Demonstration of on-chip all-optical switching of magnetization in integrated photonics

The gist: This work demonstrates single-pulse all-optical switching (AOS) of magnetization within a silicon nitride (Si3N4) photonic integrated circuit, establishing a foundation for ultrafast, energy-efficient, and fully integrated spintronic-photonic platforms.

Background and Motivation

Ultrafast all-optical magnetization switching (AOS) holds great promise for nextgeneration spintronic memory and hybrid spintronic-photonic systems However, most implementations to date rely on bulky free-space optical setups, limiting scalability and practical integration Singlepulse all-optical switching (AOS) of magnetization has emerged as a promising route toward the fastest and one of the most energyefficient magnetization reversal Optical approaches also allow for direct magnetization readout, such as through the magneto-optical Kerr effect (MOKE) Recent advances in AOS have significantly deepened our understanding of the fundamental mechanisms and enabled optimization of key performance metrics Integration of optically switchable magnetic layers into functional spintronic architectures has already been demonstrated in several works, including magnetoresistive random-access memory (MRAM).

Device Architecture and Fabrication

The platform is based on silicon nitride (Si3N4) waveguides fabricated using the LioniX TriPleX technology Trains of femtosecond laser pulses are coupled into the Si3N4 waveguides via on-chip edge couplers, enabling robust all-optical toggle switching of magnetization in an out-of-plane Co/Gd Hall cross patterned directly atop the photonic waveguide. The spintronic material stack used is Ta(2)/Pt(2)/Co(1)/Gd(3)/TaN(5)/Pt(2), which is selected for its relevance in state-of-the-art spintronics. To ensure sufficient light–matter interaction between the guided optical mode and the spintronic Hall cross, a few-microns-thick protective SiO2 cladding layer was selectively etched from the top surface and sidewalls in regions where Hall crosses were fabricated, creating sensing windows. The final integrated Hall cross is positioned atop both the Si3N4 waveguide and stage, which is shown in Fig. 1d.

Experimental Results and Switching Behavior

The symmetric Hall cross device with an arm width of 500 nm showed a clear toggling of the anomalous Hall effect (AHE) signal synchronous with the incoming pulses, evidenced by robust, all-optical toggle switching. Repeating the procedure on a device with an 1100 × 700 nm2 asymmetric arm width (the 1 µm device) yielded a similar toggle-switching pattern but with a substantially reduced signal amplitude, from 90% in the 500 nm device to no more than 30% in the 1 µm case. Finite element simulations attribute this trend to non-uniform light absorption across the device, in combination with a hypothesis including domain wall relaxation and thermally assisted (de)pinning processes.

Mechanisms of Stochasticity

The transition from deterministic to stochastic switching is linked to both device geometry and the complex spatial profile of optical energy deposition. For the 1 µm Hall cross, when pulse energy is reduced to 0.81E0, stochastic switching begins to emerge, while at 0.67E0, the switching becomes increasingly stochastic. This stochasticity is attributed to the formation of meta-stable magnetic domain patterns resulting from a combination of toggle switching, domain wall (DW) relaxation, stochastic pinning and thermally assisted depinning events. At pulse energies near the switching threshold, the DW can become pinned at sharp corners of the Hall cross or trapped by local magnetic inhomogeneities.

Theoretical Modeling and Analysis

Numerical simulations using Ansys Lumerical Finite-Difference Time-Domain (FDTD) and Finite-Difference Eigenmode (FDE) solvers were performed to evaluate the light absorption profile within the Hall cross. The occurrence of AOS in each cell was assessed with a simplified M3TM model, using relative differences in absorbed pulse energy from the FDTD output as input parameters. The microscopic three-temperature model (M3TM) captures the essential energy exchange among electrons, phonons, and spins following ultrafast excitation. AOS occurs when the peak electron temperature (Te,max) exceeds the Curie temperature of cobalt (TC,Co = 1388 K) while the lattice temperature remains below it.

Conclusion and Future Directions

Device miniaturization, specifically scaling below the characteristic domain wall (DW) width, largely suppresses stochasticity observed in larger devices. The results demonstrate the feasibility of on-chip AOS but also suggest that stochastic switching could be exploited as a resource for certain probabilistic computing or neuromorphic applications. Future implementations could benefit from improved coupling strategies, such as fiber-based approaches or the integration of on-chip pulsed laser sources. The work establishes a foundation for scalable, ultrafast spintronic–photonic hybrid systems.

