🧪 Module 2 · Chemistry and Materials Science · Chapter 2.1 · 13 min read

The Chip Side of the Periodic Table

18 elements, one chip — why exactly these?

Prerequisites

What you'll learn here

  • Map key elements (Si, Hf, O, Ta, Ti, Cu, W, Nb) to their roles in SIDRA
  • Distinguish covalent / ionic / metallic bonds and which dominates where
  • Connect valence-electron count to a class of behaviour
  • List beyond-HfO₂ and beyond-Si future material alternatives

Hook: 118 Elements, 18 Choices

The periodic table has 118 elements. A modern chip uses fewer than 20 in any meaningful way. The other 100 are too rare, too weak, too radioactive, or chemically impossible. That selection is the revolution; if any element is in the wrong place the chip doesn’t work.

SIDRACHIP is no exception. Silicon for the 28 nm CMOS base die, hafnium oxide for the gate dielectric, tantalum nitride as top electrode, titanium nitride as bottom electrode, copper for interconnect, tungsten for refractory vias, niobium oxide for the OTS selector. Each is deliberate. This chapter unpacks the chemistry behind those decisions — and shows why the next generation might switch to entirely different elements.

Intuition: Atomic Number and Position

The periodic table arranges elements by valence count and energy level. That makes chemical behaviour predictable:

  • Far left (groups 1-2): one or two valence electrons, easy to lose → metallic, conductive (Na, K, Cu).
  • Far right (groups 17-18): 7-8 valence electrons, electron-greedy → nonmetal, insulating (F, O, noble gases).
  • Middle (transition metals, groups 3-12): partly filled d orbitals → interesting oxides (HfO₂, NbOx, TiO₂).
  • Group 14 (C, Si, Ge): 4 valence electrons, can give or take → semiconductor.

A chip designer reads the table with three questions:

  1. Could it be a substrate? Group 14 + Group 13/15 (GaAs, InP).
  2. Could it be an insulator? Transition metal + O (HfO₂, Al₂O₃, Ta₂O₅).
  3. Could it be a conductor? Group 1, 11, refractory transition (Cu, W, Al, Ti).

Formalism: Bonds and Periodic Trends

L1 · Intro

Three bond types, three behaviours:

BondHowExampleChip role
CovalentElectron sharingSi-Si, Si-OSemiconductor lattice, dielectric
IonicElectron transfer (+/−)Na⁺Cl⁻, Ti⁴⁺O₂²⁻Oxide dielectrics (partly)
MetallicElectron seaCu, W, AgConductive lines
L2 · Full

Periodic trends:

  • Atomic radius: decreases left-to-right (more nuclear pull), grows down a group (new shell).
  • Ionization energy: rises right, drops down. Cu (746 kJ/mol) doesn’t want to ionize → stays metallic. F (1681 kJ/mol) doesn’t ionize, takes electrons.
  • Electronegativity (Pauling): rises right. F (3.98), O (3.44), Cu (1.90), Si (1.90), Hf (1.30). ∣ΔEN∣|ΔEN| large → ionic; small → covalent. Hf-O has ΔEN = 2.14 → mostly ionic (the source of HfO₂’s high ε_r).

Why HfO₂? εr≈25ε_r ≈ 25 (SiO₂: 3.9), Eg≈5.7E_g ≈ 5.7 eV (sufficient insulation), thermodynamically stable on Si (interface SiO₂ layer is stable). Also CMOS-compatible (ALD-depositable). Few oxides hit this combination: ZrO₂ (cousin, ε_r ≈ 25, slightly smaller E_g), La₂O₃ (hygroscopic), Y₂O₃ (sustainability concern).

Why W (tungsten)? Melting point 3422°C. Electromigration resistance ~10× Cu. SIDRA uses it for critical vias. Long lines stay Cu (low resistance), vias use W (high durability) — a tradeoff.

Why Cu? Resistivity (1.68 µΩ·cm) below Al (2.65) and W (5.6). The CMOS interconnect standard since 1997. One catch: Cu diffuses into Si → needs barrier (TaN/TiN).

L3 · Deep

Lewis structure + d-orbital participation: transition metals (Hf, Ti, Ta, Nb) have partially filled d orbitals. In their oxides O 2p hybridizes with metal d → band hybridization. That is the physical root of HfO₂’s high dielectric constant and memristive behaviour.

SIDRA filament activation energy (EaE_a) depends on doping:

  • Pure HfO₂: Ea≈1.2E_a ≈ 1.2 eV (long retention, slow program)
  • HfO₂:Al (5%): Ea≈0.9E_a ≈ 0.9 eV (10× faster program, ~20% shorter retention)
  • HfO₂:Y (3%): Ea≈1.4E_a ≈ 1.4 eV (slow but durable)

SIDRA Y1 uses pure HfO₂ (retention-priority). Y10 uses Al-doped (program speed critical). It’s a chemistry tradeoff.

Common confusion: Hf (hafnium) vs Hg (mercury) — both start “H” but unrelated; Hg is toxic, Hf is not. SIDRA has no Hg. Hf, Zr (zirconium) are chemical twins (lanthanide contraction) — almost identical chemistry, hard to separate.

