🧪 Module 2 · Chemistry and Materials Science · Chapter 2.6 · 13 min read

Plasma Etching — ICP-RIE

Carving nanometres with the dance of ions and radicals.

Prerequisites

2.5 — Lithography Chemistry

What you'll learn here

Explain why plasma (dry) etch is anisotropic, unlike wet etch
Distinguish RIE from ICP-RIE and the advantage of dual-RF control
Define and compute anisotropy, selectivity, and etch rate
Pick the right chemistry for each layer of the SIDRA memristor stack (HfO₂/NbOx/TiN)
Describe the Bosch process (SF₆/C₄F₈) and its use case

Hook: Carving the Pattern

Photolithography (Chapter 2.5) leaves a map on the resist: open windows reveal the layer underneath. Next question: how do we get down through those windows?

The answer is plasma etching. Ionize a gas, direct the cloud toward the wafer. Ions and reactive radicals work together — one chemically peels off surface atoms, the other physically launches them away. Result: nanometre-precision cavities with vertical sidewalls.

Critical steps in the SIDRA chip:

SiO₂ trenches (shallow trench isolation): CF₄/CHF₃ chemistry.
HfO₂ memristor layer: BCl₃/Cl₂ chemistry.
TiN electrodes: Cl₂/Ar chemistry.
Copper interconnects: no plasma — Cu doesn’t etch; it’s planarized by CMP (Chapter 2.7).

This chapter explains how ions, radicals, and polymer sidewall passivation all dance in sync.

Intuition: Chemical + Physical = Anisotropic

Two pure approaches fail:

Approach	Mechanism	Anisotropy	Works for SIDRA?
Wet etch (HF, KOH)	Chemical only	Isotropic (spherical hole)	❌ no nm edges
Pure sputter (Ar⁺)	Physical only	Anisotropic but slow, destroys resist	❌ no selectivity
RIE/ICP	Both	Anisotropic + selective	✅

The idea: in a plasma, electrons are hot (~3 eV), ions are cold (~0.1 eV) but accelerated vertically by the sheath electric field (100-500 V). Ions strike top-down → only horizontal surfaces erode. Sidewalls are shielded by a chemical passivation layer.

Radicals (F*, Cl* — neutral, reactive) etch any surface, but without ion bombardment the reaction is too slow → the sidewall survives. Vertical rate = chemical + physical; lateral rate ≈ chemical only.

Result: 90° sidewalls, 10:1 aspect ratios feasible.

Formalism: RIE, ICP, Anisotropy, Selectivity

L1 · Başlangıç

What is a plasma? A “fourth state” of matter in which an RF electric field (13.56 MHz) ionizes gas at low pressure (1-50 mTorr). Three species:

Electrons: fast (~3 eV), create more ions.
Ions: heavy (Ar⁺, SF₅⁺, …), accelerated by the sheath to 100-500 V onto the wafer.
Radicals: neutral but chemically reactive (F*, Cl*). Strip atoms.

Two basic reactors:

RIE (Reactive Ion Etching): single RF. Plasma density and ion energy are coupled — raise one, the other rises.
ICP-RIE (Inductively Coupled): dual RF. Top coil sets plasma density; bottom bias RF sets ion energy independently.

SIDRA uses ICP-RIE for precise control.

L2 · Tam

Three key metrics:

Etch rate (R): nm/min. Depends on radical flux + ion flux.

R = R_{chem} + R_{phys} \cdot Y(E_{ion})

where $Y$ is sputter yield (the Chapter 2.4 formula).

Selectivity (S): target etch rate / mask (or stop-layer) etch rate.

S = R_{target} / R_{mask}

SIDRA targets $S > 10$ for HfO₂/TiN (resist cap is ~100 nm).

Anisotropy (A): vertical rate / lateral rate.

A = 1 - \frac{R_{lateral}}{R_{vertical}}

$A = 1$ perfectly vertical; $A = 0$ isotropic like wet etch.

Typical SIDRA ICP-RIE parameters (HfO₂ cut):

Chemistry: BCl₃/Cl₂/Ar = 20/10/5 sccm
Pressure: 5 mTorr
ICP power: 500 W (plasma density ~10¹¹ cm⁻³)
Bias power: 50 W → ion energy ~150 V
Etch rate: 5 nm/min
Selectivity HfO₂/SiO₂: ~15

Sheath voltage (L2 intuition):

V_{sheath} \approx \frac{T_e}{2} \ln\left(\frac{M_i}{2\pi m_e}\right) \approx 3\text{-}5 \cdot T_e

$T_e \approx 3$ eV → $V_{sheath} \approx 10\text{-}15$ V. With bias, ions gain 100-500 V.

L3 · Derin

Bosch process (deep Si etch):

Step 1: SF₆ plasma — F* radicals etch Si.

\mathrm{Si} + 4\,\mathrm{F}^\ast \rightarrow \mathrm{SiF}_4 \uparrow

Step 2: C₄F₈ plasma — deposits a fluorocarbon polymer (Teflon-like) on sidewalls.

Step 3: SF₆ again — ion bombardment breaks the polymer at the bottom, sidewalls stay passivated.

Cycle time: 3-10 s. Produces 10-30 µm depth at 10:1 aspect ratio. MEMS, TSVs (through-silicon vias).

ALE (Atomic Layer Etch): the mirror of ALD.

Step 1: Cl₂ adsorption → SiCl_x monolayer on the Si surface.
Step 2: Ar⁺ pulse (low energy, < 50 V) → only one layer is removed.
Per cycle: 0.3-1 Å. Very slow but atomic control. Used in GAA fin shaping.

