🧪 Module 2 · Chemistry and Materials Science · Chapter 2.3 · 11 min read

NbOx — The Chemistry of OTS

The selector that solves crossbar sneak-path.

Prerequisites

What you'll learn here

  • Explain how a NbOx-based OTS (Ovonic Threshold Switch) works
  • Introduce the Mott insulator-to-metal transition
  • Describe the sneak-path problem in a crossbar and how OTS solves it
  • Summarize NbOx vs VO₂ Mott material comparison

Hook: The Crossbar's Headache

A 256×256 memristor crossbar. You want to read one cell — say row 5, column 12. Expected: that cell’s current. Reality: every column’s currents mix. Because neighbours in low-resistance state leak current through “side paths”. This is sneak path, the bane of analog MVM.

Fix: put a selector next to each memristor. High resistance at low voltage (closed), sudden low resistance above threshold (open). Diode-like behaviour, but CMOS-compatible and symmetric. SIDRA picked NbOx — niobium-oxide OTS.

Intuition: The Mott Transition

NbOx has two phases:

  • NbO₂ (stoichiometric): Mott insulator — electrons localized, no conduction.
  • NbO₂ under heat / field: jumps to the metal phase — electrons free.

The transition is driven by Joule heat: current → heat → transition temp (~1080 K) → Mott metal. Remove the current, it cools back to insulator. Thus volatile — only open while voltage is on.

Exactly what we want: active during reads, closed otherwise. Doesn’t touch the memristor’s persistent state, only gates the read path.

Formalism: Threshold and Transition

L1 · Intro

OTS I-V:

  • V<VthV < V_{th}: OFF — high R (MΩ).
  • V=VthV = V_{th} (~1.3 V): snap-back to ON.
  • Vth>V>VholdV_{th} > V > V_{hold} (~0.5 V): ON — low R (kΩ).
  • V<VholdV < V_{hold}: returns to OFF.

This hysteresis blocks sneak paths: weak leak currents can’t exceed VthV_{th}; the selector stays closed.

L2 · Full

Mott transition: when electron-electron Coulomb repulsion UU exceeds bandwidth WW, electrons localize (Mott insulator). Temperature or field increases WW → metal. NbO₂:

  • Transition TM≈1080T_M \approx 1080 K
  • Field threshold ∼107\sim 10^7 V/cm
  • Transition time ∼10\sim 10 ns (fast enough for SIDRA reads)

1S1R cell: memristor (R) stacked under an OTS (S). SIDRA uses NbOx 5 nm + TiN electrodes. ~100 nm² per cell.

Sneak-path math: in 256×256 without selectors, a read is buried ~90% under leak currents. Selectors boost signal/noise ~100×.

L3 · Deep

Ovonic: from Stanford Ovshinsky, who discovered threshold switching in amorphous semiconductors (1960s). Same family as PCM (GST).

NbOx stoichiometry: x = 2 (NbO₂) Mott, x = 2.5 (Nb₂O₅) ordinary insulator, x < 2 metal. SIDRA targets x ≈ 2 via ALD oxygen partial pressure.

Electro-thermal model:

T(t)=Tamb+P(t)⋅Rth,local,P=I2RNbOxT(t) = T_{amb} + P(t) \cdot R_{th,local}, \quad P = I^2 R_{NbOx}

Once TMT_M is reached R drops, current rises, more heat — positive feedback. Cooling rate is set by local RthR_{th} — sub-ns.

Experiment: 1S1R Logic

Sketch a 2×2 crossbar:

         col-0   col-1
row-0   [1S1R]  [1S1R]
row-1   [1S1R]  [1S1R]

Scenario: read row-0, col-0. Apply +0.1 V to row-0, 0 V to col-0, +0.05 V to other lines (half-selected).

  1. Selected cell (0,0): V=0.1V = 0.1 V — if Vth>0.1V_{th} > 0.1 the OTS is closed; in SIDRA the selector stays pre-primed, memristor R dominates.
  2. Neighbour (0,1): V=0.05V = 0.05 V — OTS closed, leak negligible.
  3. Far (1,1): V=0V = 0 — no current.

The selector silences half-selected cells, making single-cell reads possible. Without selectors, 256 cells’ currents mix.

Quiz

1/5What material is SIDRA's OTS selector?

Lab Task

A NbOx OTS cell: Vth=1.3V_{th} = 1.3 V, ROFF=10R_{OFF} = 10 MΩ, RON=10R_{ON} = 10 kΩ.

(a) OFF current at V=0.1V = 0.1 V? (b) ON current at V=1.5V = 1.5 V? (c) ON/OFF ratio selectivity. In a crossbar with 256 half-selected cells, how much sneak-path suppression?

Answers

(a) I_OFF = 0.1 / 10⁷ = 10 nA. (b) I_ON = 1.5 / 10⁴ = 150 µA. (c) Ratio = 15,000. 256 half-selected cells leaking = 256 · 10 nA = 2.56 µA. Selected ON = 150 µA → SNR = 60×. Without the selector, 256 memristor HRS currents (~0.1 µA each) swamp the signal → SNR ≈ 0.4 (read fails).

Cheat Sheet

  • NbOx / NbO₂ — Mott insulator; Joule drives metal transition.
  • OTS — Ovonic Threshold Switch. Volatile, diode-like threshold.
  • Vth≈1.3V_{th} \approx 1.3 V, Vhold≈0.5V_{hold} \approx 0.5 V — hysteresis kills sneak path.
  • 1S1R — one selector per memristor. SIDRA crossbar cell.
  • Mott T ≈ 1080 K. Fast (~10 ns), volatile (closes on current off).

Vision: Beyond OTS

  • VO₂ Mott selector: transition T = 340 K (near room T!) → far lower power. Active research, 2024 IEDM.
  • TaO / HfO OTS: same family, different thresholds — combinations for 3D stacks.
  • Mixed ionic-electronic (MIEC): Ag-based OTS; faster.
  • Ferroelectric selector: HZO-based — natural hysteresis; no extra selector needed.
  • Schottky diode selector: simple but unidirectional; incompatible with bipolar memristors.
  • 2D-material selector: MoS₂-based — atomic-thickness for ultra-dense crossbars.
  • Quantum-tunnel selector: RTD-based — sharp NDR threshold, sub-ns switching.
  • Selector-less architecture: bipolar memristor + smart sense-amp; ~40% area savings per cell.

Biggest lever for post-Y10 SIDRA: the VO₂ Mott selector — transition at 340 K (near room T). Joule budget 10× lower, SET/RESET energy 5× lower, crossbar size can grow from 256 → 512. 2027–2029 horizon.

Further Reading