🧪 Module 2 · Chemistry and Materials Science · Chapter 2.2 · 12 min read

HfO₂ — Why Hafnium Oxide?

SIDRA's main material, cast in two roles.

Prerequisites

2.1 — The Chip Side of the Periodic Table

What you'll learn here

Justify HfO₂'s advantage over SiO₂ as a high-k dielectric (ε_r and tunneling)
Name HfO₂'s three crystal phases (monoclinic, tetragonal, cubic) and what each does on chip
Re-state HfO₂ as memristor active layer (oxygen-vacancy filament)
Explain how doping (Al, Y, Zr) tunes endurance/retention

Hook: One Material, Two Lives

SIDRA Y1 has 419 million HfO₂ cells. The same material is both the gate dielectric and the memristor active layer. Not an accident — a design. In 2007 Intel 45 nm replaced SiO₂ with HfO₂ (the “high-k gate oxide” revolution). In 2008 HP Labs showed the same material could switch resistively. Fifteen years later SIDRA uses both roles side-by-side on one chip.

This chapter answers one question: Why Hf? There are 118 elements, but HfO₂ is the unique supplier of a very specific combination.

Intuition: Two Lives, One Lattice

Life 1 — Gate dielectric:

High dielectric constant (ε_r ≈ 25) → 6× thicker film at same C → 10³-10⁴× less tunneling.
Wide bandgap (E_g ≈ 5.7 eV) → transparent in visible, electrons can’t leak over.
Thermodynamic compatibility with Si → thin stable SiO₂ forms at the interface.

Life 2 — Memristor:

Oxygen vacancies (V_O²⁺) are mobile — voltage forms a filament.
LRS/HRS ratio 10-100× → analog multilevel (MLC).
CMOS-compatible → ALD deposition below 400°C (respects BEOL thermal budget).

Same HfO₂, in different lattice phases. Gate oxide is amorphous with controlled crystallization — isotropic insulation. Memristor is amorphous or monoclinic — structural disorder enables vacancy motion. Same molecule, different order, two different jobs.

Formalism: Phases and Properties

L1 · Intro

Three important crystal phases of HfO₂:

Phase	Temperature	ε_r	Role
Monoclinic (m)	Below 1700°C, stable at room T	~20	Conventional high-k
Tetragonal (t)	1700-2600°C	~40	Higher ε_r; stabilized with Zr/Al doping
Cubic (c)	>2600°C	~30	Via doping; key for FeFET
Amorphous (a)	Low-T ALD	~20	Memristor active phase

Production depends on post-ALD anneal. SIDRA gate oxide uses amorphous + controlled crystallization; memristor stays fully amorphous.

L2 · Full

Dielectric constant breakdown:

\varepsilon_r = \varepsilon_\infty + \varepsilon_{\text{ion}} + \varepsilon_{\text{dipol}}

Electronic ( $\varepsilon_\infty$ ): ~4 (similar across materials)
Ionic: Hf⁴⁺ and O²⁻ ions displace under field → ~16 in HfO₂
Dipole: trapped dipole moments in amorphous → ~5

Total ε_r ≈ 25. SiO₂ has only electronic + small ionic → 3.9.

EOT (Equivalent Oxide Thickness):

\text{EOT} = t_{\text{real}} \cdot \frac{\varepsilon_{\text{SiO}_2}}{\varepsilon_r}

In 28 nm HKMG actual HfO₂ thickness is ~7-8 nm, EOT ~1.1-1.2 nm. Tunneling scales with real $t$ → 6× thicker film = 10³-10⁴× less leakage.

Memristor vacancy density: $[V_O] \sim 10^{20}$ cm⁻³ typical — neighbor distance $a_0 = 0.3$ nm. A 1-10 nm diameter, 5 nm long filament needs 30-300 vacancies. Doping tunes density:

Pure: $[V_O] \sim 10^{19}$
Al (5%): $[V_O] \sim 5 \times 10^{20}$ (each Al³⁺ creates one V_O)
Y (3%): $[V_O] \sim 10^{20}$

L3 · Deep

Lanthanide contraction: Hf (Z=72) and Zr (Z=40) have nearly identical ionic radii (0.83 vs 0.84 Å). The f-orbital effect of the lanthanides makes Hf smaller than expected. Consequence: Zr and Hf are chemical twins, hard and expensive to separate. HfO₂ raw material contains 2-3% Zr — tolerable in gate-oxide design.

Ferroelectricity: pure HfO₂ is not ferroelectric; but thin films of Si- or Zr-doped HfO₂ (Hf₀.₅Zr₀.₅O₂ = HZO) settle in the orthorhombic phase and become ferroelectric. Discovery: Böscke et al., 2011 — the start of ferroelectric HfO₂ RAM. SIDRA Y100 FeFET candidate.

