Wafer-Level Atomic Vapor Cells

• Wafer-level atomic vapor cells are microfabricated alkali vapor or vacuum cells produced on bonded wafers that offer precise optical access and reliable cavity geometries for integrated quantum devices.
• They employ advanced fabrication methods such as DRIE, anodic bonding, and laser-assisted etching to create diverse architectures including glass–silicon–glass stacks, all-glass cells, and vapor–PIC hybrids.
• Applications span chip-scale atomic clocks, magnetometers, Rydberg sensors, and nonlinear optics where rigorous atmosphere engineering and surface chemistry optimize device performance and stability.

Searching arXiv for recent and relevant papers on wafer-level atomic vapor cells and closely related microfabricated alkali-cell platforms. Wafer-level atomic vapor cells are microfabricated alkali-vapor or vacuum cells produced collectively from bonded wafers, most commonly in glass–silicon–glass or all-glass stacks, and engineered to provide hermetic optical access, controlled alkali loading, and reproducible cavity geometry for chip-scale atomic clocks, magnetometers, Rydberg sensors, vapor-integrated photonics, nonlinear optics, and cold-atom packages (Li et al., 2024, Artusio-Glimpse et al., 19 Mar 2025, Kelleher et al., 30 Jan 2026). In the recent literature, the term encompasses both conventional MEMS alkali cells and newer devices in which the enclosed vapor volume is lithographically structured into optical, RF, or fluidic functions, including evacuated cells, tunable buffer-gas cells, patterned diffractive elements, multi-axis cells, and photonic-integrated vapor packages (Stern et al., 2018, Carlé et al., 29 Aug 2025, Shrestha et al., 22 Dec 2025).

1. Definition, scope, and boundaries of the field

Wafer-level atomic vapor cells are defined less by a single geometry than by a manufacturing logic: cavities, reservoirs, and optical windows are formed collectively on wafers; alkali is introduced by precursor chemistry, dispenser pills, or ex situ filling; and the final structure is sealed by anodic bonding, direct bonding, local laser sealing, or related vacuum-packaging methods (Li et al., 2024, Péroux et al., 15 Feb 2025, Kelleher et al., 30 Jan 2026). This places them in continuity with MEMS atomic devices, but the cited work shows that the category now extends well beyond rectangular spectroscopy chambers.

The scope of the field is clarified by comparison with adjacent platforms. “Atomic vapor spectroscopy in integrated photonic structures” places a silicon nitride chip inside an external rubidium vapor environment and uses evanescent coupling to exposed waveguides, but explicitly does not implement a sealed wafer-level atomic vapor cell; its relevance is architectural and methodological rather than hermetic (Ritter et al., 2015). By contrast, “Laser cooling in a chip-scale platform” demonstrates a micro-fabricated Si/glass vacuum cell that is actively pumped to ultra-high vacuum for a grating MOT, showing that wafer-derived alkali packaging can extend from hot-vapor devices toward cold-atom micro-UHV systems (McGilligan et al., 2020).

This boundary matters because a recurring misconception is that all chip-scale atom–photon platforms are equivalent. The literature instead separates at least three classes: hermetically sealed wafer-level vapor cells, open or chamber-immersed integrated-photonics platforms, and microfabricated UHV cells for laser cooling. Their fabrication steps overlap, but their atmosphere control, reliability constraints, and optical functions differ substantially (Ritter et al., 2015, McGilligan et al., 2020).

2. Structural architectures and material systems

A standard architecture remains the bonded silicon–glass cell. Wafer-scale evacuated rubidium arrays were fabricated on a 4-inch, 1 mm thick silicon wafer with through-etched cavity pairs and borosilicate glass caps (Li et al., 2024). Rydberg spectroscopy has been demonstrated in a wafer-scale fabricated Pyrex–Si–Pyrex cell with a $2 \times 2 \times 1.4~\mathrm{mm}^3$ interaction cavity connected to a dispenser region by a small aperture (Giat et al., 13 Apr 2025). For Rydberg electrometry, a glass-silicon-glass sandwiched structure using a specially customized 6 mm thick silicon wafer with resistivity exceeding $10{,}000~\Omega\cdot\mathrm{cm}$ was developed to provide a 4-fold improvement in optical interrogation length while reducing RF disturbance (Ma et al., 2 Sep 2025).

