Free-Space-Coupled Frequency-Locked Microtoroid Resonators with Reactive Polymer Functionalization for Part-Per-Trillion Gas Detection

Abstract

Whispering gallery mode (WGM) microtoroid resonators combined with frequency locking offer single-molecule detection sensitivity, but their dependence on fragile fiber tapers has limited their use to controlled laboratory environments. Here, we overcome this barrier by integrating frequency-locked WGM microtoroids with a robust free-space optical coupling scheme, enabling ultra-sensitive and stable gas-phase chemical sensing. Using polymer materials synthesized via reversible addition–fragmentation chain-transfer (RAFT) polymerization, we demonstrate selective and durable interaction with volatile thioethers, with 2-chloroethyl ethyl sulfide (2-CEES, a common mustard gas simulant) serving as a model target. Importantly, the polymer functionalization preserves ultra-high Q factors (>10⁷) after coating, ensuring performance is not compromised. Housed in a compact, coin-sized chamber, the sensor achieves a room-temperature detection limit of 25 parts per trillion—three orders of magnitude more sensitive than prior reports—while maintaining mechanical resilience under ambient conditions. Continuous frequency locking enables unattended, long-duration monitoring. This first demonstration of free-space–coupled WGM microtoroids for chemical sensing with novel polymer materials establishes a scalable platform for ultra-sensitive, rugged detectors with broad applications in defense, security, and environmental monitoring.

Introduction

Ultra-sensitive and reliable gas-phase chemical sensors are critically needed for applications ranging from environmental monitoring and industrial safety to defense and security. They also play a central role in the exposome framework, which emphasizes comprehensive measurement of environmental exposures across the human lifespan¹. Achieving part-per-trillion (ppt) sensitivity under real-world conditions is particularly important for early warning and exposure prevention. Chemical warfare agents such as sulfur mustard (HD, bis(2-chloroethyl) sulfide) exemplify the need for rapid, ultra-sensitive detectors, yet their extreme toxicity complicates laboratory development^2,3. Structurally and chemically related simulants, including 2-chloroethyl ethyl sulfide (2-CEES), are therefore widely used in detection, decontamination, and protective-system studies^4–6. While CEES provides a convenient model system, the broader demand extends well beyond chemical warfare agents, encompassing volatile organics and toxic industrial chemicals relevant to both security and environmental monitoring. Developing compact, rugged, and field-deployable sensors capable of ppt-level detection remains an unmet challenge.

A variety of techniques have been investigated for trace-level detection of CEES and related volatile organics, including gas chromatography⁷, chemiresistive sensors^8–17, surface acoustic wave devices^18,19, fluorescent chemosensors^20–24, photonic crystal reflection spectroscopy²⁵, microbalance sensors²⁶, and surface-enhanced Raman scattering (SERS) spectroscopy^27,28. While these approaches can achieve ppb-level sensitivity, they often require high operating temperatures or rely on complex, expensive instrumentation that limits their practicality. The most sensitive optical methods demand long integration times and bulky systems. To date, no method has demonstrated ppt-level CEES detection or related volatile organics under ambient conditions, and no existing sensor architecture simultaneously offers ultra-high sensitivity, mechanical robustness, and long-term unattended operation.

Whispering gallery mode (WGM) microtoroid optical resonators provide a promising platform for ultra-sensitive chemical sensing. By confining light in ultra-high-Q (Q > 10⁷) cavities, these devices strongly amplify light–matter interactions and can resolve single-molecule binding events²⁹. While various on-chip ultra-high-Q resonators exist, microtoroids were specifically chosen for this platform due to their combination of ultra-high quality (Q > 10⁷) factors, small mode volumes, and compact on-chip format. Furthermore, the curved rim of the toroid geometry provides a large, accessible evanescent field that maximizes the interaction between the optical mode and its surroundings.^30,31 When combined with frequency locking, resonance shifts can be tracked continuously, enabling real-time operation with unparalleled sensitivity ^29,32–38. In gas sensing applications, chemoselective polymer coatings transduce molecular adsorption events into local refractive index changes within the resonator’s evanescent field ^39,40. However, conventional WGM implementations rely on fragile fiber tapers for optical coupling, which are highly susceptible to drift and vibration, restricting their use to carefully controlled laboratory settings. This fragility has prevented translation of WGM sensors into rugged, deployable devices. We recently demonstrated that free-space coupling using a long-working-distance objective and digital micromirror device could provide robust temperature sensing ⁴¹, suggesting a pathway toward stable, field-ready chemical sensors.