Experiment: SIDRA Periodic Table

Click on:

  1. Hf — high-k gate + memristor.
  2. O — vacancies form the memristor filament.
  3. Si — substrate.
  4. Cu — interconnect.
  5. W — refractory via.
  6. Nb — OTS selector.
  7. Al — HfO₂ doping.
  8. B / P — Si dopants (acceptor / donor).

Each element’s role is shown in the side panel.

Quiz

1/5What advantage does HfO₂ have over SiO₂ as gate dielectric?

Lab Task

(a) HfO₂ monoclinic unit cell: a=5.12a = 5.12 Å, b=5.17b = 5.17 Å, c=5.29c = 5.29 Å, β=99.2°\beta = 99.2°. Volume ≈ abc⋅sin⁡βabc \cdot \sin\beta. Compute.

(b) HfO₂ molar mass: Hf=178.49, O=16.00. M = ?

(c) For a 5 nm × 100 nm × 100 nm volume, how many HfO₂ formula units? (Density 9.7 g/cm³.)

Answers

(a) V = 5.12 · 5.17 · 5.29 · sin(99.2°) ≈ 140.0 · 0.987 ≈ 138 ų = 1.38×10⁻²² cm³.

(b) M = 178.49 + 32 = 210.49 g/mol.

(c) V_cell = 5×10⁻⁹ · 100×10⁻⁹ · 100×10⁻⁹ = 5×10⁻²³ m³ = 5×10⁻¹⁷ cm³. Mass = 5×10⁻¹⁷ · 9.7 = 4.85×10⁻¹⁶ g. Mol = 4.85×10⁻¹⁶ / 210.49 = 2.30×10⁻¹⁸. Formula units = 2.30×10⁻¹⁸ · 6.022×10²³ ≈ 1.39 × 10⁶. So one memristor cell holds ~1.4 million HfO₂ formula units (~4 million atoms). If 100 vacancies make a filament, only 0.007% of atoms move — explains the 10⁶ endurance ceiling.

Cheat Sheet

  • Fewer than 20 elements make up ~99% of a modern chip.
  • Si: substrate (4 valence, covalent, p/n via doping).
  • HfO₂: high-k dielectric + memristor. ε_r ≈ 25, E_g ≈ 5.7 eV.
  • Cu: main interconnect (1.68 µΩ·cm); barrier (TaN/TiN) required.
  • W: refractory via, electromigration-resistant.
  • Ta/Ti: TaN/TiN barriers + memristor electrodes.
  • Nb: NbOx OTS selector.
  • B (acceptor), P (donor): Si dopants.
  • Al, Y: HfO₂ doping → endurance/retention tuning.
  • Bond trio: covalent (Si-Si), ionic (Hf-O), metallic (Cu).

Vision: Beyond HfO₂, Beyond Si

SIDRA Y1/Y10 sit on the current material palette. Y100 and post-Y100 research watch these alternatives:

Memristor material alternatives:

  • HZO (Hf₀.₅Zr₀.₅O₂) — ferroelectric HfO₂ variant; not just resistance, polarization carries memory. 10⁹ endurance target.
  • PCM (GST: Ge₂Sb₂Te₅) — phase change; amorphous↔crystalline. Basis of Intel/Samsung Optane; faster (~10 ns) but hotter.
  • MTJ (Magnetic Tunnel Junction) — spintronics; CoFeB/MgO stack. Angle between ferromagnetic layers carries data. STT-MRAM is already commercial; durable, infinite endurance, lower density.
  • Organic / molecular memristors — bio-compatible, flexible electronics. Still research.

Substrate alternatives (post-Si):

  • GaN, SiC — wide-bandgap (E_g 3.3-3.4 eV); high-voltage / high-T power electronics. Tesla EV inverters already use SiC.
  • 2D materials (MoS₂, WSe₂, hBN) — atom-thick transistors. Sub-2 nm channel possible. IBM 2024 demo.
  • Diamond (C) — extreme thermal conductivity; experimental transistor.
  • Graphene — high mobility but no bandgap → hybrid solutions (graphene-on-hBN heterostructures).

Interconnect alternatives (post-Cu):

  • Carbon nanotubes (CNT) — 1000× higher current density than copper; integration unsolved.
  • Co (cobalt) — below 7 nm, Cu’s resistivity blows up; Co is the new barrierless via.
  • Ru (ruthenium) — leading candidate for 3 nm node.

Beyond-architecture vision:

  • Photonic-CMOS hybrids — Y100 target; Si photonic waveguides + CuM memristors. Optical MVM in, analog compute, optical out.
  • Cryogenic memristors for quantum — 4 K memristor synapses; not AI accelerator but quantum-algorithm interface. Research only.
  • DNA / biological storage — petabit/cm³ density; very slow (hours) but enormous retention (millennia). Outside SIDRA’s focus, but post-Y100 academic interest.

Key takeaway: today’s chip chemistry is rewritable from the bottom up. The SIDRA architecture you’ll see in Module 5 is material-agnostic — the same crossbar geometry could be built tomorrow with HZO or CNT. This shows SIDRA is an architectural choice, not a chemical one.

Biggest lever for post-Y10 SIDRA: the triple of HZO ferroelectric memristor + Ru interconnect + 2D-material channel. HZO pushes endurance to 10⁹; Ru improves resistance/current-density by 3×; MoS₂ channels raise density by 2×. Together: 10-15× more performance in the same footprint. 2028–2032 horizon.

Further Reading