A second major lineage replaces silicon in the RF-active region. An all-glass wafer-level process based on a direct-bonded Borofloat 33 triple stack uses a 2 mm middle wafer to define either an open cell with one open reservoir $19~\mathrm{mm}\times 13~\mathrm{mm}$ and evacuated volume $494~\mathrm{mm}^3$, or a supported cell with seven $19~\mathrm{mm}$ long by $1.3~\mathrm{mm}$ tall trenches connected by one $1~\mathrm{mm}$ wide vertical trench and evacuated volume $354.2~\mathrm{mm}^3$ (Artusio-Glimpse et al., 19 Mar 2025). A distinct all-glass vacuum route uses selectively etched fused silica wafers, Al$_2$O$_3$ permeation barriers, Ti micro-knives, and a single bonding interface to realize evacuated vapor cells and atom-beam cells (Kelleher et al., 30 Jan 2026).

A third architectural direction embeds additional optical functionality into the cell body. “Chip-scale atomic diffractive optical elements” forms a silicon-based etched cavity structure filled with rubidium vapor and capped by borosilicate glass, with the enclosed vapor volume patterned into a lamellar grating of period $10{,}000~\Omega\cdot\mathrm{cm}$0 and depth $10{,}000~\Omega\cdot\mathrm{cm}$1, or into a 2 mm diameter atomic Fresnel lens with designed focal length 70 mm (Stern et al., 2018). “Microfabricated multi-axis cell for integrated atomic devices” uses a five-wafer stack—three borosilicate glass wafers and two silicon wafers—to create three orthogonal optical pathways, including embedded lateral windows formed by laser-assisted etching and thermal reflow (Péroux et al., 26 Sep 2025).

Photonic-integrated variants now place the vapor cell directly over a PIC. “Enabling atom-clad waveguide operation in a microfabricated alkali vapor-photonic integrated circuit” uses a three-layer bonded stack comprising a SiN PIC die, a micromachined borosilicate glass frame with a center chamber and side chamber, and a borosilicate lid; the air-clad waveguide interaction lengths are $10{,}000~\Omega\cdot\mathrm{cm}$2 and 3 mm (Shrestha et al., 22 Dec 2025). This suggests that wafer-level atomic vapor cells have become structural platforms for co-designed optical, microwave, and photonic functions rather than passive containers alone.

3. Wafer-level fabrication, sealing, and alkali loading

The dominant fabrication vocabulary is still DRIE, laser machining or laser-assisted etching, anodic bonding, dispenser insertion, and post-seal activation. In the atomic diffractive-optics platform, two wafer-level routes were explored: direct blind etching into a 2 mm thick silicon 4-inch wafer, and an etch-stop-enabled stack with a 170 $10{,}000~\Omega\cdot\mathrm{cm}$3m silicon upper layer, 200 nm SiO$10{,}000~\Omega\cdot\mathrm{cm}$4, and 830 $10{,}000~\Omega\cdot\mathrm{cm}$5m handle layer, followed by anodic bonding to borosilicate glass and activation of RbMbO$10{,}000~\Omega\cdot\mathrm{cm}$6/AlZr dispenser pills using approximately 1 W of 980 nm light (Stern et al., 2018).

Wafer-scale evacuated rubidium cells use a different sequence. A 4-inch, 1 mm thick silicon wafer is DRIE-etched to form paired cell and deposition cavities, then patterned with $10{,}000~\Omega\cdot\mathrm{cm}$7 deep and $10{,}000~\Omega\cdot\mathrm{cm}$8 wide surface channels that run to the wafer edge and pass within 1 mm of each cell. After first anodic bonding to borosilicate glass, BaN$10{,}000~\Omega\cdot\mathrm{cm}$9 and RbCl are dispensed into deposition cavities, rubidium is generated in situ at $19~\mathrm{mm}\times 13~\mathrm{mm}$0, transferred laterally, and the cells are finally sealed by anodic bonding at $19~\mathrm{mm}\times 13~\mathrm{mm}$1 and around 1300 V (Li et al., 2024).

Local rather than global final sealing has also emerged. “Locally-sealed microfabricated vapor cells filled from an ex situ Cs source” fabricates a glass/silicon/glass stack in which a 200 $19~\mathrm{mm}\times 13~\mathrm{mm}$2m top borosilicate wafer carries cylindrical $19~\mathrm{mm}\times 13~\mathrm{mm}$3 channels made by laser-assisted selective etching, a 500 $19~\mathrm{mm}\times 13~\mathrm{mm}$4m silicon wafer is through-etched by DRIE, and the final closure is achieved after external filling by local CO$19~\mathrm{mm}\times 13~\mathrm{mm}$5-laser-induced collapse of the glass microchannels (Péroux et al., 15 Feb 2025). The laser is a 10.6 $19~\mathrm{mm}\times 13~\mathrm{mm}$6m Keyence ML-9110 system with 12 W maximum power and a 1 mm beam width at the channel.

