Skip to content

Scalable Photonic Quantum Interconnect Platform

FieldValue
JournalPhysical Review X
Published2026-02-06
AuthorsDaniel Riedel, Teodoro Graziosi, Zhuoxian Wang, Chawina De-Eknamkul, Alex Abulnaga
DOI10.1103/nfrg-zsts
Citations4

Silicon-vacancy quantum memories in thin-film diamond reached a measured cavity cooperativity of 46. The authors report best devices with C near 50, while 10 of 12 measured interfaces across three chips had C > 5. The membrane transfer process reported near-unity yield and required less than 1 minute per bond.

The authors developed single-crystal diamond membranes on semiconductor handling chips as an alternative to bespoke bulk-die nanofabrication. This separates high-temperature defect processing from lower-temperature substrate and photonic fabrication. Each membrane carried a matrix of nanophotonic devices, while the substrate could carry control lines. Silicon-vacancy centers were targeted through aligned nanoapertures before membrane release and bonding. Targeted implantation reduces background SiV fluorescence from unintended, weakly coupled centers, which raises the final device yield.

A 0.5 MeV helium implantation at a 7° tilt and 5 × 10¹⁶ cm⁻² dose creates a subsurface damage layer. Annealing at 1300 °C, 1 × 10⁻⁶ Pa, and 120 min graphitizes that layer. Masking the diamond edge leaves a frame that anchors the membrane during undercutting and bonding. Electrochemical etching removes the graphitized layer after trenches define tethered membranes.

Homoepitaxial overgrowth produces a typical 0.5 μm layer with less than 10 nm thickness variation per millimeter. The as-grown surface roughness is below 0.2 nm rms.

Aligned 100 nm by 100 nm nanoapertures restrict silicon-28 implantation to 125 kV and 1 × 10¹² cm⁻². A 1250 °C anneal at 1 × 10⁻⁸ Pa for 90 min creates typically two to three SiV centers per implantation spot. Gold thermocompression bonding transfers each membrane to a matching gold pad on a silicon handling chip. The metal layers are 50 nm titanium and 100 nm gold, although the methods also permit silicon dioxide bonding. Bonding occurs at 150 °C, 5 kPa, and 15 s. The process supported 25 consecutive transfers from one parent diamond die. A 3 × 6 grid contained 18 membranes with 5 mm spacing. The malleable metal interface accommodated contamination, hillocks, and pits from overgrowth.

After bonding, membranes were thinned to 160 nm for cavity fabrication and reached roughness below 0.3 nm rms. Each membrane received a 4 × 32 array of photonic crystal cavities.

At 20 mK, the authors tuned cavity resonances by depositing nitrogen ice and sublimating it with controlled laser heating. Tapered-fiber reflection spectra resolved several SiV centers coupled to the cavity mode. A representative fit gave C = 46 and g = 2π × 3.52 GHz. The total cavity linewidth was 2π × 11.46 GHz, and the SiV linewidth was 2π × 94 MHz. The fit assumed an SiV transition at 406.755 THz. Across three chips, 10 of 12 preselected devices had C > 5. The authors attribute most variation to random defect positions relative to the cavity-mode maximum.

An external magnetic field along the SiV axis split the spin states and produced spin-dependent reflection. Spin jumps under weak illumination gave T1 = 32 s, which the authors report as a lower bound. They identify weak probe illumination as the main relaxation channel in this measurement. The bound therefore reflects the probe and field conditions. The ultimate T1 depends on phonon relaxation, field magnitude and direction, strain, and temperature.

Four membranes bonded 15 mm apart on one handling chip each received a 4 × 32 cavity array. The authors measured 48 devices per membrane at room temperature using a supercontinuum source and spectrometer. They excluded devices with known proximity-effect shifts. The representative cavity had Q = 3.20 × 10⁴ at 743.5 nm and 92.5% reflection contrast. Mean Q values remained between 2.18 × 10⁴ and 2.36 × 10⁴. Resonance wavelength spread stayed below 2 nm within each membrane.

The measurements produced these distributions for the four membranes. Each row lists the paper’s mean and spread.

MembraneResonance wavelengthTotal Q, in units of 10⁴
1744.7 ± 1.9 nm2.36 ± 0.41
2744.2 ± 1.8 nm2.20 ± 0.49
3751.2 ± 1.6 nm2.21 ± 0.38
4747.3 ± 1.4 nm2.18 ± 0.35

The sub-2 nm spread was smaller than the approximately 10 nm cryogenic tuning range from nitrogen condensation. The authors therefore report a near-unity yield of usable resonant cavities across a wafer. By comparison, angled-etched devices typically placed only about 10% of cavities within the SiV transition tuning range on one chip.

