3–4× improvement in process control.
• Fin and nanosheet trimming: Silicon ALE enables sub-nanometer CD control for fin width adjustment, critical for threshold voltage tuning at the 3 nm node.
• Self-aligned contact etch: Selective oxide ALE with > 100:1 selectivity to SiN etch-stop layers, enabling reliable contact formation without spacer erosion.
• Damage-sensitive interfaces: ALE of high-k dielectrics (HfO₂, ZrO₂) preserves interface quality, maintaining equivalent oxide thickness (EOT) within 0.1 nm of target.

The primary tradeoff is throughput: ALE cycles take 5–30 seconds each, making full-wafer etch of thick films impractical. A common strategy is to use continuous etch for bulk material removal, then switch to ALE for the final precision etching — a hybrid approach that balances throughput with atomic-level control.

3) Pulsed Plasma and Pulsed Bias Techniques

3.1 The Principle of Pulsed Etching

In conventional continuous-wave (CW) plasma etching, the source RF power and substrate bias operate at steady state. Pulsed plasma techniques modulate one or both of these power sources on and off at controlled frequencies (typically 100 Hz – 10 kHz) and duty cycles (10–90%). This temporal modulation provides control knobs that do not exist in CW operation:

• Source pulsing: Modulates plasma density and radical production. During the off-phase, high-energy electrons thermalize while lower-energy electrons sustain negative ion populations — shifting the electron energy distribution function (EEDF) toward lower energies and reducing UV damage to the substrate.
• Bias pulsing: Modulates ion bombardment energy independently of plasma chemistry. During the off-phase, surface charging dissipates, reducing charge-induced damage in high-AR features and gate dielectrics.
• Synchronous pulsing: Source and bias pulse in phase, providing maximum control over the ion-to-neutral ratio during each pulse cycle.
• Asynchronous pulsing: Source and bias pulse at different frequencies or phase offsets, enabling independent tuning of radical flux (chemistry) and ion energy (directionality).

Figure 2: Pulsed Plasma Operating Modes — Timing diagrams comparing synchronous pulsing (source and bias in phase) and asynchronous pulsing (independent frequencies), illustrating how duty cycle and phase offset provide additional process control dimensions beyond CW operation

3.2 Benefits and Quantitative Impact

Pulsed techniques deliver measurable improvements across several critical etch metrics:

• Selectivity enhancement: Pulsed-bias SiO₂/SiN etch achieves selectivity of 15–20:1 compared to 8–10:1 in CW mode, due to reduced ion-driven sputtering of the etch-stop layer during bias-off periods.
• Damage reduction: Gate oxide integrity improves by 2–5× (measured by charge-to-breakdown Q_bd) because charging damage and UV exposure are substantially reduced during off-phases.
• Profile control in HAR features: Pulsing reduces bowing in deep silicon trenches by 30–50% compared to CW operation at the same etch rate, as ion angular distribution narrows during the lower-plasma-density off-phase.
• Etch rate tunability: By adjusting duty cycle from 20% to 80%, etch rates can be modulated over a 4:1 range without changing gas chemistry or pressure — providing a convenient process tuning knob.

Pulsed operation is particularly valuable as a retrofit capability: many modern ICP-RIE platforms support pulse mode through RF generator upgrades without requiring chamber hardware changes, making it an accessible first step toward advanced etch control.

4) High Aspect Ratio (HAR) Etching Challenges

4.1 Why HAR Etching Is Increasingly Critical

The aspect ratio (depth ÷ width) of features that plasma etching must produce has increased dramatically. 3D NAND channel holes now exceed 100:1 AR at depths of 8–10 μm; DRAM capacitor trenches approach 50:1; and TSV interconnects require 10:1 to 20:1 AR through hundreds of micrometers of silicon. At these extreme geometries, the physics of ion and neutral species transport within the feature fundamentally change.