All-glass direct bonding uses higher thermal budgets but eliminates silicon from the active RF region. In the Borofloat 33 triple-stack process, the bonding chamber is evacuated to $19~\mathrm{mm}\times 13~\mathrm{mm}$7, the wafers are brought together, heated to $19~\mathrm{mm}\times 13~\mathrm{mm}$8 to $19~\mathrm{mm}\times 13~\mathrm{mm}$9, loaded with $494~\mathrm{mm}^3$0, and held for 20 hours before natural cooling (Artusio-Glimpse et al., 19 Mar 2025). By contrast, the micro-knife platform forms a grain-boundary thermo-compression bond through plastic deformation of Ti knives into a thick compliant metal layer; low-temperature test bonds are reported at $494~\mathrm{mm}^3$1, and final devices show shear-force strength $494~\mathrm{mm}^3$2 (Kelleher et al., 30 Jan 2026).

The interaction between sealing temperature and internal surface chemistry has become a central design issue. “MEMS Vapor Cells With Passivated Internal Cavities” first forms a high-temperature base frame, then deposits OTS on the internal cavity, loads $494~\mathrm{mm}^3$3–$494~\mathrm{mm}^3$4 of pure Cs by piezo-electric dispensing, and completes a second low-temperature capping seal at $494~\mathrm{mm}^3$5, $494~\mathrm{mm}^3$6 to $494~\mathrm{mm}^3$7, for 90–120 min so that the organic monolayer survives (Pandiyan et al., 28 Nov 2025).

Post-fabrication atmosphere programmability has also been demonstrated. In tunable He–Ne cells, a silicon wafer is formed by double-sided lithography and two-step DRIE, one anodic bond traps helium in sealed reservoirs, Cs pill dispensers are loaded, a second anodic bond seals the main cavities under neon, and thin silicon break-seals are later opened by laser ablation to incrementally adjust the final He fraction (Carlé et al., 29 Aug 2025).

4. Atmosphere engineering, surfaces, and stability

The internal atmosphere of wafer-level cells is no longer limited to a single final-bond condition. Evacuated rubidium arrays reduce trapped gas by shortening the pumping path from up to 50 mm to below 1 mm with surface channels, reaching residual gas pressure below 0.5 kPa (4 Torr) in more than 50% of cells and a yield of 51% relative to the cells initially loaded with precursor (Li et al., 2024). The linewidth analysis uses the standard power-broadening relation

$494~\mathrm{mm}^3$8

with excess linewidth converted to an upper bound on residual pressure (Li et al., 2024).

Ex situ filling with local sealing changes the process logic. In locally sealed cesium cells, the bonded stack is baked for a few days at $494~\mathrm{mm}^3$9, the chamber typically reaches $19~\mathrm{mm}$0 mbar, cesium migrates into the cells within one hour under a thermal gradient, and local channel collapse yields cells that support saturated absorption at $19~\mathrm{mm}$1 with a 0.5 mm path length and a Lorentzian FWHM of 60 MHz (Péroux et al., 15 Feb 2025). One monitored cluster showed two saturated cells with stable atmosphere over 60 days at $19~\mathrm{mm}$2, with absorption contrast between 30% and 40% and linewidth constant at 400 MHz (Péroux et al., 15 Feb 2025).

Buffer-gas composition has also become tunable after sealing. In He–Ne cesium cells, the clock frequency is modeled as

$19~\mathrm{mm}$3

with

$19~\mathrm{mm}$4

Sequential opening of helium reservoirs shifts the inversion temperature from $19~\mathrm{mm}$5 in pure neon to $19~\mathrm{mm}$6, $19~\mathrm{mm}$7, and $19~\mathrm{mm}$8, corresponding to estimated He concentrations of 1.2%, 2.3%, and 4.5%; one fabricated cell achieved fractional frequency stability of $19~\mathrm{mm}$9 at one-day integration time (Carlé et al., 29 Aug 2025).