In this work, we utilize the Frequency-Locked Optical Whispering Evanescent Resonator (FLOWER) platform for trace detection of volatile thioesters. FLOWER is an electronic feedback architecture that continuously matches the laser frequency to the center of the microtoroid’s resonance.^{29,32,33,37,42–44} A proportional-integral-derivative (PID) controller monitors the optical signal; as the resonance shifts due to molecular binding, the controller applies a correction voltage to the laser’s internal piezo-tuner to re-center the wavelength. This locking approach transforms the resonator from a passive filter into an active probe, allowing for continuous, high-speed acquisition of the resonance position without the ‘dead time’ associated with traditional wavelength sweeping. While we have previously implemented FLOWER using fiber-taper coupling, this work represents its first integration with a robust free-space optical interface for trace gas detection.

Polyvinylamine coatings synthesized via reversible addition–fragmentation chain-transfer (RAFT) polymerization were immobilized on the resonator surface, providing selective and durable binding to thioethers, with 2-CEES serving as a representative target. Sulfur mustard and its analogues form episulfonium intermediates that rapidly alkylate nucleophilic sites^45,46, a reactivity we exploit by trapping them with polyvinylamine (PVAm), a polymer rich in primary amine side chains (Scheme 1). The binding of 2-CEES to the PVAm polymer layer modified surface perturbs the microtoroid’s optical mode, resulting in a measurable shift in its optical resonance. While aminopyridine-substituted polyallylamines have been used for 2-CEES degradation⁴⁷, this is the first application of PVAm for WGM sensing⁴⁸. The resulting sensor, housed in a compact coin-sized chamber, achieves a 25 ppt detection limit (17 ppt theoretical), nearly three orders of magnitude below prior reports, all at room temperature and atmospheric pressure. The system maintains ultra-high Q factors after polymer functionalization, operates stably under repeated mechanical shocks, and enables continuous “set-and-forget” monitoring over extended durations. To reduce system complexity and enhance robustness for potential field-deployable sensing, the present design replaces the DMD-based programmable pinhole used previously⁴¹ with a compact flippable mirror. The mirror enables side-view imaging during alignment and facilitates collection of resonance-scattering signals during frequency locking. This modification eliminates the 4-f relay, imaging lens, and associated alignment degrees of freedom, shortening the optical path length by approximately 60% and reducing the number of optical components by about half. This first demonstration of free-space–coupled WGM resonators for chemical sensing establishes a scalable platform for portable, ultra-sensitive, and mechanically resilient detectors with immediate applications in battlefield monitoring, environmental compliance, and industrial safety.

Scheme 1.

Proposed Mustard-Induced Alkylation of Polyvinylamine

Methods

Polymer synthesis.

Poly(N-vinylformamide) was synthesized via RAFT polymerization of N-vinylformamide at 35 °C using xanthate CTA-1 and 2,2′-azobis(4-methoxy-2,4-dimethylvaleronitrile (V-70) (Scheme 2A). CTA-1 containing a succinamide ester was employed to reliably install the desired polymer end group for toroid functionalization. Moreover, V-70, a low temperature initiator, was used given that the thermal instability of xanthate terminal end groups of poly(vinyl amides) has been previously reported.⁴⁹ Following purification of the polymer, analytical gel permeation chromatography indicated that the poly(N-vinylformamide) was synthesized with a molecular weight of 9.0 kDa and polydispersity (Đ) of 1.239.

Scheme 2.

Synthesis of Poly(N-vinylformamide), Deposition onto Microtoroid and Hydrolysis to Afford Polyvinylamine

Functionalization protocol (four steps)

1. Microtoroid fabrication A 150 μm-diameter silica microdisk was patterned by standard photolithography and etching, then reflowed with a CO₂ laser to yield a smooth, bare microtoroid.

