Ion trap on borosilicate substrate with integrated femtosecond-laser-written waveguide

Abstract: We present an ion-trap platform on borosilicate glass with an integrated femtosecond-laser-written waveguide for on-chip light delivery. The optical layer is physically separated from the electrode substrate and bonded atop the trap, remaining compatible with silicon-based integration. We engineer single-mode low-loss guidance at 729 nm with tunable mode-field diameter and achieve low-loss curved waveguides down to a radius of curvature of 6 mm. We also extend single-mode operation to a wavelength of 405 nm. The fabrication process is compatible with the industrial fabrication of a single-metal-layer surface-electrode trap, including active fiber alignment and bonding. We validate the platform in a cryogenic trapped-ion system with $<sup>{40}$Ca$<sup>+$, demonstrating trapping, shuttling the ion to a zone in front of the waveguide, and coherent operations driven by 729 nm light delivered through the integrated waveguide. We characterize the effect of the exposed dielectric on the ion and measure stray electric fields that show slow drift at a timescale of hours. The architecture is compatible with hybrid micro-optics (e.g. pick-and-place lenses) to realize single ion addressing and provides a robust, scalable route to integrated light delivery for trapped-ion devices.

• The paper demonstrates a surface electrode ion trap on borosilicate glass with an integrated femtosecond-laser-written waveguide for efficient qubit control of 729 nm light for trapped $^{40}$Ca$^+$ ions; the waveguides are inscribed at 170 µm height and emit light in-plane at the trapping position, with single-metal-layer surface trap that uses aluminum for patterned segmented electrodes
• Ion trap uses symmetric RF rails and segmented DC electrodes; the fabricated trap achieves an RF amplitude below 140 V and DC voltage less than 40 V, whereas radial secular frequencies around 2-3 MHz
• Demonstrates coherent Rabi oscillations driven by light through the integrated optical path: guided modes had sub-dB/cm loss at bend radii down to 6 mm and a mode field diameter ranging from 3 to 8 µm connected to reliable single-ion qubits and trapping operations

Overview and motivation

This paper demonstrates a surface-electrode ion trap fabricated on a borosilicate glass substrate with a femtosecond-laser-written (FLW) waveguide integrated for on-chip delivery of 729 nm qubit light to trapped $^{40}$Ca$^+$ ions (2608.13207). The central architectural idea is the physical separation of the optical layer from the electrode substrate: a structured borosilicate optics block is anodically bonded on top of the trap surface, with FLW waveguides inscribed at the ion height of $170~\mu$m so that light is emitted in-plane at the trapping position. This eliminates the out-of-plane coupling elements—grating couplers or microfabricated mirrors—that dominate existing planar photonic approaches beneath the electrodes, which require tens-of-nanometers lithographic tolerances and consume chip area for multi-wavelength addressing.

The choice of borosilicate glass is motivated by its coefficient of thermal expansion matched to silicon, enabling wafer-level anodic bonding and future integration with silicon ASICs via through-substrate vias. The fabrication was performed at Infineon Technologies Austria AG on 200 mm wafers using a single metal layer, with active fiber alignment and UV-cure bonding as part of the assembly flow. The work validates the full platform end-to-end in a cryogenic system: trapping, shuttling into an addressing zone 1 mm in front of the waveguide facet, and coherent carrier Rabi oscillations driven by light delivered entirely through the integrated optical path.

Waveguide engineering in borosilicate glass

A notable material-physics finding is that guiding in this borosilicate (SCHOTT BF33) occurs not within the laser-modified track but in spatially displaced regions above and/or below it, separated by roughly $10~\mu$m. The authors attribute this to thermally driven material redistribution away from the elongated focal volume, producing densified cores outside the modification region—an extension of the mechanism reported by Tan et al. in Eagle XG glass, where only a single displaced core was observed. The appearance of two vertically separated guiding regions is specific to this material and writing regime.