Surface chemistry is equally consequential. OTS-coated MEMS cells reduce Cs surface coverage from 2.9% on bare glass and 3.0% on bare Si to 0.2% and 0.3%, respectively, and yield Rydberg spectral linewidths down to $1.3~\mathrm{mm}$0 with inferred electric fields below $1.3~\mathrm{mm}$1 (Pandiyan et al., 28 Nov 2025). In Rydberg all-glass cells, repeated EIT measurements over 23 months in a supported cell bonded on 25 January 2023 and activated on 6 March 2023 showed stable operation for 707 days and counting (Artusio-Glimpse et al., 19 Mar 2025). In fused-silica micro-knife cells, vapor cells exhibit long lifetimes $1.3~\mathrm{mm}$2 year, low residual gas pressures $1.3~\mathrm{mm}$3, and leak rates below fine-leak testing sensitivity $1.3~\mathrm{mm}$4 (Kelleher et al., 30 Jan 2026).

A persistent practical issue is alkali deposition on nearby photonic surfaces. Open vapor-clad SiN waveguides showed build up of a Rb layer on their surface over time (Ritter et al., 2015). In a sealed vapor–PIC, successful operation depended on low-power pulsed dispenser activation and a counter-propagating 801 nm desorption laser that completely suppresses Rb-induced losses and enables waveguide-based atomic vapor spectroscopy (Shrestha et al., 22 Dec 2025).

5. Optical, microwave, and quantum functions enabled by wafer-level cells

Wafer-level cells now support functions far beyond absorption references. In atomic diffractive optical elements, the enclosed alkali vapor is itself one arm of a diffractive phase–amplitude profile, with far-field response calculated from

$1.3~\mathrm{mm}$5

and grating orders varying qualitatively as $1.3~\mathrm{mm}$6 and $1.3~\mathrm{mm}$7. The grating showed diffraction efficiency modulation up to 50% by changing laser frequency, and the Fresnel lens switched between “on” and “off” states with optical power contrast exceeding 13 dB by changing laser frequency by only 1 GHz (Stern et al., 2018). Nonlinear optics has likewise moved on-chip: micromachined Rb cells with $1.3~\mathrm{mm}$8 mm interaction length generated coherent blue light of 17 $1.3~\mathrm{mm}$9W and collected coherent mid-IR power of 50 nW, with blue linewidth $1~\mathrm{mm}$0 MHz and higher coherent blue-light generation efficiency than a 7 cm glassblown cell under the reported conditions (Krelman et al., 3 May 2026).

Integrated photonics has become a second major axis. The open vapor-clad SiN platform demonstrated straight-waveguide spectroscopy and a Mach–Zehnder interferometer with about 17% of the TE mode overlapping rubidium vapor, showing OD $1~\mathrm{mm}$1 at $1~\mathrm{mm}$2 and extracted atom-induced phase shifts up to $1~\mathrm{mm}$3 (Ritter et al., 2015). The sealed vapor–PIC platform went further by integrating a pill-type Rb dispenser with an air-clad SiN waveguide and borosilicate microcell, enabling repeatable control of vapor density by activation pulse length, duty cycle, and device temperature, together with 3 mm atom-clad waveguide spectroscopy (Shrestha et al., 22 Dec 2025). At still smaller thicknesses, a wafer-fabricated sub-micron Rb cell at $1~\mathrm{mm}$4 showed that wall-collision-induced relaxation in 500 nm and 1 $1~\mathrm{mm}$5m cells suppresses optical pumping into uncoupled states and leaves the telecom-adjacent $1~\mathrm{mm}$6 cycling transition dominant, creating an effective near-infrared two-level system at 1529.37 nm (Orr et al., 22 Jan 2026).

Rydberg electrometry has become a primary driver of new wafer-level designs. Wafer-scale fabricated Pyrex–Si–Pyrex cells supported EIT/Autler–Townes electrometry with line widths about 20 MHz, spatial resolution $1~\mathrm{mm}$7 at $1~\mathrm{mm}$8 GHz, inferred internal DC fields from about $1~\mathrm{mm}$9 to about $354.2~\mathrm{mm}^3$0, and estimated sensitivity as low as $354.2~\mathrm{mm}^3$1 (Giat et al., 13 Apr 2025). All-glass direct-bonded cells measured a 34.009 GHz field resonant with the $354.2~\mathrm{mm}^3$2 transition; the Autler–Townes splitting was linear versus $354.2~\mathrm{mm}^3$3, and a linear fit yielded $354.2~\mathrm{mm}^3$4 (Artusio-Glimpse et al., 19 Mar 2025). Wafer-level MEMS cesium cells with 6 mm high-resistivity silicon achieved a minimal detectable microwave field of $354.2~\mathrm{mm}^3$5 (Ma et al., 2 Sep 2025).