2. Silanization with APTES The chip was oxygen-plasma cleaned for 1 min, immersed in 1 % (v/v) (3-aminopropyl)triethoxysilane (APTES) in a glass vial for 10 min on a nutator, and rinsed with the same solvent to remove unbound silane.

3. Attachment of poly(N-vinylformamide) (PNVF) A 5 mM PNVF solution in deionized water was prepared. The silanized chip was submerged in this solution and incubated on a 40 °C hotplate for 2 h, then rinsed with water and dried under nitrogen.

4. Hydrolysis to polyvinylamine (PVAm) The chip was placed in 5 mM NaOH at 50 °C for 1 h in a glass vial to convert PNVF to PVAm, rinsed, dried, and baked at 60 °C for 30 min to remove residual solvent.

After these steps, the PVAm-functionalized microtoroid was ready for gas-sensing experiments.

Fig. 1a outlines the experimental platform. The apparatus comprises three modules: (i) an optical free-space-coupling path, (ii) a digital frequency-locking loop, and (iii) a trace-gas generation and delivery line.

Fig. 1. Experimental setup and device architecture of the free-space coupled FLOWER (Frequency Locked Optical Whispering Evanescent Resonator) gas sensing platform.

(a) Schematic of the free-space–coupled, frequency-locked microtoroid gas sensing system, comprising a microtoroid free-space coupling module, a frequency locking module, and a trace gas delivery system. The optical path includes a tunable laser, collimator, half-wave plate (HWP), non-polarizing beamsplitter (BS), tube lens, flippable mirror, 20× objective lens, CCD camera, and photodetector (PD). Red and blue beams indicate coupling-in and scattering paths, respectively. The frequency locking module connects to both the photodetector and the frequency modulation port of the laser, forming a feedback loop. The gas chamber housing the microtoroid was connected to the gas generator via two stainless steel ports; nitrogen (carrier gas) was flowed in and out, with a secondary dilution line branching from the outlet as indicated by a black arrow. (b) Photograph of the stainless steel gas chamber with the microtoroid mounted inside, shown alongside a U.S. quarter for scale. (c) Cross-sectional schematic of the microtoroid structure. The inset shows the simulated optical mode distribution with a corresponding electric field intensity color bar. (d) Schematic of free-space coupling to a microtoroid, shown with a Cartesian coordinate system. The red and blue beams indicate the coupling-in and scattering paths, respectively. The inset shows an image captured by a CCD camera in the yz-plane during microtoroid coupling. (e) Lorentzian fit of the resonance dip in the frequency domain, yielding an optical quality factor (Q) of 3.87 × 10⁷ post polymer coating.

Optical free-space coupling.

A tunable laser (Newport TLB-6712) was butt-coupled to an FC/APC adapter and a 50 mm plano-convex lens to form a collimator. The resulting beam (red in Fig. 1a) passed sequentially through a half-wave plate (HWP), a non-polarizing beam-splitter cube (BS), and a 20 × objective (NA = 0.4) before being focused onto the rim of the microtoroid through the front window of a hermetically sealed gas chamber. Resonantly scattered light (blue beam) was collected by the same objective, reflected at the BS, relayed by a tube lens, and directed by a flippable mirror either to a CCD camera for alignment or to a photodetector (PD) for signal acquisition. The objective–tube-lens pair formed an afocal imaging system that projected the toroid’s side view onto the CCD (Fig. 1d). When this image was in sharp focus, the toroid lay precisely in the objective focal plane, maximizing coupling efficiency. Fine adjustment of the HWP and toroid position while scanning the laser wavelength produced the characteristic Lorentzian resonance profile shown in Fig. 1e.

Frequency locking.

The PD output and the laser’s frequency-modulation port were connected to a digital locking module. A small dither signal applied to the laser produced a modulated transmission; multiplication with the dither and time-averaging yielded an error signal proportional to the first derivative of the resonance. A PID controller drove this error to zero, thereby locking the laser wavelength to the resonance center. The locked wavelength was streamed to a data-acquisition unit for real-time tracking of resonance shifts.