4.2 Key HAR Defect Mechanisms

Figure 3: HAR Etch Defect Mechanisms — Cross-section views illustrating (a) ARDE: reduced etch rate in narrower features, (b) bowing: lateral widening from scattered ions, (c) twisting: angular deviation of the feature axis, and (d) necking: constriction near the feature opening due to polymer buildup

Several defect mechanisms emerge at high aspect ratios, each driven by distinct physical processes:

• ARDE (Aspect Ratio Dependent Etching): Etch rate decreases as aspect ratio increases because fewer ions and neutrals reach the bottom of deeper, narrower features. In extreme cases, etch rate at 100:1 AR can be 50–70% lower than at 10:1 AR in the same wafer. This causes depth non-uniformity across features of different widths — a critical problem for DRIE and 3D NAND processes. See our detailed guide on DRIE and the Bosch process for more on ARDE mitigation.
• Bowing: Ions that scatter off sidewalls or are deflected by local electric fields cause lateral etching below the mask opening, widening the feature mid-depth. Bowing is exacerbated by charging effects in insulating materials and can cause shorts between adjacent features.
• Twisting and distortion: In dense arrays, asymmetric ion shadowing and stochastic variations in local etch rates cause features to deviate from their intended vertical axis. Twisting becomes significant at pitches below 40 nm and AR above 60:1.
• Necking and clogging: Polymer deposition and mask erosion products accumulate near the feature opening, constricting the aperture and starving the bottom of reactive species. Necking can cause complete etch stop in extreme cases.

4.3 Mitigation Strategies

Addressing HAR etch challenges requires a combination of equipment capability and process innovation:

• Pulsed bias for ARDE compensation: Time-modulated bias allows ions to accumulate directionality during off-phases before the next acceleration pulse, narrowing the ion angular distribution and improving transport to feature bottoms.
• Gas modulation and multi-step recipes: Alternating between etch-dominant and passivation-dominant steps (similar to the Bosch process concept) can manage polymer buildup while maintaining etch directionality. For 3D NAND, multi-step recipes with progressive chemistry changes compensate for ARDE as the feature deepens.
• Tilt and rotation control: Wafer tilting during etch can compensate for systematic twisting in dense arrays by rebalancing ion incidence angles. Some advanced tools incorporate in-situ tilt stages for this purpose.
• Higher plasma density sources: ICP-RIE sources generating plasma densities of 10¹¹–10¹² cm⁻³ produce higher ion flux at the wafer, improving species transport into deep features compared to capacitively coupled systems.

5) Low-Damage and Selective Etching

5.1 Damage Mechanisms in Plasma Etching

Every plasma etch step introduces some degree of substrate damage. At advanced nodes where device layers are only a few nanometers thick, even minor damage degrades performance. The primary damage pathways are:

• Ion bombardment damage: Energetic ions (> 50 eV) displace lattice atoms, creating vacancy-interstitial pairs that degrade carrier mobility. In III-V channels (InGaAs, GaN), ion damage can increase contact resistance by 10× or more.
• UV and VUV radiation: Photons with energies above 9 eV (vacuum UV) generate electron-hole pairs in gate dielectrics, leading to trapped charge and threshold voltage shifts. VUV damage is particularly problematic for high-k/metal gate stacks.
• Radical penetration: Reactive fluorine and chlorine radicals can diffuse several nanometers into exposed surfaces, altering composition and creating subsurface defect layers. This is especially damaging to 2D materials (MoS₂, graphene) where even a single disrupted atomic layer significantly impacts electrical properties.

5.2 Low-Damage Process Approaches

Several strategies minimize etch damage while maintaining acceptable throughput:

• Multi-frequency RF: Dual-frequency systems (e.g., 2 MHz + 60 MHz) decouple ion energy control (low frequency) from plasma density (high frequency). By running the bias frequency at 2 MHz with reduced power, ion energies can be kept below 20 eV — well under the displacement threshold for most semiconductors — while maintaining adequate plasma density for reasonable etch rates (50–200 nm/min).
• Remote plasma and downstream processing: Generating plasma upstream and delivering only neutral radicals to the wafer eliminates ion bombardment entirely. Downstream ashing and surface cleaning with O₂ or forming gas (H₂/N₂) plasmas achieve damage-free removal of photoresist and surface contaminants at rates of 1–5 μm/min.
• ALE for ultimate damage control: As described in Section 2, the self-limiting nature of ALE allows ion energies to be set at or just above the modified-layer removal threshold, minimizing subsurface penetration. ALE of SiN achieves surface roughness < 0.2 nm RMS with no detectable damage layer by XPS or TEM.
• Gentle chemistries: Replacing aggressive fluorocarbon plasmas with milder alternatives — such as HBr/O₂ for silicon or BCl₃/N₂ for III-V compounds — reduces the chemical aggressiveness while maintaining selectivity. For 2D materials, low-power XeF₂ vapor etching or remote O₂ plasma provides layer-by-layer removal with sub-nanometer control.