The authors exploit spherical aberration as a deliberate process parameter. Using an objective with adjustable correction collar, they define a focus offset between writing depth and aberration-free correction depth and find that performance degrades sharply near zero offset, with optimal transmission and confinement at moderate negative offsets of $170$–$240~\mu$m. They concede that their qualitative model does not explain why the degradation remains so localized around zero offset—a point where the formation mechanism remains incompletely understood.

Mode-field diameter (MFD) tuning is achieved through two levers: pulse energy and lateral overscan offset $\Delta x$. The MFD is tunable from approximately $3$ to $8~\mu$m. Counterintuitively, the MFD decreases non-monotonically with $\Delta x$, reaching minima near $^+$0m at offsets of $^+$1–$^+$2m; the authors interpret this consistently with weak-guidance theory, where confinement is best when the effective guiding region matches the intrinsic modal extent supported by the index contrast. These MFD values are smaller than typically reported for FLW waveguides in borosilicates.

Two performance consequences follow directly from this tunability:

• Far-field divergence: divergence ranges from 84–103 mrad for the smallest modes ($^+$3–$^+$4m) down to 40–60 mrad for MFDs near $^+$5m, saturating beyond that point as the mode profile deviates from Gaussian due to the surrounding modified index landscape.
• Bending loss: strongly confined modes tolerate tight curvature. Waveguides with $^+$6m MFD maintain losses below $^+$7 down to a bending radius of approximately $^+$8, whereas $^+$9m MFD structures begin losing at around $170~\mu$0. The authors state these radii are smaller than commonly reported for single-mode FLW waveguides in glass, which typically require tens of millimeters.

The writing strategy also extends single-mode operation to 405 nm—the wavelength relevant for Doppler cooling, state preparation, and fluorescence detection—with stable single-mode profiles verified under fiber displacement. This is significant because short wavelengths normally push FLW waveguides multimode unless dimensions are very small; the strong confinement here makes blue-violet guidance feasible without specialized multi-pass morphologies.

Trap design, fabrication, and dielectric effects

The trap uses a symmetric RF rail pair ($170~\mu$1m) flanking a central DC electrode ($170~\mu$2m), with segmented outer DC electrodes ($170~\mu$3m) and uniform $170~\mu$4m gaps, all patterned in a single $170~\mu$5m aluminum layer via sputtering, optical lithography, and reactive-ion etch. The geometry was optimized against competing constraints: wider RF rails reduce required RF amplitude (minimum $170~\mu$6 V at $170~\mu$7m width) but inflate DC voltages; the chosen $170~\mu$8m keeps RF below 140 V and DC well under the 40 V hardware limit.

FEM simulations quantify the perturbation of the bonded dielectric: at the adopted $170~\mu$9 mm glass-to-ion distance, the RF null shifts by only $10~\mu$0m vertically and $10~\mu$1m radially. The fabrication flow separates optics-wafer processing (waveguide inscription and ablation trenches at Joanneum Research) from trap-wafer metallization at Infineon, followed by anodic bonding, dicing, and fiber attach. Two functional devices were produced, with total waveguide transmission (including reflection losses) of $10~\mu$2 (device A) and $10~\mu$3 (device B).

A clear limitation acknowledged by the authors is the outcoupling facet quality. Facets are formed by manual cleaving initiated at laser-ablation trenches, a low-repeatability process yielding slanted or curved surfaces. Profilometry shows height deviations up to $10~\mu$4m over $10~\mu$5m along one axis, though local tilt within $10~\mu$6m of the waveguide stays below 1°. Device B's facet angle bound ($10~\mu$7) is worse than device A's ($10~\mu$8), correlating with poorer beam alignment at the ion height. At the ion position 1 mm from the facet, the beam has diverged to radii of $10~\mu$9m and $170$0m in the two transverse directions—so without hybrid lenses, the integrated beam functions as a global rather than individually addressing beam.