Other sensing modes are likewise now realized. The multi-axis cesium cell with reflowed lateral windows demonstrated magnetic sensitivities of $354.2~\mathrm{mm}^3$6 through the main windows and $354.2~\mathrm{mm}^3$7 through lateral windows or in dual-beam operation, with a zero-field resonance FWHM of 22 nT (Péroux et al., 26 Sep 2025). Micro-fabricated Si/glass UHV cells have also supported laser cooling and a grating MOT, including a 6-beam MOT with $354.2~\mathrm{mm}^3$8 atoms and an inferred pressure of $354.2~\mathrm{mm}^3$9 during operation (McGilligan et al., 2020). These results suggest that wafer-level atomic vapor cells now span hot-vapor spectroscopy, nonlinear optics, RF metrology, magnetometry, photonic integration, and cold-atom precursor platforms.

6. Design tensions, misconceptions, and future directions

A first design tension concerns silicon. Silicon remains attractive because DRIE, anodic bonding, and wafer handling are mature, but several Rydberg papers treat it as electromagnetically non-neutral. The all-glass Rydberg-cell work argues that silicon is less ideal for microwave and millimeter-wave compatibility because of its high dielectric constant and conductive loss, whereas the thick-silicon MEMS work argues that ultra-high-resistivity silicon is a practical compromise that minimizes RF field distortion relative to typical silicon (Artusio-Glimpse et al., 19 Mar 2025, Ma et al., 2 Sep 2025). Numerical full-vector FEM over 0.05 GHz to 150 GHz sharpens the point: structured all-glass supported cells exhibit localized RF power enhancements exceeding 8x, while highly doped silicon support layers suppress the resonant features through material loss (Maurya et al., 9 Sep 2025). This suggests that “wafer-level” does not imply a single optimal material stack; the correct choice is application-specific.

A second tension concerns internal surfaces and alkali-source geometry. Rydberg sensing in wafer-fabricated Pyrex–Si–Pyrex cells revealed a complex internal electrostatic landscape tied to surfaces, adsorbates, liquid Rb, and the dispenser region (Giat et al., 13 Apr 2025). OTS passivation, reservoir separation, and low-temperature capping are direct responses to that problem (Pandiyan et al., 28 Nov 2025). In vapor–PIC devices, the integrated source becomes both an enabling element and a contamination risk, requiring pulsed activation and desorption-laser mitigation (Shrestha et al., 22 Dec 2025).

A third misconception is that wafer-level fabrication automatically solves reliability. Several architecture papers explicitly do not provide detailed hermetic leak rates, long-term lifetime, outgassing, contamination, or yield. The evacuated Rb array paper reports 51% yield; the local-seal paper shows only a subset of cells in a monitored cluster filling and sealing successfully; the vapor–PIC paper does not quantify leak rate, bond yield, or long-term dispenser lifetime (Li et al., 2024, Péroux et al., 15 Feb 2025, Shrestha et al., 22 Dec 2025). Conversely, the most reliability-focused platforms—micro-knife fused-silica cells and long-lived all-glass Rydberg cells—show that lifetime and leak-rate evidence can be made central rather than implicit (Kelleher et al., 30 Jan 2026, Artusio-Glimpse et al., 19 Mar 2025).

The current trajectory is therefore not a single linear progression from silicon MEMS to one “best” cell, but a diversification of wafer-level atomic packaging strategies: direct-bonded all-glass cells for RF transparency, locally sealed cells for ex situ filling and coating compatibility, laser-actuated break-seal cells for tunable gas mixtures, patterned vapor volumes for diffractive optics, reflowed side-window cells for multi-axis access, micro-knife sealed fused-silica cells for low leak and low permeation, and vapor–PIC packages for atom-clad nanophotonics (Artusio-Glimpse et al., 19 Mar 2025, Péroux et al., 15 Feb 2025, Carlé et al., 29 Aug 2025, Stern et al., 2018, Péroux et al., 26 Sep 2025, Kelleher et al., 30 Jan 2026, Shrestha et al., 22 Dec 2025). A plausible implication is that future wafer-level atomic vapor cells will be evaluated not only by whether they confine alkali vapor, but by how precisely they co-engineer atmosphere, surfaces, RF response, optical transfer function, and manufacturable integration.