5.3 Selectivity at Advanced Nodes

Selectivity — the ratio of etch rates between the target material and adjacent layers — becomes increasingly challenging as film thicknesses shrink. A selectivity of 10:1 that was adequate when etch-stop layers were 20 nm thick becomes insufficient when those layers are only 2–3 nm. Current selectivity targets at the 3 nm node include:

• SiO₂-to-SiN: > 20:1 (self-aligned contact etch)
• SiGe-to-Si: > 100:1 (nanosheet release)
• TiN-to-HfO₂: > 50:1 (metal gate patterning)
• Low-k-to-Cu: effectively infinite (no Cu attack during dual-damascene etch)

Achieving these targets requires precise control of ion energy (< 30 eV for many selective steps), gas chemistry, and wafer temperature — a combination that often points toward ALE or pulsed-mode processing.

6) EUV Resist Removal and Post-Lithography Cleaning

Extreme ultraviolet (EUV) lithography at 13.5 nm wavelength has introduced a new class of resist materials — metal-oxide (e.g., tin-oxide-based) and chemically amplified resists (CARs) with organic-inorganic hybrid compositions — that present unique etch and strip challenges distinct from conventional 193 nm photoresists.

6.1 Challenges with EUV Resists

• Thinner films: EUV resists are typically 20–40 nm thick (vs. 100–300 nm for DUV resists), requiring etch processes with sub-nanometer depth control to avoid pattern distortion during development or transfer.
• LWR/LER sensitivity: Line width roughness (LWR) and line edge roughness (LER) in EUV patterns are 2–4 nm — a significant fraction of the feature width at sub-20 nm pitch. Any plasma process step that increases roughness by even 0.5 nm can push LWR out of specification.
• Metal-containing residues: Tin-oxide-based resists leave non-volatile metallic residues after conventional O₂ plasma strip, requiring additional wet or plasma cleaning steps that add cost and can damage underlying layers.
• Stochastic defects: EUV's low photon count creates stochastic variations in resist profiles. Etch processes must be tolerant of local variations in resist thickness and composition without amplifying these defects.

6.2 Process Solutions

• Low-damage strip: Downstream H₂/N₂ plasma or remote O₂ plasma at reduced power (< 200 W) removes organic components without sputtering metal residues into underlying layers. Typical strip rates: 50–200 nm/min with < 0.2 nm roughness increase.
• Multi-step clean sequences: A combination of plasma strip (organic removal) → wet clean (metal residue removal with dilute HF or SC-1) → surface treatment achieves complete resist removal without pattern degradation.
• In-situ metrology: Optical emission spectroscopy (OES) endpoint detection is critical for stopping the strip process precisely at the resist-substrate interface, avoiding over-etch that would degrade LWR/LER.

7) Cryogenic and Variable-Temperature Processing

Temperature is a powerful but underutilized control parameter in plasma etching. By operating at temperatures far outside the conventional 20–80 °C range, cryogenic and elevated-temperature processes unlock etch behaviors that are difficult or impossible to achieve at room temperature.

7.1 Cryogenic Etching (−100 °C to −40 °C)

At cryogenic temperatures, etch chemistry shifts fundamentally:

• Passivation without polymer: In SF₆/O₂ cryogenic silicon etching, oxygen radicals form a thin SiO_xF_y passivation layer on sidewalls. At −100 °C, this layer is stable enough to prevent lateral etching, but thin enough (1–2 nm) to be removed by directional ion bombardment at the feature bottom. The result is smooth, scallop-free vertical profiles without the cyclic passivation/etch switching of the Bosch process. For a detailed comparison, see our article on Cryogenic Etching vs. the Bosch Process.
• Surface roughness: Cryogenic processes achieve sidewall roughness of < 5 nm Ra — an order of magnitude smoother than Bosch-process scalloping (50–200 nm) — making them suitable for photonic waveguides and MEMS devices where surface quality directly impacts performance.
• Etch rate: Cryogenic silicon etch rates of 3–8 μm/min are achievable with SF₆ flow rates of 100–300 sccm at pressures of 5–20 mTorr, comparable to Bosch process throughput but with superior profile quality.

