On chip glucose sensing using guided waves at terahertz frequencies Authors: Mohsen Haghighat, Thomas Darcie, Levi Smith Published: 2024 ABSTRACT This paper demonstrates an on-chip anhydrous D-glucose sensor using a coplanar stripline (CPS) on a thin (1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m) silicon nitride membrane at terahertz (THz) frequencies. A thin layer (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx$$\end{document} 10 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m) of D-glucose was placed in close proximity to the CPS and the transmission response was measured using a modified THz-TDS setup. The D-glucose introduces frequency-dependent changes to the effective permittivity of the CPS resulting in a modified spectral response at the receiver. Measurement results show absorption signatures at 1.42 THz and 2.07 THz corresponding to the first two significant resonances beyond 1 THz for D-glucose allowing for label-free detection. The frequency-dependent attenuation coefficient was estimated by simulation for several D-glucose layer thicknesses using a modified Lorentz model. Measurement results align with simulations and other literature that use free-space THz radiation. This work verifies on-chip THz sensing of D-glucose and presents a pathway toward on-chip sensing of other materials at THz frequencies. Introduction Glucose detection is important across various industries, including the food, beverage, and medical sectors1. The escalating consumption of sugar in diets has been linked to chronic health issues, notably cardiovascular diseases, diabetes, and obesity2,3. Given the importance of these health issues, accurate detection and monitoring of glucose both in the human body and food become imperative. In response to this need, researchers have been developing glucose-sensing technologies for blood glucose monitoring in the last couple of decades4,5. Also, industries such as food, beverage, and fermentation manufacturing are required to shrink and control sugar levels while ensuring the quality and safety of their products1. This necessitates continuous monitoring and measurement of various samples, including sugars, throughout the manufacturing process and in the final product. On the other hand, selective detection of glucose is essential for investigation in the food and beverage industry because natural glucose should not be entirely replaced with artificial sweeteners since they have other health side-effects such as sugar craving and mood disorders6. The necessity to either control excessive sugar levels or detect natural sugars from artificial sweeteners motivates researchers to explore advanced sugar-sensing methodologies. D-glucose is one of the main sugars that exist in food and the common naturally occurring simple sugar7. In the context of this paper, the term ‘glucose’ exclusively denotes anhydrous D-glucose8. The pursuit of glucose sensing methods aligns with the broader objective of promoting health and well-being, either glucose detection in food, or blood glucose monitoring. Glucose sensing methods have been reviewed in4,5,9 from traditional enzymatic and electrochemical methods to optical and spectroscopic techniques. Regarding in-body glucose sensing, the category of optical sensors is favorable for their non-invasive nature and potential for continuous monitoring. These methods include fluorescence, Raman spectroscopy, near-infrared (NIR), mid-infrared (MIR), and far-infrared (FIR) spectroscopy5. However, optical methods face challenges due to scattering4,9,10 with reduced scattering coefficient of 30–50 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {cm}^{-1}$$\end{document}11, contributing to signal-to-noise ratio (SNR) reduction12, relatively weak glucose absorption peaks especially in NIR range13, and interference from surrounding light1. On the other hand, THz waves have less scattering compared to IR and optical beams10,12 with estimated reduced scattering coefficient up to 1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {cm}^{-1}$$\end{document} based on analysis in11, and have strong glucose absorption peaks8,14 which enables easier glucose detection with higher signla-to-noise ratios9. Moreover, THz waves penetrate into human tissue up to 300 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m15, potentially reaching the dermis layer where blood and glucose are present whereas the MIR range only penetrates up to 100 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m, which corresponds to the thickness of the epidermis layer that lacks blood13. However, glucose can still be detected in the MIR and THz range within the interstitial fluid (ISF) of the epidermis, where a slight concentration of glucose is present