Terahertz (THz) waves have unique advantages, such as low photonic energy, good penetrability, and abundant spectrum band, and attract extensive research in THz sensing [1–4]. The state-of-the-art importance is to promote biological analysis by providing vibrational information complementary to traditional spectroscopy [1,3,5,6]. The frequency of collective behaviors (vibration and rotation) for biomolecules, such as carbohydrates [7–10] and proteins [11–13], lies in the THz band mostly. The substance could be identified by analyzing the THz absorption spectrum. However, the direct interaction between THz waves and the substance is restricted by the mismatch of the absorption cross section, which reduces the detection sensitivity and accuracy. Metasurfaces [2,4,14–22], metal microstrip lines [23–27], and silicon waveguide structures [28–30] have been reported to compress the THz mode and enhance the interaction to improve the sensing performance. Metasurfaces enhance detection capabilities, but they make it challenging to recover absorption spectra restricted by the large free spectral range (FSR). The waveguide structures recover the absorption spectrum successfully, but the improvement in detection sensitivity is limited. Therefore, there is an urgent desire for a novel fingerprint spectrum detection scheme that combines enhanced ability and fully recovering absorption spectra characteristics.

The whispering gallery mode resonator (WGMR) possesses the advantages of miniaturization and a high Q factor [31–44]. THz waves fulfill accumulation in the resonator; therefore, the electric fields in the resonator are multiple times higher than those in the waveguide. Consequently, benefiting from the stronger interaction between THz waves and the substances, the THz-WGMR could achieve better sensing performance. The THz-WGMR has been reported in humidity sensing [37], leaf-hydration monitoring [38], crystal water detection [39], and particle sensing [40] based on outstanding sensing ability. These works demonstrate the advantages of WGMR with the high Q value, but most are separate devices and are mainly based on refractive index sensing. To our knowledge, no integrated WGMR is currently applied in vibrational absorption spectroscopy, which is an essential index in THz biomedical sensing.

In this work, we first studied the application of silicon WGMR for measuring fingerprint absorption characteristics. The classic lactose was applied, and we demonstrated the high sensitivity through the high-Q WGMR. Thanks to its compact periodicity, the envelope of the fingerprint spectrum was reproduced fully. As a comparison, glucose showed no fingerprint during measuring the frequency band to illustrate the effect of spectral recovery. We also compared the WGMR and the straight waveguide to demonstrate the state-of-the-art performance of the proposed platform. Considering the effective spectral recovering ability and promising sensitivity, the proposed sensing platform in our work will improve the accuracy of THz biomedical identification and promote the development of THz sensing.

Generally, different substances are distinguished by the difference in refractive index, including the difference in frequency shift caused by the real part of the refractive index and the difference in transmissivity caused by the imaginary part [37,39,40]. As mentioned above, in the fingerprint identification of substances, the absorption coefficient of substances is related to the frequency, and specific identification can be carried out by measuring the characteristic absorption spectrum of substances. In the analysis of absorption characteristics, the Beer–Lambert law is a classic tool and is usually applied as [45] Iout=Iine−αL,(1)Tw=10lg(IoutIin)∝−αL,(2)where α is the absorption coefficient, L is the interaction length, Iin and Iout are the input and output intensity of THz waves, and Tw is the transmissivity. According to the formula, for a straight waveguide, the Tw is linear with the interaction length, and the sensitivity is restricted by the absorption coefficient. In high-sensitivity sensing, the absorption coefficient is almost tiny. Therefore, it usually requires a long interaction length to fulfill a detectable change, which restricts the applications for tiny substance identification. However, it is usually possible to reduce the interaction length by allowing THz waves to propagate and interact with the substance multiple times by a resonant structure. The Q value generally manifests this improvement. WGMR, with extremely high Q, can significantly improve sensitivity, whose structure is shown in Fig. 1(a).

The WGMR consists of a coupling waveguide and a resonator. The substance covers the resonator as depicted in the inset. The thicknesses of ridge H1 and slab H2 are 60 μm, respectively, and the width of the waveguide is 300 μm to guarantee the single-mode transmission. The radius R of the resonator is 5 mm, which is a trade-off between the transmission loss and the bend loss to promote a high Q. And the gap g is 50 μm to promote a slightly under-coupling state, maintaining comparatively excellent sensing performance. The substances interact with the top and side faces of the ridge. The resonator fulfills the multiple interactions between THz waves and the substance, strengthening the sensing sensitivity. For a WGMR, the transfer matrix method is applied to analyze the transmission characteristics. The intensity transmissivity and phase transmissivity are concluded as [31] a=a0exp(−αL),(3)Tr=r2+a2−2arcosφ1+r2a2−2arcosφ,(4)ϕ=π+φ+arctanrsinφa−rcosφ+arctanarsinφ1−arcosφ,(5)respectively, where α represents the absorption coefficient resulting from the substance, L represents the interaction length of the substance, r represents the self-coupling coefficient between the waveguide and the resonator, which is approximately 0.99, a is the single-pass amplitude transmission coefficient of the resonator, and φ represents the single-pass phase shift of the resonator. The numerical results could be solved according to the above formulas.