Trace-gas delivery.

Trace-gas concentrations of 2-CEES were generated using a high-precision gas standards generator (FlexStream^™, KIN-TEK Analytical, schematized as the glass bottle in Fig. 1a) equipped with a certified permeation tube (Trace Source^™, KIN-TEK Analytical). The permeation tube, maintained at a constant temperature of 30 ± 0.01 °C, provides a NIST-traceable emission rate calibrated via gravimetric weight loss. To achieve parts-per-trillion (ppt) concentrations, the primary emission mixed with a 1000 sccm nitrogen carrier gas and was further processed through a secondary-dilution manifold, allowing for total dilution ratios exceeding 10,000:1. The system utilizes internal mass flow controllers with an ‘as-left’ certification accuracy of ±1.5% of the reading, resulting in a total estimated concentration uncertainty of approximately 2%.

The gas was delivered to a compact, coin-sized stainless-steel chamber (Fig. 1b) holding the microtoroid chip and a thermistor for temperature compensation. To suppress specular reflections, the front window of the chamber was tilted at 66.8°. Gas lines entered and exited through Swagelok fittings at opposite ends, with the outflow routed to a scrubber. This robust calibration and housing framework ensures the accuracy of the ultra-low (part-per-trillion) detection limits while enabling stable, field-ready operation.

Results and Discussion

Using the free-space–coupled, frequency-locked microtoroid platform, we tracked resonance shifts of a polyvinylamine-coated toroid during alternating exposures to nitrogen and 2-CEES vapor. Nitrogen at 1,000 sccm was flowed through the sealed chamber for 10 min to purge the system, followed by 10 min of 2-CEES carried in the same nitrogen stream. This 20-min sequence constituted one measurement cycle. Linear fits were applied to the blank and analyte periods to extract wavelength shift rates (shown as grey and red traces, respectively, in Fig. 2a). Across four successive cycles with concentrations ranging from 25 to 200 ppt, we observed a monotonic blue shift upon 2-CEES exposure, and the shift rate increased linearly with concentration (Fig. 2b). Extrapolation of the calibration curve yielded a theoretical limit of detection of ~17 ppt, nearly three orders of magnitude below the previous best report (Table 1). Notably, we observed only trace-level sensing using polyvinylamine coatings with molecular weights of 4.37 kDa and 20 kDa, respectively, under exposures ranging from 25 ppt to 200 ppt (Fig. S1). We propose that the surface density of the polymer brush on the microtoroid is influenced by the chain length of the polymer and impacts sensing ability. Detection of 2-CEES at higher, parts-per-billion (ppb) concentrations is demonstrated in Fig. S1.

Fig. 2. Trace-level detection of 2-CEES vapor using the FLOWER platform.

(a) Real-time resonance shift (Δλ) of the frequency-locked microtoroid is shown as a thin blue line. Grey and red segments represent linear fits to the nitrogen and 2-CEES exposure periods, respectively. Δλ is defined relative to the initial resonance wavelength. Each 10-minute exposure corresponds to a specific 2-CEES concentration, as labeled. (b) Response slopes extracted from (a) are plotted as red dots versus 2-CEES concentration and fitted with a red dashed line. The linear regression equation is shown in the figure. The black dashed line indicates the baseline noise level; its intersection with the fitted curve was used to estimate the theoretical limit of detection (LOD).

Table 1.

Comparison of 2-CEES Trace Gas Sensors: Limits of Detection and Operating Temperatures