but blood capillaries are not reached at this depth13. Absorption coefficients extracted from THz time-domain spectroscopy (THz-TDS) measurements have shown a high sensitivity to the glucose level in blood samples16. Additionally, THz-TDS combined with weak value amplification was demonstrated to be sensitive to small changes in concentrations of glucose both in solid and liquid samples17,18. However, challenges remain in terms of miniaturization and cost-effectiveness of THz sensors5, which can be addressed by on-chip THz sensors, that are the focus of this research. Glucose absorbs electromagnetic radiation at certain frequencies in the THz band. The first peak of absorption spectra for glucose beyond 1 THz occurs in the range of 1.40–1.44 THz8,14,19–22. In literature, the majority of glucose sensing methods at THz frequencies involve directing a free-space THz beam towards a thin glucose sample8, a metasurface20,21,23–25, or nano-antennas (i.e., also a form of metasurface)14 coated by sugar (glucose) then detecting the transmitted wave. Using metasurfaces results in strong subwavelength field concentration at a resonance frequency, making them sensitive to small features and slight variations in the nearby environment, thereby enhancing their overall sensitivity23. In24,25 authors designed metasurface unit cells that have resonance frequencies below 0.5 THz and then measured the shift in resonance frequency in the presence of glucose or other materials. Others14,20 have designed the metasurface unit cell so that it targets the first THz-band absorption frequency of glucose near 1.4 THz. In14 the authors proposed a metasurface for THz transmission by nano-slot-antennas which increases the molecular absorption cross-sections, thereby enabling the detection of sugar molecules at low concentrations. Table 1 provides an overview of various THz sensing methods for sugar detection including key details such as the reference source, resonance frequency in THz, frequency range covered by the sensor, structural design, mechanism of wave transmission, spectral detection method employed, and the type of sugar targeted in the respective study. Our work, highlighted as ‘This Work’ in Table 1, proposes on-chip glucose sensing operating in the broad frequency range of 0.1–2.2 THz, employing guided waves which as previously mentioned, is the key difference from the other works. The aforementioned studies use free-space THz radiation for sensing which offers simplicity and flexibility in accommodating various sample geometries and configurations; however, free-space methods suffer from path loss and require bulky THz optical components26. Alternatively, guided-wave methods are attractive because they are compact and do not require the alignment of a THz beam. Also, a guided-wave sensor can offer lower channel loss26 and increased sensitivity by confining the electromagnetic field within the sample region making it suitable for on-chip integrated sensors in various applications27. Additionally, guided-wave methods are more resilient to sample roughness and heterogeneity owing to the guided propagation of electromagnetic waves. There are several studies that use guided waves for sensing at THz frequencies. In28 the authors reported high-loss liquid sensing including water using a coplanar waveguide (CPW) up to 1.1 THz where the signal decays below the noise floor. A similar approach is taken with microstrip lines (MSL) in29 for liquid sensing (i.e., water), and authors mentioned that the experimental results for frequencies greater than 1 THz are close to the noise floor, although spectral data is reported up to 1.5 THz. This is mainly due to the high loss of MSL configuration at THz frequencies and also partly because of the thickness of the waveguide cover layer29. Similarly, a proof-of-concept on-chip planar Goubau line, integrated with a microfluidic channel is presented in30 to measure the permittivity of liquids up to 0.8 THz and thereby sensing different types of alcohols. The aforementioned studies demonstrate experimental measurements up-to 1.5 THz which is suitable for many applications, but does not address the need to resolve signatures at 1.42 THz and 2.07 THz for label-free glucose sensing. Additionally, there are other THz dielectric waveguide approaches for sensing reported in31,32 that do not use of metallic conductors. Though these approaches mitigate ohmic losses, they operate at sub-1 THz frequencies and are relatively narrow band (fractional bandwidths < 45 %) making them unsuitable for glucose detection. This work explores a unique approach for on-chip glucose sensing using guided waves via coplanar stripline (CPS) on a thin silicon nitride (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document}) substrate