Guangdong Juyao Intelligent Technology Co., Ltd. · Laser Software & Process R&D
23°07′N 113°15′E · GUANGZHOU · PURE STATIC MONOGRAPH
← Back to news
Research Watch2026.11.06· 4 min read

Research Watch: Thin-film lithium niobate directly hosts all-optical memristors — 3-bit multilevel non-volatile optical memory at hundreds of femtojoules per bit

A photonic memristor is the memory cell of optical computing: light must be able to write a state and hold it non-volatilely before a neural network can truly run on light. The difficulty has always been fabrication — grafting memristive function onto a waveguide platform usually means heterogeneous integration of several materials, at high cost and low yield. The Shandong University team took a different approach: instead of assembling materials, modify the material itself.

JUYAO editorial visual: a micro-ring resonator waveguide on a thin-film lithium niobate chip glows softly, with the layered chip edge faintly visible
COVERJUYAO editorial visual (concept schematic, not a measured experimental image).
EXECUTIVE SUMMARY

On 6 November, Nature Photonics published research from Shandong University: using CMOS-compatible oxygen-ion implantation, spatially separated oxygen-doped and defect-rich layers are formed in thin-film lithium niobate, turning this widely used photonic platform itself into an all-optically controlled memristive material — photonic memristors can be fabricated directly on lithium niobate waveguides, with no heterogeneous integration. The devices demonstrate 3-bit multilevel non-volatile memory, switching energies of hundreds of femtojoules per bit and submicrosecond optical response. The full text is subscription-only; this item is compiled from the publicly available abstract and publication record.

01 / The idea: skip heterogeneous integration and turn the lithium niobate film itself into a memristive material

Photonic memristors are essential to photonic neural networks, but integrating them with waveguide platforms has traditionally relied on heterogeneous integration: stacking different materials layer by layer, with complex fabrication and high manufacturing costs. This paper’s route is homogeneous: thin-film lithium niobate (TFLN) is already one of the most widely adopted photonic platforms, and the team used a CMOS-compatible oxygen-ion implantation process to form two spatially separated layers inside the film — an oxygen-doped layer and a defect-rich layer — giving lithium niobate itself all-optically controlled refractive index modulation.

The team is from the State Key Laboratory of Crystal Materials, School of Physics, Shandong University (Xu Gao and colleagues); the paper was published in Nature Photonics on 6 November 2026. One caveat: the journal is subscription-based, so this item is based on the paper’s publicly available abstract and publication record; the full text was not read, and specific parameters beyond the abstract were not independently verified.

02 / The results: eight states in a micro-ring, hundreds of femtojoules, submicrosecond response

The device-level demonstration used an oxygen-implanted thin-film lithium niobate micro-ring resonator: the resonance transmission spectrum shifts with the optically controlled state, showing eight distinct non-volatile tuning states, equivalent to 3-bit memory capacity. Switching energy is on the order of hundreds of femtojoules per bit, optical response reaches the submicrosecond regime, and data retention remains stable across extended operation.

The three figures should be read together: multilevel storage answers “how much”, femtojoule-scale energy answers “at what cost”, and submicrosecond response answers “how fast” — only when all three hold can a photonic memristor enter a neural network. The abstract gives no exact retention time or endurance-cycle figures; those await the full text or follow-up work.

03 / Significance and limits: a wafer-scale, programmable photonic platform — but don’t order one yet

The authors stress a homogeneous, scalable and programmable platform: thin-film lithium niobate is already produced at wafer scale, and oxygen-ion implantation is CMOS-compatible — so all-optical memristors could plausibly scale along existing semiconductor lines rather than remaining laboratory samples of heterogeneous integration. The long-term target is large-scale photonic neural networks, where light on a chip can transmit, remember and compute.

The same measure applies: this is a validation of a material and device principle, and the abstract discloses no array scale, crosstalk, yield or system-level demonstration; from “eight states in a single micro-ring” to “a large-scale photonic neural network” is still a long road. Every figure in this item comes from the public abstract and a secondary report, and the subscription-only full text limits independent verification — read it as a direction worth tracking, not as a technology already in production.

PRIMARY REFERENCES & LINKS

  • All-optical thin-film lithium niobate memristor
    Nature Photonics · 6 November 2026 · Subscription-only, full text not open

    Abstract of the original paper verified: published 6 November 2026, DOI 10.1038/s41566-026-02048-1; CMOS-compatible oxygen-ion implantation, spatially separated oxygen-doped and defect-rich layers, all-optically controlled refractive index modulation, 3-bit multilevel non-volatile memory, hundreds of femtojoules per bit, submicrosecond response and long data retention all come from the paper’s publicly available abstract; the full text is subscription-only and details beyond the abstract were not independently verified.

    Open source link ↗
  • All-optical photonic memristors fabricated directly on thin-film lithium niobate
    ScienMag · 7 November 2026

    Used for cross-checking: eight distinct states in a micro-ring resonator corresponding to 3-bit memory, hundreds of femtojoules per bit, submicrosecond response and stable data retention across extended operation are all consistent with the paper’s abstract.

    Open source link ↗