Sensing method	Experimental LOD	Theoretical LOD	Operating temperature	Reference
COF coated quartz crystal microbalance	5.6 ppm	0.96 ppm	RT	26
4-mercaptocoumarins fluorescent chemosensors	2.5 ppm	9 ppb	RT	20
Triazole AIEE fluorescent chemosensor	0.55 ppm		RT	21
ZIF–67 3D photonic crystals	0.5 ppm	16.5 ppb	RT	25
Al-doped ZnO quantum dot	0.5 ppm		450°C	7
Hydrogen-bond acidic polymer-coated SAW sensor	0.2 ppm	2 ppb	RT	18
Au-ZnFe2O4 semiconductor sensor	0.1 ppm	<0.1ppm	250°C	8
WO3/WS2 heterostructures	0.1 ppm	<0.1 ppm	240°C	9
Au nanoparticles decorated 3D porous graphene	0.1 ppm	5.8 ppb	80°C	12
F@Zr-BTC fluorescent probe	50 ppb	48 ppb	RT	22
Hierarchical Fe2O3 nanotube arrays	30 ppb		170°C	10
WO3 porous thin film	15 ppb		260°C	11
Surface-enhanced Raman scattering spectroscopy	10 ppb		RT	28
Free space FLOWER system	25 ppt	17 ppt	RT	This work

^*RT represents room temperature.

The quantitative response of the sensor is characterized by the rate of change of the resonance frequency, rather than a static total shift. Because the reaction between the 2-CEES vapor and the polymer brush is a covalent, irreversible alkylation process, the analyte continuously accumulates on the resonator surface during exposure. The FLOWER platform’s continuous frequency-locking allows us to track this accumulation in real-time, where the resulting slope of the resonance shift is directly proportional to the gas concentration. This kinetic sensing metric is particularly advantageous for trace-level detection, as it allows for rapid quantification (within seconds of exposure) and provides inherent immunity to slow environmental baseline drifts that might otherwise obscure a total shift measurement.

Because the sensing mechanism relies on irreversible covalent reaction between 2-CEES and nucleophilic binding sites within the polymer, the total number of available reactive sites is finite. As a result, the device operates in an accumulative mode, where signal increases with cumulative exposure until the available binding capacity is approached. In the early-time regime relevant to low-concentration detection (as shown in Figure 2), the number of occupied sites remains small compared to the total capacity, and the response rate is approximately proportional to analyte flux. Under prolonged exposure or high cumulative dose, gradual saturation of binding sites would reduce the incremental response rate. For the intended deployment as a passive, leave-behind exposure indicator, detection occurs well before significant site depletion, and full regeneration is not required.

To evaluate selectivity, the functionalized microtoroid was exposed to other hazard gases, including DIMP (diisopropyl methylphosphonate) and DMMP (dimethyl methylphosphonate), as well as DES (diethyl sulfide), which is structurally similar to 2-CEES, following the same procedure used for 2-CEES exposure. Two response cycles were recorded for each analyte. As shown in Fig. 3a–c, the microtoroid exhibited a pronounced response to 2-CEES at 200 ppt, while showing minimal response to the comparison vapors, confirming high chemical selectivity toward 2-CEES.

Fig. 3. Selectivity of the FLOWER platform for 2-CEES vapor.

(a–c) Real-time resonance shift (Δλ) measurements during exposure to DIMP (a) DMMP, and (b) DES, (c) each shown at 100 ppt and 200 ppt. Blue lines represent the raw FLOWER signal, while red and grey segments denote linear fits to the 2-CEES exposure and nitrogen baseline periods, respectively. Δλ is plotted relative to the initial resonance wavelength. (d) A control experiment using a PNVF-coated microtoroid without undergoing the hydrolysis protocol is conducted to assess the response to 200 ppt 2-CEES. The sensor shows negligible response, confirming that PVAm is necessary to enable sensitivity to 2-CEES. (e) Bar chart comparing the sensor's response slopes at 200 ppt for 2-CEES, DIMP, DMMP, and DES. The sensor exhibited a strong negative response to 2-CEES, while responses to the comparison vapors were negligible, confirming high chemical selectivity for 2-CEES at this concentration. Error bars on the CEES response represent the standard deviation of three repeated measurements from independent experiments.

We attribute the sensor’s selectivity for 2-CEES to the specific reaction pathway involving the intramolecular formation of a highly reactive, cyclic thiorane (episulfonium) intermediate. While DIMP and DMMP are categorized as electrophiles, and are notably more electrophilic than common laboratory confounders such as ethyl acetate, they yielded negligible shifts in the resonant frequency. This is because the thiorane intermediate formed by 2-CEES is a significantly more potent alkylating agent than the phosphorus centers in DIMP or DMMP under ambient conditions. The nucleophilic primary amines of the PVAm polymer brush effectively “trap” this intermediate through a covalent ring-opening reaction, leading to an irreversible change in the polymer’s refractive index and thickness.