which presents a platform that mitigates the limitations of standard THz waveguide systems for sensing applications. A CPS is commonly used in conjunction with photoconductive switches since they are compatible with two-terminal devices which allows for excitation and detection using the sliding-contact method33. Also, the CPS on a thin membrane enables the interaction of the electromagnetic field with materials (such as glucose) beneath the membrane (see Fig. 1). Our experiments show the prominent THz-band absorption frequency of 1.42 THz measured with the fabricated sensor, which aligns with the absorption frequency reported in Table 1. We note that label-free sensing involves the detection of the multiple characteristic absorption signatures which occur at different frequencies and the measurement of a broad spectrum facilitates this capability. To confirm label-free sensing, we simultaneously resolve the glucose absorption features at 1.42 THz and 2.07 THz.8,19.Figure 1Illustration of glucose sensing by CPS waveguide on a thin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document} membrane.Table 1Summary of experimental sugar sensing methods based on THz spectroscopy.ReferencesResonance freq. (THz)Freq. range (THz)StructureWave transmissionSpectral detection methodSugar type81.43, 2.080.3–3.0NA (direct exposure)Free-SpaceAbsorption frequencyD-glucose141.4, 1.70.5–2.5Nano-slot antennasFree-SpaceAbsorption frequencyD-glucose, Fructose, Sucrose201.431.2–1.6MetasurfaceFree-SpaceAbsorption frequencyD-glucose211.4351.2–1.6MetasurfaceFree-SpaceAbsorption frequencyD-glucose240.260.2–0.5MetasurfaceFree-SpaceResonance shiftD-glucose250.320.2–0.5MetasurfaceFree-SpaceResonance shiftD-glucose340.60.3–1.2Meta-material absorberFree-SpaceResonance shiftD-glucoseThis work1.42, 2.070.1–2.2CPSGuided-WaveAbsorption frequencyD-glucose Illustration of glucose sensing by CPS waveguide on a thin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document} membrane. Table 1Summary of experimental sugar sensing methods based on THz spectroscopy.ReferencesResonance freq. (THz)Freq. range (THz)StructureWave transmissionSpectral detection methodSugar type81.43, 2.080.3–3.0NA (direct exposure)Free-SpaceAbsorption frequencyD-glucose141.4, 1.70.5–2.5Nano-slot antennasFree-SpaceAbsorption frequencyD-glucose, Fructose, Sucrose201.431.2–1.6MetasurfaceFree-SpaceAbsorption frequencyD-glucose211.4351.2–1.6MetasurfaceFree-SpaceAbsorption frequencyD-glucose240.260.2–0.5MetasurfaceFree-SpaceResonance shiftD-glucose250.320.2–0.5MetasurfaceFree-SpaceResonance shiftD-glucose340.60.3–1.2Meta-material absorberFree-SpaceResonance shiftD-glucoseThis work1.42, 2.070.1–2.2CPSGuided-WaveAbsorption frequencyD-glucose Summary of experimental sugar sensing methods based on THz spectroscopy. Results Design, modeling, and simulation The sensor presented in this work operates on the principle of frequency-dependent dielectric loss from a nearby material (i.e., glucose). Figure 2 illustrates a cross-section of the sensor geometry which is a CPS transmission line (S = 130 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m and W = 30 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m) on a suspended thin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document} membrane (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_1$$\end{document} = 1 µm) which is in contact with a thin glucose layer (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_2$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx$$\end{document} 10 µm). The effective permittivity of the propagating wave, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon _{\text {eff}}$$\end{document}, depends on the individual material parameters (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon _{\text {0,1,2}}, ~\mu _{\text {0,1,2}},~\text {and}~\sigma _{\text {0,1,2}}$$\end{document}) and the dimensions of the transmission line and substrate (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S, W, H_{1,2}$$\end{document}). Here, of primary concern, is detecting frequency-dependent variations in glucose permittivity, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon _{\text {2}} (\omega )$$\end{document}, which can be achieved when the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document} layer is thin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_1 \approx$$\end{document} 1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m.Figure 2CPS sensor cross section. CPS sensor cross section. The CPS sensor structure is fabricated by depositing a 1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document} layer on a 500 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m silicon substrate. Next, a rectangle of silicon