Selectivity engineering (fluorocarbon chemistry):

C₄F₈, CHF₃ deposit sidewall polymer that’s more stable on SiO₂ than on Si.
SiO₂/Si selectivity: 20-40.
HfO₂/SiO₂: 10-15 with BCl₃-based chemistry (strong Cl-metal bonds).

Cryogenic etch (-100°C): wafer cooled with liquid N₂. SF₆/O₂ plasma → SiO_xF_y sidewall passivation forms spontaneously. Smoother than Bosch, single-step. ASML/Oxford tools.

Microloading: at 10% open area etch is nominal; at 90% the radicals deplete → slows down. SIDRA design rules add “dummy fill” to suppress this.

Experiment: Mental Simulation — Bias On/Off

An ICP-RIE has two knobs: ICP power (plasma density), bias power (ion energy). Thought experiment:

ICP 500 W + Bias 0 W: high radical density, low ion energy. Chemistry dominates → isotropic. Sidewalls round off — approaches wet etch.
ICP 500 W + Bias 100 W: ions strike at ~250 V. Vertical rate ↑, lateral unchanged → anisotropy rises. Vertical sidewalls.
ICP 0 W + Bias 100 W: weak plasma (capacitive mode). Selectivity collapses, ion bombardment eats the mask. Bad.
ICP 1000 W + Bias 300 W: too aggressive. Fast, but the mask doesn’t survive. Overkill.

Sweet spot for SIDRA memristors: ICP 500 W + Bias 30-80 W. Low bias means minimal ion damage to the memristor layer (the oxygen-vacancy balance must stay intact).

Quick Quiz

1/5Main advantage of plasma etch over wet etch?

Lab Exercise

Etch the SIDRA memristor stack — top-down 50 nm TiN + 5 nm HfO₂ + 10 nm TaN.

Chemistry: BCl₃/Cl₂, ICP 500 W, Bias 60 W.

Etch rates (nm/min): TiN = 25, HfO₂ = 5, TaN = 20. Resist (80 nm): 10 nm/min.

(a) Etch time per layer? (b) Total etch time? (c) Remaining resist after etching? (Include 20% over-etch margin.) (d) Selectivity TiN/HfO₂ and HfO₂/TaN? (e) What if you raise the bias from 60 W to 150 W at the HfO₂ step?

Answers

(a) TiN: 50/25 = 2.0 min. HfO₂: 5/5 = 1.0 min. TaN: 10/20 = 0.5 min.

(b) Nominal total: 3.5 min. With 20% over-etch → 4.2 min.

(d) TiN/HfO₂ = 25/5 = 5. HfO₂/TaN = 5/20 = 0.25 → TaN etches faster than HfO₂. In practice this means HfO₂ serves as a “rate drop” signal: reaching TaN accelerates and is caught by OES (optical emission spectroscopy) endpoint detection.

(e) Bias ↑ → ion energy ↑ → vertical rate ↑, but selectivity ↓ (resist etches faster too) and memristor damage ↑ (ions penetrate the layer, disrupt the oxygen-vacancy regime). HRS state can be lost. That’s why low bias is chosen.

Cheat Sheet

Plasma etching: chemical (radicals) + physical (ions) = anisotropic, selective.
ICP-RIE: top-RF sets plasma density, bottom-RF sets ion energy (decoupled).
Three numbers: etch rate, anisotropy $A$ , selectivity $S$ .
SIDRA chemistries: SiO₂ → CF₄/CHF₃; HfO₂ → BCl₃/Cl₂; TiN → Cl₂/Ar; Si → SF₆/C₄F₈ (Bosch).
No plasma for Cu — planarize via CMP.
Low bias for memristors (50-80 W) — ion damage = oxygen-vacancy disruption.

Vision: The Future of Etching

ALE (Atomic Layer Etch): mirror of ALD, one monolayer per cycle. Critical for GAA fins, 3D NAND, sub-2 nm features.
Cryogenic etch (-100°C): Si wafer cooled with liquid N₂; sidewall passivation forms spontaneously. Smoother than Bosch, single-step.
Pulse plasma: RF on/off duty cycle → less surface charge damage, better selectivity. Routine below 28 nm.
Neutral beam etch: neutralize the ion, keep only kinetic energy. Damage-free → for 2D material etching.
Plasma-free etch (HF vapor, XeF₂): thermal or spontaneous reactions on oxides and silicon. Critical on soft masks.
Ultra-HAR etch: 50:1 aspect ratio for 200+ layer 3D NAND (Lam, TEL).
Self-limiting etch: stops when a reactant is depleted — extreme uniformity, helps EUV LER.
In-situ metrology: OES + reflectometry measure etch depth live; sub-1 nm accuracy.

Biggest lever for post-Y10 SIDRA: ALE for the memristor stack — ion energy < 50 V, one monolayer removed per cycle. HfO₂ can be etched without disrupting the oxygen-vacancy balance → memristor variance drops from ~5% to ~2%. 2027–2029 horizon.

Prerequisites

What you'll learn here

🪝 Hook: Carving the Pattern

🧭 Intuition: Chemical + Physical = Anisotropic

📐 Formalism: RIE, ICP, Anisotropy, Selectivity

🧪 Experiment: Mental Simulation — Bias On/Off

📝 Quick Quiz

🛠️ Lab Exercise

🗂️ Cheat Sheet

🔮 Vision: The Future of Etching

📚 Further Reading