Band diagram:

Si: E_g = 1.12 eV
SiO₂: E_g = 9 eV (barrier ~3.1 eV above Si conduction band)
HfO₂: E_g = 5.7 eV (barrier ~1.5 eV above Si conduction band)

HfO₂’s barrier is lower than SiO₂’s; net leakage is still lower only because the film is thicker. This tradeoff is a key SIDRA design decision.

Experiment: Phase-Role Matching

Match these 5 scenarios on paper:

Scenario	Phase	Role	Reason
(1) 28 nm MOSFET gate oxide	Amorphous + controlled	Dielectric	Stable, low leakage
(2) SIDRA Y1 memristor cell	Amorphous	Resistive switching	Free vacancy motion
(3) Y10 Al-doped memristor	Amorphous + Al³⁺	Fast SET	More V_O
(4) Y100 FeFET (candidate)	Orthorhombic (HZO)	Non-volatile gate	Ferroelectric
(5) High-T power FET	Tetragonal + Y	High-T insulator	Crystal stability

Write your own sentence answering “why this phase for this role?” for each row.

Quiz

1/5HfO₂'s dielectric constant is about:

Lab Task

(a) For a 28 nm HKMG MOSFET with target EOT = 1.1 nm, what is the real HfO₂ thickness?

(b) What SiO₂ thickness would give the same C_ox? Tunneling ratio?

Answers

(a) t_HfO₂ = EOT · (ε_r/ε_SiO₂) = 1.1 · (25/3.9) = 7.05 nm.

(b) t_SiO₂ = EOT = 1.1 nm. HfO₂ is ~6.4× thicker. Tunneling goes like exp(−2κ·Δd). With V_0 ≈ 3.1 eV → κ ≈ 9×10⁹ /m. Δd = 5.95 nm. 2κΔd = 107. So HfO₂ leaks roughly 10⁻⁴⁷ × less (effectively zero; in real chips other mechanisms cap this math).

(c) HfO₂ molar mass ≈ 210 g/mol, density 10 g/cm³ → 2.87×10²² HfO₂/cm³. 5% Al substituting Hf → 1.4×10²¹ Al/cm³. Each Al³⁺ creates one V_O → ~1.4×10²¹ cm⁻³ extra V_O, ~10× the pure-HfO₂ background.

Cheat Sheet

HfO₂ — dual role: high-k gate dielectric + memristor active layer.
ε_r ≈ 25, E_g ≈ 5.7 eV. Stable interface with Si.
Phases: amorphous (memristor), tetragonal/cubic (high-k stability), orthorhombic HZO (ferroelectric).
EOT = t_real · (ε_SiO₂/ε_r); 28 nm HKMG ≈ 1.1-1.2 nm.
Doping: pure → slow/long, Al → fast/short, Y → slow/durable, Zr → ferroelectric HZO.
Intel 45 nm (2007) launched HKMG; SIDRA inherits that infrastructure.

Vision: Beyond HfO₂

HfO₂ will likely remain in CMOS for another 20 years; but parallel tracks:

HZO (Hf₀.₅Zr₀.₅O₂) FeFET: 10⁹ endurance, non-volatile. Micron + UMC 2024 demo. Strong Y100 candidate.
Antiferroelectric HfO₂ (Si-doped): double-stable state — high-density analog memory.
La₂O₃: ε_r ≈ 30, but hygroscopic; used in sealed packages (research).
SrTiO₃ (STO): ε_r ≈ 300 — extreme capacitance in cryogenic applications.
2D dielectrics (hBN, CaF₂): atomic thickness, pristine interfaces — gate dielectric for 2D transistors.
High-ε memristor: Ta₂O₅, TiO₂, ZrO₂ — different endurance/retention tradeoffs.
Polymer dielectrics: flexible electronics; PVDF (ferroelectric polymer) for biocompatible neural interfaces.

Key takeaway: HfO₂ isn’t “the one”, it’s “the best we have now”. A material shift arrives every 5-10 years. SIDRA’s architecture was designed to be material-agnostic: Y1/Y10 use HfO₂; Y100 can switch to HZO or a new candidate.

Biggest lever for post-Y10 SIDRA: an HZO transition — same ALD process, 50% Zr doping, orthorhombic phase stabilization. Retention > 10 years, endurance 10⁹, set voltage 1.5 V. 100× endurance over HfO₂ within the same BEOL budget. 2026–2028 horizon, Samsung and Micron racing in parallel.

Prerequisites

What you'll learn here

🪝 Hook: One Material, Two Lives

🧭 Intuition: Two Lives, One Lattice

📐 Formalism: Phases and Properties

🧪 Experiment: Phase-Role Matching

📝 Quiz

🛠️ Lab Task

🗂️ Cheat Sheet

🔮 Vision: Beyond HfO₂

📚 Further Reading