The minimal response given by the thioether analog, DES, supports this hypothesis because it does not contain a chloroethyl group and cannot form the reactive thiorane intermediate, making it inert to reaction with PVAm. In contrast to general electrophilicity or molecular polarity, this specific intermediate-forming capability ensures that the FLOWER–polymer platform remains highly specific to 2-CEES. A control experiment using a PNVF-coated microtoroid without hydrolysis, shown in Fig. 3d, was conducted at the same 2-CEES concentration (200 ppt) and exhibited a minimal response. Fig. 3e presents a bar chart comparing the sensor responses to various vapors at 200 ppt, highlighting the negligible signal from the control and the high selectivity toward 2-CEES.

In the present work, our objective is to establish the intrinsic sensitivity limits and mechanistic selectivity of the nucleophilic polymer toward 2-CEES under controlled conditions. For real-world deployment in humid environments, environmental robustness can be achieved through several complementary strategies, including the use of sorbent chemistries specifically engineered for humidity tolerance (as demonstrated in our prior work)⁴⁰, upstream flow conditioning to regulate moisture levels prior to entry into the sensing chamber, and differential measurement using a reference resonator to compensate for bulk refractive index fluctuations. Importantly, these approaches do not require modification of the underlying optical cavity platform.

The vibration resilience of the FLOWER sensing system was evaluated by dropping a 450 g weighted beanbag from heights of 20 cm, 40 cm, and 60 cm onto the optical workstation (VIS3660-PG4-325A), simulating mechanical shocks representative of real-world field conditions (Fig. 4e). Upon impact, the microtoroid visibly oscillated in the top-view camera, with displacements exceeding its own diameter, and the resonance signal temporarily disappeared from the photodetector output (Supplementary Video 1). With frequency locking enabled, each shock event produced a transient but trackable shift in the resonance wavelength. Real-time tracking over a 4-minute interval, during which the bag was dropped at 1-minute intervals, revealed discrete resonance shifts associated with each impact (Fig. 4a). Zoomed-in views of the drops (Figs. 4b–d) showed maximum shifts of ~1.5 pm, followed by rapid return to baseline. The raw laser wavelength with dithering is shown in blue, and the resonance shift (median filtered) in red. These results confirm that the free-space–coupled FLOWER system maintains stable optical coupling and reliable frequency locking even under significant mechanical disturbance.

Fig. 4. Mechanical shock response of the FLOWER sensing system.

(a) Real-time resonance tracking over a 4-minute period during which a 450 g weighted beanbag was dropped onto the optical workstation at 1-minute intervals from heights of 20 cm, 40 cm, and 60 cm. (b–d) Zoomed-in views of the resonance shifts corresponding to each drop, showing transient spikes followed by full recovery. Blue lines represent raw laser wavelength data; red lines show median-filtered results. (e) Photograph of the free-space–coupled microtoroid system during a shock test, with the yellow weighted beanbag positioned above the optical table.

In gas sensing experiments, changes in test gas composition or concentration can alter the refractive index of the surrounding medium, potentially deviating the optical beam or modifying the optical path length. Such effects may compromise the accuracy and continuity of resonance tracking. Based on our previously reported results⁴¹, the microtoroid resonator, when free-space coupled under typical conditions, exhibits an efficient coupling zone approximately 10.9 μm × 2.7 μm × 3.8 μm in the x, y, and z directions (as shown in Fig. 1d), respectively, at ≥50% of maximum coupling efficiency. However, refractive index variations in the gas phase can shift the beam’s spatial alignment, potentially moving it outside the optimal coupling zone. This misalignment perturbs the microtoroid’s optical mode intensity and introduces artifacts in the measured resonance shift. This phenomenon is further explored in the following section through both simulation and experimental validation.