7.2 Variable-Temperature Processing

Beyond cryogenic operation, dynamically varying wafer temperature during etch opens additional possibilities:

• Temperature ramping: Starting cold (−80 °C) for profile establishment, then warming (+20 °C) for bulk removal, combines the profile quality of cryogenic etch with the throughput of room-temperature processing.
• Elevated temperature (100–250 °C): For certain material systems — particularly III-V compound semiconductors and metal oxides — elevated temperatures promote volatile etch-product formation. InP etching with CH₄/H₂ at 200 °C produces volatile In(CH₃)₃ byproducts, achieving smooth, damage-free surfaces that are difficult to obtain at room temperature.

7.3 Equipment Requirements

Variable-temperature etching places stringent demands on chuck and chamber design:

• Electrostatic chuck (ESC): Must maintain reliable clamping and thermal contact from −150 °C to +250 °C. Helium backside cooling pressure of 5–20 Torr ensures < 5 °C temperature uniformity across 300 mm wafers.
• Condensation management: Below −40 °C, moisture condensation on chamber surfaces becomes a contamination source. Load-lock isolation and dry-gas purge cycles are essential.
• Temperature transition speed: For variable-temperature recipes, the chuck must ramp at > 5 °C/min to keep cycle times practical. Advanced systems use multi-zone heating with liquid nitrogen cooling to achieve 10–20 °C/min ramp rates.

8) AI-Assisted Process Control

8.1 The Data Opportunity in Plasma Etching

Modern etch chambers generate vast quantities of real-time process data — optical emission spectra with hundreds of wavelength channels, RF impedance measurements at millisecond intervals, mass spectrometer readings, and temperature/pressure traces — yet most of this data goes unused in conventional recipe-based process control. Machine learning (ML) and artificial intelligence (AI) can extract actionable information from these data streams to improve etch outcomes.

8.2 Key AI/ML Applications

• Advanced endpoint detection: ML models trained on OES data can detect subtle spectral changes that indicate etch completion of ultra-thin layers (< 2 nm), where conventional threshold-based endpoint fails due to noise. Neural network classifiers achieve endpoint accuracy of ± 0.5 nm, compared to ± 2 nm for traditional intensity-ratio methods.
• Fault detection and classification (FDC): Real-time anomaly detection using RF impedance and OES signatures identifies process excursions (arcing, gas flow deviations, chamber leak) within seconds, enabling automatic recipe abort before wafer damage occurs. Modern FDC systems achieve > 95% detection rates with < 1% false alarm rates.
• Virtual metrology (VM): ML models predict post-etch CD, depth, and profile shape from in-situ sensor data, reducing the need for time-consuming ex-situ metrology (SEM, ellipsometry) and enabling 100% wafer disposition without physical measurement. VM accuracy of ± 0.5 nm CD prediction has been demonstrated in production environments.
• Adaptive recipe optimization: Closed-loop systems that adjust recipe parameters (power, pressure, gas flow, bias) in real-time based on sensor feedback can compensate for chamber drift, incoming wafer variation, and consumable wear. This "digital twin" approach maintains etch performance within specification across thousands of wafers between chamber maintenance events.

8.3 Integration Readiness

For labs and fabs evaluating AI-ready etch platforms, the key infrastructure requirements are:

• Sensor suite: OES (full-spectrum, not filtered), RF V-I probes, mass flow controller feedback, chamber pressure (capacitance manometer), and wafer temperature (pyrometry or embedded sensors).
• Data bandwidth: Minimum 1 kHz sampling across all channels, with on-tool data storage for at least 10,000 wafer runs to build training datasets.
• Control interface: Recipe parameters must be adjustable via software API (not just front-panel controls) to enable closed-loop optimization.
• Edge computing: On-tool GPU or FPGA for real-time inference at < 100 ms latency, essential for within-wafer adaptive control.

9) Equipment Implications and Practical Takeaways

9.1 Platform Requirements for Next-Generation Etching

The technologies described above converge on a common set of ICP-RIE platform requirements:

• Independent source and bias control: Separate RF generators for ICP source (plasma density) and substrate bias (ion energy) are essential for ALE, pulsed processing, and low-damage etch. Both must support pulsing at 100 Hz – 10 kHz with programmable duty cycle.
• Wide temperature range: Electrostatic chuck with operational range from −120 °C to +250 °C for cryogenic and variable-temperature processes, with ≤ 5 °C wafer uniformity.
• Gas delivery flexibility: Multi-gas manifold supporting 6+ gas lines with fast-switching valves (< 200 ms) for ALE cycling and multi-step HAR recipes. Mass flow controllers with 0.1 sccm resolution for precise chemistry control.
• Sensor integration: Full-spectrum OES, RF V-I probes, and data infrastructure for AI/ML-ready operation as described in Section 8.
• Chamber materials: Yttria (Y₂O₃) or alumina (Al₂O₃) coated chamber components to minimize metal contamination and resist aggressive halogen chemistries. Anodized aluminum is acceptable for research tools but insufficient for production environments processing III-V or 2D materials.