with a dimension of 2 mm \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document} 10 mm is etched under the structure where the sample material will be placed. Metallic conductors are placed using photolithography and metal deposition (Ti/Au: 10 nm/200 nm). The overall design of the is similar to our prior work35. The transmitter and receiver are photoconductive switches (PCSs) (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{dark} \approx 50 \text {M}\Omega$$\end{document}) that are placed onto the CPSs to provide a low frequency bias for the transmitter and a method to sample the transmitted electric field at the receiver. The transmitter is directly connect to a DC voltage supply (24V). The receiver is directly connected to the current input of a lock-in amplifier (1 kHz reference to optical chopper, 300 ms integration time). Note that care must be taken to eliminate cavities, i.e, the bias lines bend to radiate undesired signals and mitigate resonances, and the receiving PCS is placed at a junction to ensure the reflections occurring from the separate conductor and substrate discontinues are not resolved. In this study, we incorporate a frequency-dependent permittivity model for glucose21 into the existing model of a CPS that is depicted in Fig. 2. Our objective is to obtain the attenuation coefficient of the CPS for glucose layers of varying thicknesses for frequencies ranging from 0.5 to 2.5 THz. The simulation was performed using ANSYS HFSS 2023 and characterizes the CPS behavior in the presence of glucose at different frequencies and thicknesses. The material properties of the model components include the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document} substrate with 1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m thickness and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon _r$$\end{document} = 7.6, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{Si_3N_4}=0$$\end{document}, tan \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _e$$\end{document} = 0.0052636, and gold conductors with 200 nm thickness and the conductivity of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{Au} = 4.1 \times 10^7$$\end{document} S/m. The dielectric permittivity of the glucose layer with single resonance at THz frequencies was previously modeled using the modified Lorentz model, which is widely used for modeling a single resonance in material’s permittivity, as presented in21. Our work includes the second resonance of glucose to the glucose Lorentz model as observed in14 and also our experiments. Equation (1) shows the relation of this permittivity model with two resonances:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\varepsilon }(\omega ) = \varepsilon _\infty + \frac{\omega _p^2}{\omega _{0,1}^2 - \omega ^2 + i\gamma _1\omega } + \frac{\omega _p^2}{\omega _{0,2}^2 - \omega ^2 + i\gamma _2\omega } \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon _\infty = 3.231$$\end{document} is the dielectric coefficient at infinite frequency21, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega _{0,1} = 2\pi (1.42 \, \text {THz}$$\end{document}) is the first resonance frequency beyond 1 THz, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega _{0,2} = 2\pi (2.07 \, \text {THz}$$\end{document}) is the first resonance frequency, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega _p = 2\pi (0.11 \, \text {THz}$$\end{document}) is bulk plasma frequency of the dielectric for the thin film of glucose, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _1 = 2\pi (0.04 \, \text {THz}$$\end{document}) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _2 = 2\pi (0.08 \, \text {THz}$$\end{document}) represents the corresponding damping constant at the resonance frequencies. Next, we simulated the attenuation coefficient for the CPS illustrated in Fig. 3a with a glucose layer (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon _{\text {2}}$$\end{document} in Fig. 2) using (1) for several different glucose thicknesses (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {H}_2$$\end{document} = 5, 10, 15 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m). The results of this simulation are plotted in Fig. 3b along with a reference without any glucose which has an attenuation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx$$\end{document} 0.6 dB/mm at 1.4 THz. As previously mentioned, the average thickness of glucose is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx$$\end{document} 10 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m based on the average of the profile plot in Fig. 6 measured by a profilometer along the top CPS line depicted on the inset. According to Fig. 3b this thickness corresponds to an attenuation