The microtoroid was free-space coupled using a focused paraxial beam (Fig. S3). The beam was directed through a 20× objective lens and passed through the tilted front window of the gas chamber, a 0.3 mm thick glass plate, before being focused near the coupling region of the microtoroid. To assess the impact of gas composition on beam alignment, we performed ray-tracing simulations in which the refractive index (RI) of the gas inside the chamber was varied to model focal point shifts. This phenomenon is shown in Supplementary Video 2. Owing to the planar symmetry of the optical system with respect to the xz-plane, changes in RI did not induce beam deviation along the y-axis. To simulate RI-induced beam displacement, we used the Lorentz–Lorenz equation to compute the refractive index as a function of 2-CEES vapor volume fraction. The relationship is given by⁵⁰:

$$\frac{n_{mix}^2-1}{n_{mix}^2+2}=\sum _i\left(\frac{n_i^2-1}{n_i^2+2}\cdot \frac{V_i}{V}\right)$$

Here, n_mix, n_i, and V_i/V represent the refractive index of the gas mixture, the refractive index of pure component i under the same conditions, and the volume fraction of component i in the mixture, respectively. Based on the Lorentz–Lorenz equation and ray-tracing simulations (Fig. S2), the focal point deviations along the x- and y-axes (Δx and Δy) were calculated to be 1.25 nm and 0.44 nm, respectively, at a 2-CEES concentration of 1 part per million (ppm). At 1 part per thousand (ppt-thou), Δx and Δy increased to 1.31 μm and 0.46 μm, respectively, both of which remain significantly smaller than the previously reported free-space coupling zone dimensions. This indicates that the coupling remains stable even in the presence of significant RI variation.

To experimentally evaluate the potential wavelength shift caused by such RI-induced beam deviations, we scanned the microtoroid along the optical axis using a nanopositioner while keeping the free-space beam fixed as shown in Fig. S4(a). The resonance spectra were recorded at each position. The normalized resonance power (P_res) and the corresponding resonance wavelength shift at each position are shown in Fig. S4(b–c). The coupling profile’s full width at half maximum (FWHM) was 11.2 μm, in agreement with previous results⁴¹. Notably, the resonance wavelength remained nearly constant over the entire scan range, suggesting that the refractive index–induced beam displacement leads to negligible additional wavelength shift. These results indicate that the free-space coupled FLOWER system exhibits strong tolerance to refractive index–induced beam perturbations, maintaining stable optical alignment across a wide dynamic range of 2-CEES concentrations, from low ppt to over one part per hundred.

Conclusion

In summary, we present a frequency-locked, free-space–coupled microtoroid sensor for ultra-sensitive (part-per-trillion) detection of volatile thioethers. As a representative example, we demonstrate detection of the mustard-gas simulant 2-CEES at concentrations as low as 25 ppt, with a theoretical limit of detection of 17 ppt, nearly three orders of magnitude below the best previously reported value. As a representative demonstration, we used the mustard-gas simulant 2-CEES, achieving ultra-sensitive detection at room temperature and atmospheric pressure. The sensor combines high sensitivity with mechanical robustness by eliminating fragile fiber tapers and housing the chip in a sealed, coin-sized gas chamber. Polyvinylamine functionalization confers strong chemical selectivity through nucleophilic trapping, while comparative exposures to DIMP, DMMP, and DES produced minimal responses. Ultra-high Q (Q > 10⁷) factors were maintained even after polymer coating, due to operation in the undercoupled regime. Unlike earlier WGM gas-sensing experiments that demanded constant operator supervision, our platform remains locked and aligned even after repeated mechanical shocks (for example, weighted objects dropped onto the optical table). Users can therefore “set and forget” the device, enabling unattended, long-duration monitoring that was previously impractical. These results establish a generalizable, ultra-sensitive, and rugged sensing architecture for compact, passive optical systems that require neither complex signal amplification nor vacuum operation. The approach is positioned for portable, field-deployable sensors targeting chemical-warfare agents, toxic industrial chemicals, and environmental pollutants, with future integration into scalable photonic packaging and wireless readout technologies enabling real-time deployment in battlefield monitoring, environmental compliance, and industrial safety.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.