9.2 Adoption Roadmap for Labs and Fabs

Not every facility needs to implement all of these technologies simultaneously. A practical adoption sequence based on impact and implementation complexity:

1. Pulsed plasma/bias (immediate): Often a software/generator upgrade on existing ICP-RIE tools. Provides immediate benefits in selectivity and damage reduction. Start with simple synchronous pulsing at 1 kHz, 50% duty cycle, then optimize.
2. ALE-ready operation (near-term): Requires fast gas switching and recipe sequencing capability. Begin with ALE characterization on one or two critical etch steps while running continuous etch for the rest. Most modern ICP-RIE platforms from leading suppliers can support ALE with appropriate gas panel configuration.
3. AI/ML integration (medium-term): Start with data collection and endpoint improvement, then progress to fault detection and virtual metrology as training datasets accumulate. Partner with data science resources for model development.
4. Cryogenic capability (when needed): Requires dedicated chuck and chamber hardware. Justify based on specific application requirements (smooth-walled photonic waveguides, MEMS resonators, or deep silicon features where Bosch process scalloping is unacceptable).

10) Conclusion

The future of plasma etching is not a single breakthrough technology but a convergence of complementary techniques — ALE for atomic precision, pulsed processing for damage control, advanced HAR methods for 3D scaling, and AI for process intelligence. The common thread is a shift from brute-force continuous etching toward precisely controlled, self-limiting, and data-driven processes that match the atomic-scale demands of next-generation devices.

For process engineers and equipment planners, the practical message is clear: invest in ICP-RIE platforms with the flexibility to support these techniques — independent source/bias, pulsing capability, wide temperature range, and sensor infrastructure — even if your current processes don't yet require them. The technology roadmap is moving fast, and the cost of retrofitting capability later far exceeds the cost of specifying it upfront.

Explore NineScrolls Etch Solutions: ICP Etcher Series · RIE Etcher Series · ICP-RIE Technology Guide · Contact Us for Consultation

References

1. Kanarik, K. J., et al. "Overview of atomic layer etching in the semiconductor industry." Journal of Vacuum Science & Technology A, 33(2), 020802 (2015). doi:10.1116/1.4913379
2. Cardinaud, C., Peignon, M.-C. & Tessier, P.-Y. "Plasma etching: principles, mechanisms, application to micro- and nano-technologies." Applied Surface Science, 164(1–4), 72–83 (2000). doi:10.1016/S0169-4332(00)00328-7
3. IRDS (IEEE International Roadmap for Devices and Systems). irds.ieee.org
4. Faraz, T., et al. "Atomic layer etching: what can we learn from atomic layer deposition?" ECS Journal of Solid State Science and Technology, 4(6), N5023–N5032 (2015). doi:10.1149/2.0051506jss
5. Donnelly, V. M. & Kornblit, A. "Plasma etching: yesterday, today, and tomorrow." Journal of Vacuum Science & Technology A, 31(5), 050825 (2013). doi:10.1116/1.4819316
6. Banna, S., et al. "Pulsed high-density plasmas for advanced dry etching processes." Journal of Vacuum Science & Technology A, 30(4), 040801 (2012). doi:10.1116/1.4716176
7. George, S. M. & Lee, Y. "Prospects for thermal atomic layer etching using sequential, self-limiting fluorination and ligand-exchange reactions." ACS Nano, 10(5), 4889–4894 (2016). doi:10.1021/acsnano.6b02991
8. Huard, C. M., et al. "Atomic layer etching of 3D structures in silicon: self-limiting and nonideal reactions." Journal of Vacuum Science & Technology A, 35(3), 031306 (2017). doi:10.1116/1.4979661
9. Marchack, N. & Chang, J. P. "Perspectives in nanoscale plasma etching: what are the ultimate limits?" Journal of Physics D: Applied Physics, 44(17), 174011 (2011). doi:10.1088/0022-3727/44/17/174011