Methods Overview

The integrated photonic platform was based on a silicon nitride (Si3N4) waveguide system fabricated using low-pressure chemical vapor deposition (LPCVD). Inverse-tapered edge couplers were integrated at the chip periphery to enable efficient free-space coupling. The spintronic Hall cross was patterned by e-beam lithography and a lift-off process, and deposited directly onto the Si3N4 waveguide stage within the sensing window.

Theoretical Modeling Details

The local absorption density was calculated from the divergence of the Poynting vector as Pabs = 1/2 ω E2 Im (ε), (B1). The model was reduced to a twotemperature model (2TM) description, sufficient to capture the coupled electron–phonon temperature dynamics. AOS occurs when Te,max > 2020 K > Tp,max for the studied multilayer.

Experimental Setup

Free-space, femtosecond laser pulses (720 nm wavelength) were focused onto the edge coupler using a highnumerical-aperture objective. A current was supplied to one arm of the Hall cross and the anomalous Hall voltage was recorded along the orthogonal arm using a lock-in amplifier to allow precise electrical readout. The coupling efficiency was low due to practical limitations and suboptimal alignment.

Contribution

P. Li conceived and designed the project, P. Li, G.W.A. Simons, and T. Zhang developed the experimental protocol, performed simulations, and collected data. G.W.A. Simons, P.P.J., Schrinner, and S Kamyar were responsible for the fabrication of the devices. R Dekker provided technical supervision of the photonic circuit, D C Leitao supervised the post-processing fabrication, and R Lavrijsen and Y Jiao provided feedback on the experimental work. P Li and B Koopmans supervised the project, P Li wrote the original manuscript draft.

Competing Interests

The authors declare no conflicts of interest.

References Cited

[1] B Dieny, I L Prejbeanu, K Garello, P Gambardella, P Freitas, R Lehndorff, W Raberg, U Ebels, S O Demokritov, J Akerman, A Deac, P Pirro, C Adelmann et al., “Opportunities and challenges for spintronics in the microelectronics industry,” Nat. Electron., vol. 3 no. 8 pp.

1. Bold header: Improved Spintronic Memory Write Mechanism

The improved AI system can utilize single-pulse all-optical switching (AOS) to write binary data with up to 90% contrast for 500 nm-wide devices, offering a switching speed improvements by up to three orders of magnitude compared to conventional spintronic writing approaches.

2. Bold header: Stochastic Switching Analysis and Mitigation

The improved AI system can predict device performance under non-ideal conditions by modeling the transition from deterministic to stochastic switching, linking it to non-uniform light absorption across the device, in combination with a hypothesis including domain wall relaxation and thermally assisted (de)pinning processes.

3. Bold header: Optimized Device Geometry Design

The AI system can design Hall cross geometries that maximize switching contrast by understanding the critical role of device scaling in achieving robust on-chip AOS, specifically targeting dimensions below the characteristic DW width to suppress stochasticity.

Ultrafast all-optical magnetization switching (AOS) holds great promise for nextgeneration spintronic memory and hybrid spintronic-photonic systems. However, most implementations to date rely on bulky free-space optical setups, limiting scalability and practical integration. As a critical step toward integrated applications, we demonstrate single-pulse AOS within a silicon nitride (Si3N4) photonic integrated circuit. Using trains of femtosecond laser pulses guided through on-chip waveguides, we achieve deterministic toggle switching in a sub-micron out-of-plane Co/Gd Hall cross patterned directly atop the photonic waveguide. Electrical readout via the anomalous Hall effect reveals a switching contrast of up to 90% for 500 nm-wide devices. In larger Hall crosses, the contrast decreases and switching becomes stochastic, consistent with spatially non-uniform optical absorption as confirmed by finite-element simulations. This behavior is hypothetically attributed to domain wall relaxation and thermally assisted (de)pinning processes within partially switched regions. Our results highlight the critical role of device scaling in achieving robust on-chip AOS and establish a foundation for ultrafast, energy-efficient, and fully integrated spintronic-photonic platforms.