depth of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx$$\end{document} 4.5 dB/mm which is the height of the absorption peak at 1.42 THz and attenuation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx$$\end{document} 2.2 dB/mm at 2.07 THz. As observed in Fig. 6, there are step-like discontinuities in the thickness of the glucose layer. This nonideality will be mitigated in future works; however, the effect is investigated here. To quantify the impact, we performed full-wave simulations using ANSYS HFSS to obtain the S-parameters for step discontinuities for several glucose layer thicknesses (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {H}_2$$\end{document} = 0, 5, 10, 15, 20 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m). The simulation data is found in the Supplementary Material S1. For a glucose layer thickness of 10 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m at 1.4 THz we found that the insertion loss is less than 0.7 dB and the return loss is greater than 22 dB which ensures the discontinuities will not drastically degrade the sensor performance.Figure 3(a) CPS on the thin membrane with glucose layer. (b) Attenuation of CPS with modified Lorentz model for glucose layer with different thicknesses under the sensor’s thin membrane. (a) CPS on the thin membrane with glucose layer. (b) Attenuation of CPS with modified Lorentz model for glucose layer with different thicknesses under the sensor’s thin membrane. Experimental results and discussion The measurement results of unloaded (no glucose) and loaded sensor (with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx$$\end{document} 10 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m glucose layer) are shown in Fig. 4. Note that the spectral amplitude decay with increasing frequency in the experimental results is an expected consequence of applying a THz finite duration pulse as an input signal rather than an ideal THz impulse with a flat spectral response37. From the simulations (Fig. 3b), the depth of peak absorption should be approximately 3 mm \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document} 4.5 dB/mm = 13.5 dB which is observed in measurement results (spectral response in Fig. 4). We find our experimental peak absorption frequency (1.42 THz) lies within the range (1.40–1.435 THz) found in other literature (see Table 1). The frequency-dependent absorption observed in experimental results demonstrates the sensor functionality. It is important to recognize that this work is primarily a proof-of-concept and that future methods will be developed to improve sensitivity through the optimization of sample distribution methodology and circuit topology. We note that glucose shows broad absorption starting from 1.10 THz up to around 1.28 THz8,14,19. This is observed in Fig. 4 as a minor depression before the main peak at 1.42 THz. Moreover, another THz absorption peak of glucose at 2.07 THz is observed in the measurement results and is expected to occur at 2.05–2.10 THz8,14,19,20. We predict the attenuation of the dip at 2.07 THz relative to the attenuation at 1.42 THz using the data from14 which results in an expected attenuation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx$$\end{document} 5.8-6.6 dB at 2.07 THz. As shown in Fig. 4 the experimental results find a depth of absorption peak is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx$$\end{document} 6.2 dB which is within reasonable agreement with the expected value. It is important to consider practical aspects of sensors. The broadband nature of the received spectrum has several benefits which can reduce the probability of false positives and false negatives. First, the absorption at 1.42 THz can be measured relative to the adjacent region of the spectrum (i.e., 1.5 THz), but this method would be susceptible to false positives since other materials, such as sucrose, exhibit weak absorption characteristics near 1.4 THz. However, since the presented sensor is capable of resolving broad spectral information that corresponds to two signatures for glucose (1.42 THz and 2.07 THz) it becomes straightforward to reduce the probability of false positives by verifying that both signatures exist simultaneously with the appropriate relative magnitudes. Next, the sensor can exhibit false negatives if the glucose film is poorly distributed away from the CPS. Our proof-of-concept sensor is susceptible to this issue [see Fig. 5b] due to our current drop-coating method. After a precise method for uniformly distributing the glucose layer is developed, then it will become possible to determine the thickness of the glucose layer (without using a profilometer) by measuring the magnitude of attenuation at the absorption peaks. Also, we have