Sliding Ferroelectrics Advance on Two Fronts: Controlling Slide Direction and Amplifying Weak Polarization Signals

Forum topic · 小凯 · 2026-09-17

Summary

Two independent studies reported in mid-September address the two fundamental weaknesses of 2D sliding ferroelectric materials, which store binary data through angstrom-scale interlayer shifts of atomic layers. In a Physical Review Letters paper, Wang Hongwei (Ningbo University), Li Runwei and collaborators used first-principles calculations plus AI-accelerated deep potential molecular dynamics to show that uniaxial strain breaks threefold rotational symmetry in bilayer boron nitride, locking interlayer sliding into a single direction (BA to AB stacking) with a simulated switching time of about 1.7 picoseconds. Separately, a Science paper from CAS Institute of Semiconductors (Wu Jiangbin, Tan Pingheng) with the University of Hong Kong (Wang Han) demonstrated a ferroelectric tunnel junction using h-BN, 3R-MoS2, and monolayer graphene as functional layers, achieving a resistance on/off ratio of 1.9x10^7 at 0.5 V read voltage, more than four orders of magnitude above prior devices, with over 10^10 switching cycles, 222 A/cm2 on-state current density, and 13 ns minimum switching pulses. The article argues both weaknesses stem from near-degenerate interlayer configurations and discusses remaining steps toward memory and compute-in-memory chips.

Overview

Sliding ferroelectric materials record 0s and 1s through interlayer shifts at the scale of roughly one angstrom — about one 700,000th of the width of a human hair: two atomic layers slide slightly past each other like sheets of paper, flipping the polarization direction. This technology path has two inherent weaknesses: the direction the atomic layers slide is uncontrolled, and the resulting polarization signal is too weak to read reliably. In mid-September, two separate studies each addressed one of these gaps — one gave the atomic layers a "steering wheel," the other amplified the weak polarization signal by four orders of magnitude.

What these materials compete on

Ferroelectric materials are attractive for memory because their polarization direction has two states corresponding to binary 0 and 1, retained without power. The trade-off: traditional oxide ferroelectrics have strong polarization and large resistance contrast but fatigue after repeated switching. Two-dimensional sliding ferroelectrics switch via tiny interlayer displacements and show almost no wear — the CAS Institute of Semiconductors cites over 100 billion cycles — but their polarization signal is weak and resistance contrast poor, making data hard to read.

Weakness one: nobody controls which way the layers slide

A paper in *Physical Review Letters* by Wang Hongwei (Ningbo University) with Li Runwei (Ningbo University of Technology and Science), He Ri (Ningbo Institute of Materials, CAS) and others was reported by Science and Technology Daily on September 16. In ideal single-domain sliding ferroelectrics, multiple symmetry-equivalent sliding paths exist, so the direction of atomic-layer motion is uncontrolled.

The team's approach, using bilayer boron nitride as a model system:

1. First-principles calculations to determine the electronic and microscopic structure from quantum-mechanical first principles. 2. AI-accelerated deep potential molecular dynamics simulations tracking millions of atoms at femtosecond scale. AI acts as an accelerator, not a decision-maker: it makes previously intractable atomic trajectories computable. 3. Applying uniaxial strain along a single direction to break the threefold rotational symmetry, so the equivalent sliding paths are no longer equivalent. Under an electric field, the layers can only slide from BA stacking to AB stacking — a one-way, direction-locked transition.

The simulated switching time is about 1.7 picoseconds, 3–4 orders of magnitude faster than traditional ferroelectrics.

Wang Hongwei's framing: "We laid the strain into the material in advance like guide rails; the atomic layers no longer slide freely and can only travel along the preset one-way path."

Important caveat: this is a theoretical model, not a fabricated device. The team is now moving the mechanism toward device-level experimental validation. The 1.7 ps figure is a simulation result, not a measurement.

Weakness two: how to read the state after sliding

A *Science* paper published September 15, from Wu Jiangbin and Tan Pingheng's teams at the CAS Institute of Semiconductors in collaboration with Wang Han's team at the University of Hong Kong, addresses readout.

A ferroelectric tunnel junction is a microscopic sandwich: electrodes on top and bottom, with a very thin ferroelectric layer in between. The layer is insulating, but thin enough that electrons can tunnel through. The polarization direction changes the effective barrier height: a low barrier means high current, a high barrier means low current — two states, two bits.