Electronic integration and optical packaging

Section titled “Electronic integration and optical packaging”

Functionalized silicon substrates provided buried coplanar waveguides for microwave delivery to color centers in bonded diamond. One substrate had nine independent control lines and a roughly 1 μm SiO2 cladding layer. Nitrogen-vacancy centers were implanted about 60 nm below the surface for this proof of concept. Resonant microwave pulses through the buried waveguide produced ODMR and Rabi oscillations. At 15 dBm, the measured tπ was 55 ns, while the ensemble Rabi decay constant was 0.93 μs. The authors attribute the fast decay to the spatially non-uniform bias field across an ensemble. Placing the waveguide below the membrane reduces the heat load of parallel microwave addressing compared with traces on the diamond.

Optical packaging attached a tapered fiber to a tapered waveguide coupler without active alignment. Comparing cavity reflection with a retroreflector gave approximately 85% coupling efficiency, corresponding to less than 1 dB loss. The paper cites angled-etched devices with less than 0.5 dB loss as the packaging benchmark. Oxide bonding was demonstrated as a metal-free route on coplanar waveguide chips. Full foundry-level oxide bond yield remains a projection.

Full wafer reconstitution was proposed as an extension, but the demonstrations covered 18 membranes in a grid and 25 transfers from one die. Metal thermocompression bonding remains the demonstrated high-yield route. The authors expect oxide bonding to approach similar yield only after foundry development. Defect placement still varies because implanted SiV centers do not land at the cavity-mode maximum. That placement variation primarily produces the spread in cooperativity.

Run-to-run shifts and differences between membranes still require compensation. The proposed controls include metrology during fabrication, witness chips, and lithography offsets. The uniformity test sampled 48 cavities from each 128-cavity array, leaving the untested fraction without reported Q or wavelength data. The preselected 12-device cooperativity sample is not a wafer-level yield measurement.

The supporting data are not publicly available because they contain commercially sensitive information. They are available from the authors upon reasonable request. The current optical packaging has higher loss than the less than 0.5 dB result shown in earlier angled-etched devices. The number of optical inputs and outputs remains limited by membrane size. The authors propose heterogeneous integration with lithium niobate or silicon nitride to raise that limit.

Engineering implications for diamond selection

Section titled “Engineering implications for diamond selection”

Material selection for this architecture must address final membrane thickness, lateral size, and roughness. The principal cavity process used 160 nm membranes with less than 0.3 nm rms roughness. Overgrowth quality also controls thickness variation before thinning. Defect implantation must align the SiV target with the optical mode, because position drives cooperativity. The bonding interface must tolerate surface irregularities without reducing transfer yield. Separating control electronics into the substrate allows lower-temperature processing and parallel microwave delivery.

The measured results cover membrane arrays, three-chip SiV sampling, and four-membrane cavity uniformity. The stated development goals are full wafer reconstitution, foundry oxide bonding, and heterogeneous photonic integration. These goals should not be treated as completed results.

The photonic quantum memory platform uses single-crystal diamond membranes, typically 100 to 300 nm thick, with Au-bearing bond layers in the main bonding route. 6CCVD’s diamond materials catalog lists MPCVD-grown single-crystal diamond, thickness from 0.1 to 500 micrometers, and Au metallization. Use engineering support for process-specific membrane preparation, metallization, or polishing questions, with surface and bond-interface specifications qualified against the paper.

View Original Abstract

Many quantum networking applications require efficient photonic interfaces to quantum memories which can be produced at scale and with high yield. Synthetic diamond offers unique potential for the implementation of this technology as it hosts color centers which retain coherent optical interfaces and long spin coherence times in nanophotonic structures. Here, we report a technique enabling wafer-scale processing of thin-film diamond that combines ion implantation and membrane liftoff, high-quality overgrowth, targeted color center implantation, and serial, high-throughput thermocompression bonding with yields approaching unity. The deterministic deposition of thin diamond membranes onto semiconductor substrates facilitates consistent integration of photonic crystal cavities with silicon-vacancy (SiV) quantum memories. We demonstrate reliable, strong coupling of SiVs to photons with cooperativities approaching 50. Furthermore, we show that photonic crystal cavities can be reliably fabricated across several membranes bonded to the same handling chip. Our platform enables modular fabrication where the photonic layer can be integrated with functionalized substrates featuring electronic control lines such as coplanar waveguides for microwave delivery. Finally, we implement passive optical packaging with sub-decibel insertion loss. Together, these advances pave the way to the scalable assembly of optically addressable quantum memory arrays which are a key building block for modular photonic quantum interconnects.