Trapping performance and stray fields

Both experimental systems operate at $170$1 K in closed-cycle cryostats with RF amplitudes of 110–160 V at 31.7 MHz, yielding radial secular frequencies of 2–3 MHz. Axial heating rates measured via resolved sideband spectroscopy lie between 0.5 and 1.5 phonons/s at 1 MHz axial frequency, with a power-law exponent of $170$2 versus axial frequency—consistent with technical noise on the DC electrodes filtered by first-order RC filters inside the cryostat, rather than anomalous surface-related heating. No influence of the dielectric optics block on heating was measured, consistent with the model of Teller et al. predicting negligible dielectric heating at 1 mm distance. It should be noted that this conclusion rests on the millimeter-scale standoff; the paper does not demonstrate tolerance at closer proximity.

Stray electric fields from the exposed glass were characterized spatially and temporally. A FEM model with uniform surface charge density fitted to measurements yields an inferred charge density of $170$3, with good agreement between the measured spatial profile and the scaled simulation—indicating the position-dependent stray field is dominated by the glass block. Over a 9-hour measurement, the radial stray field drifted at only $170$4, and ion reloading events did not produce fast stray-field changes, suggesting neutral flux and photoionization lasers are not dominant short-term charging sources. The practical implication stated by the authors is that micromotion compensation recalibration is needed only every few hours. The measurement method itself achieves a resolution of $170$5 by modulating axial curvature and monitoring ion position.

Integrated qubit operations

Qubit light at 729 nm was routed from a room-temperature diode laser through a vacuum fiber feedthrough, through the 40 K and 4 K radiation shields, into the chip-bonded fiber, and out of the integrated waveguide onto the ion. With the ion shuttled to positions in front of the waveguide in 20 $170$6m steps, the authors measured position-dependent Rabi frequencies on the $170$7 carrier transition, demonstrating coherent operations driven exclusively by the integrated optical path.

The extracted intensity profile agrees well in shape and width with both an independent camera measurement and an ideal Gaussian of radius $170$8m. One discrepancy remains unexplained: the ion-based data show a secondary intensity peak near $170$9m absent from the camera profile. Both datasets exhibit significant local intensity fluctuations, attributed plausibly—but not proven—to wavefront perturbations from defects on the manually cleaved facet. Because the emitted polarization is unknown, absolute intensity cannot be inferred from Rabi frequencies alone, limiting quantitative characterization.

Limitations and open questions

The paper is explicit about several constraints. First, FLW waveguide curvature is limited to radii of order a few millimeters, constraining routing density relative to lithographic photonics. Second, the unlensed output delivers a $240~\mu$0m beam radius at the ion, so the demonstrated capability is global illumination of the addressing zone, not single-ion addressing; achieving micrometer spots requires hybrid pick-and-place micro-optics (e.g., ball lenses yielding $240~\mu$1m waists), which the authors argue are compatible with the platform but do not demonstrate here. Third, the manual cleaving process introduces facet variability that directly degrades beam pointing and produces unexplained intensity structure; a manufacturing-grade facet process is an open question. Fourth, the off-axis waveguide formation mechanism—and specifically the sharp performance optimum at non-zero focus offset—is not fully explained theoretically. Finally, stray-field stability is characterized over hours at 1 mm standoff; longer-term behavior and closer dielectric proximities remain untested.

Conclusion

This work establishes a complete, industrially fabricated ion-trap platform combining a single-metal-layer surface trap on a silicon-compatible borosilicate stack with bonded FLW waveguides emitting in-plane at the ion height. Key quantitative results include MFD-tunable single-mode guidance at 729 nm and 405 nm, sub-6 mm bend radii with sub-dB/cm loss, heating rates of 0.5–1.5 quanta/s, stable glass-induced stray fields drifting at $240~\mu$2 V/mm per hour, and coherent Rabi oscillations driven through the fully integrated optical path. The demonstration stops short of single-ion addressing, leaving the integration of hybrid micro-optics onto this platform as the principal open step toward scalable, individually addressed trapped-ion processors.