begun investigating alternative microfluidic methods to address this issue and to develop a pathway towards measuring glucose level monitoring (in blood or aqueous solution) - both of which are beyond the scope of this work.Figure 4Measurement results of the proposed CPS-based glucose sensor. Measurement results of the proposed CPS-based glucose sensor. Conclusion An on-chip glucose sensor is demonstrated using guided waves at THz frequencies on a thin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document} membrane. The sensor is based on the interaction between the propagating electromagnetic waves at THz frequencies and a thin layer of glucose placed in close proximity to a transmission line which results in frequency-dependent notches in the received spectrum. Our experiments reveal a resonance at 1.42 THz and 2.07 THz which aligns with findings in existing literature. Our approach stands out for its guided-wave nature and on-chip microfluidic integration potential. This work demonstrates pathways towards on-chip label-free sensing of glucose and other materials with applications such as the food industry and healthcare. Methods Sample preparation We made a mixture of glucose and isopropanol (5 g: 50 mL) and sonicated that in a beaker to reduce the grain size. Then, we placed drops of the liquid using a micropipette in steps of 10 µL on the backside of the membrane and waited for the isopropanol to evaporate. Figure 5 illustrates microscopic and close-up photos of the sensor with and without glucose on the backside of the membrane. The thickness of the resulting glucose layer was measured using a profilometer (Fig. 6) after placing 60 µL of the mix. It is important to note that with our current simple drop placement methodology precise control of the glucose distribution was not possible. Future work will address this limitation by the introduction of lithographically-defined microfluidic channels. For this specific experiment, the glucose film thickness is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_2 \approx$$\end{document} 10 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu$$\end{document}m, and the glucose interaction length is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \approx$$\end{document} 3 mm (Fig. 5b). We note that to avoid absorption saturation, a thin sample layer is desired8.Figure 5Sensor and membrane pictures. (a) Clean (unloaded) membrane back side. (b) Dehydrated glucose on the back side membrane (loaded sensor). (c) Fabricated wafer. (d) glucose grains on a glass after 10 min of sonication with isopropanol. (e) CPS sensor windows (wafer backside). (f) Membrane front: CPS side microscope image. (g) Membrane back: Material side microscope image. (h) Membrane back image with dried drop-coated glucose with a micropipette.Figure 6Thickness profile of the glucose layer, and back side of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document} membrane with non-uniform drop-coated glucose (inset). Sensor and membrane pictures. (a) Clean (unloaded) membrane back side. (b) Dehydrated glucose on the back side membrane (loaded sensor). (c) Fabricated wafer. (d) glucose grains on a glass after 10 min of sonication with isopropanol. (e) CPS sensor windows (wafer backside). (f) Membrane front: CPS side microscope image. (g) Membrane back: Material side microscope image. (h) Membrane back image with dried drop-coated glucose with a micropipette. Thickness profile of the glucose layer, and back side of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {Si}_3\hbox {N}_4$$\end{document} membrane with non-uniform drop-coated glucose (inset). Experimental setup The experimental measurements were carried out using a modified THz-TDS setup that is in Fig. 737,38. A femtosecond pulsed laser with a wavelength of 780 nm and average output power of 20 mW was split and directed towards thin-film low-temperature grown gallium arsenide (LT-GaAs) photo-conductive switches (PCS) placed on both sides of the tapered CPS to generate and detect a broadband THz pulse signal37. The fabrication of the LT-GaAs PCS devices is detailed in38,39. The transmitted signal was reconstructed by photo-conductive sampling of the output of the THz pulse, using the mechanical delay line and measuring the receiver current using a lock-in amplifier, similar to conventional THz-TDS techniques40. Other specific details about the experiment are the optical chopping frequency of 1 KHz and the bias voltage of transmitting PCS of 24 Volts.Figure 7Measurement setup picture and diagram for glucose sensing. Measurement setup picture and diagram for glucose sensing. Supplementary Information Supplementary Information 1. Supplementary Information 2. Supplementary Information 1. Supplementary Information 2.