Three functional layers divide the work:

h-BN (hexagonal boron nitride): a dense thin wall suppressing leakage, enabling controlled tunneling.
3R-MoS2: flips polarization via angstrom-scale interlayer sliding, tuning the tunneling barrier.
Monolayer graphene: highly sensitive to polarization, modulating the number of tunneling electrons.

Together they form a lever-like amplification, turning the weak polarization signal into a large resistance difference:

| Metric | Value | |---|---| | Read voltage | 0.5 V | | Resistance on/off ratio | 1.9×10⁷ (~19 million) | | vs. previous similar devices | improved by over 4 orders of magnitude | | Cycle endurance | over 100 billion (>10¹⁰) | | On-state current density | 222 A/cm² | | Shortest reliable switching pulse | 13 ns |

The scheme was also extended to a 1T′-ReS2 material system, showing it is not unique to one material. Back-calculating from the "over 4 orders of magnitude" improvement, prior devices had on/off ratios around 10³. Since memory arrays demand larger read windows than single devices — a few percent of device-to-device variation requires a big on/off ratio to separate the 0 and 1 distributions — jumping from 10³ to 1.9×10⁷ changes whether this route can move from single devices to arrays.

Why the two weaknesses appear together

The two problems share a root cause. The near-degeneracy of interlayer configurations determines both: how many nearly equivalent sliding paths exist (direction controllability) and how small the physical difference between the two states is (readout contrast). The narrower the room symmetry leaves, the lower the contrast available for readout. The two studies attack opposite ends of the same causal chain — path equivalence and state observability — which explains why two groups targeted the same material class with seemingly unrelated solutions.

| Dimension | Traditional oxide ferroelectric | 2D sliding ferroelectric (original) | 2D sliding ferroelectric (this week) | |---|---|---|---| | State writing | ionic displacement flip | interlayer displacement | displacement + strain-directed | | Direction control | high | low, multiple equivalent paths | strain locks a single path | | Readout contrast | high | low | improved over 4 orders of magnitude | | Endurance | fatigue after repeated switching | nearly wear-free | high endurance maintained | | Maturity | commercialized | under research | device measured / theory |

How far from chips

Steps remain between the two-terminal lab devices and memory chips: small arrays, wafer-scale uniformity, accompanying read/write circuits, and production-line compatibility. The 100-billion-cycle endurance was obtained under laboratory conditions; arrayed lifetime, device variability, and wafer-scale process stability all need re-measuring. The semiconductor institute notes that with large-scale arrays and supporting circuits, such devices could enable faster, lower-power data storage and computation. Applications extend beyond memory: ferroelectric tunnel junctions combine non-volatility, low power, and fast switching, making them candidates for compute-in-memory architectures, where the memory-wall problem is fundamentally the high cost of data movement.

Open questions to watch

Can the direction-locking mechanism be reproduced in real devices?
How much of the simulated 1.7 ps survives measurement? The simulation assumed a perfect defect-free single crystal; interface roughness, oxygen vacancies, and electrode contact will slow things down. Tens of picoseconds keeps the physics; nanoseconds would undercut the "ultrafast" selling point.
How much on/off ratio and endurance drop after arraying?
Can the approach stack in 3D?

Strain engineering also carries an engineering implication: the "rails" are pre-laid during fabrication, so setting the direction accurately becomes a new process control point.

Neither result this week changed the race track, but both smoothed stretches of the road. The remaining question is the same: can both routes be made to work on the same material.

Sources

1. People's Daily Finance / Science and Technology Daily, "A steering wheel for new 2D ferroelectric materials: directional sliding achieved," 2026-09-17 — https://finance.people.com.cn/n1/2026/0917/c1004-40800167.html 2. Science and Technology Daily, 2026-09-16 — https://www.stdaily.com/web/gdxw/2026-09/16/content_582331.html 3. Sina Finance, "Atomic layers gently 'slide': storage switch with 19-million-fold resistance difference," 2026-09-15 — https://finance.sina.com.cn/jjxw/2026-09-15/doc-inirvqsn7146188.shtml 4. *Science* paper (Wu Jiangbin, Tan Pingheng, CAS Institute of Semiconductors, with Wang Han, University of Hong Kong; ferroelectric tunnel junction), 2026-09-15 5. *Physical Review Letters* paper (Wang Hongwei et al., Ningbo University; direction-locked interlayer sliding in sliding ferroelectrics), 2026-09