As a demonstration, the fingerprint absorption characteristic of lactose is taken as an example. A simulation of the WGMR for lactose sensing is performed and depicted in Fig. 1(b), with a frequency range from 500 GHz to 550 GHz. The extinction ratios (ERs) of the modes around 532 GHz decrease. Because the absorption peak of the lactose is located at 532 GHz and the full width at half-maximum (FWHM) is approximately 21 GHz [7], there is an absorption dip in the single-pass amplitude transmission coefficient of the resonator, as illustrated in Fig. 1(d). In the simulation, the absorption coefficient of the lactose is 6mm−1, and the height and width of the lactose are 200 μm and 600 μm, respectively. The lactose covers the waveguide, and the interaction length along the waveguide is 1 mm. When unloaded, the single-pass amplitude transmission coefficient a is a constant, and the spectrum of the WGMR should be uniform as in Fig. 1(c). It could be considered that the absorption of the lactose modulates the modes of the WGMR. Thanks to the large length of the THz-WGMR, its FSR is approximately 2.6 GHz according to FSR=c2nπR,(6)in which c is the speed of light, n is the mode refractive index, and R is the radius of the resonator. The FSR of the THz-WGMR is much smaller than the FWHM of the absorption peak of lactose, thus enabling the full recovery of its fingerprint absorption spectrum. The sensing sensitivity is not only related to the absorption α of the substance but also related to the constants a and r of the WGMR. The values of these constants are calculated using the finite element method [46]. Figure 1(e) illustrates the normalized electric field distribution. The substance affects the effective mode refractive index by interacting with the evanescent field of the ridge waveguide. The transmissivity of the WGM closest to 532 GHz and the straight waveguide at 532 GHz for lactose sensing is calculated and depicted in Fig. 1(f), respectively. The horizontal axis means the interaction length of the lactose along the resonator and the straight waveguide. The vertical axis represents the transmissivity of the WGMR and the straight waveguide after lactose loading. For the WGMR, with the increase of the length, the transmissivity increases rapidly at first and then reaches saturation as depicted by the blue curve. With the increase in the interaction length, the absorption loss increases; therefore, THz waves are difficult to store in the resonator. As a result, the resonance disappears and finally reaches saturation. For the waveguide, the transmissivity has a linear correlation with the interaction length as depicted by the red line, and the sensitivity is 0.22 dB/mm. Noteworthily, the transmissivity of the WGMR changes significantly around 0 mm, which is 10.7 dB/mm; therefore, the WGMR possesses a state-of-the-art response to a tiny absorption in the substance. This simulation proves that the resonator possesses high sensitivity for tiny absorption owing to the enhanced interaction with the substance. Thanks to the narrow FSR, the absorption spectrum is recovered fully.

To demonstrate this capability, we fabricated a high-Q WGMR based on CMOS-compatible technology for fingerprint spectra measurement. The processes are to spin the photoresist (AZ9260) on the silicon wafer (resistivity greater than 10kΩ·cm), then expose it to transfer the pattern, and finally carry out deep etching (ICP Oxford Plasmalab System 100 ICP108). The experiment setup is illustrated in Fig. 2(a). Frequency-domain spectroscopy (Toptica TeraScan 1550) is applied to measure the THz spectrum. The maximum measuring range is 0–1.22 THz, and the dynamic range is 80 dB. The smallest frequency step of the system is 1 MHz, which satisfies measuring requirements. THz waves are generated and emitted at the emitter by beating two lasers. After the collimating lens, THz waves are collimated into parallel rays. Then, after the focusing lens, THz waves are collected into the horn antenna, which depresses the THz fields into 254μm×500μm. Next, THz waves are coupled into the THz-WGMR from the cross section of the straight coupling waveguide, and the single coupling loss is approximately 2.9 dB. THz waves interact with the lactose at the WGMR and then pass through the following lenses. Finally, THz waves are received by the THz receiver. The α-lactose monohydrate powder and glucose powder (Aladdin Co.) are applied in the process to accomplish the absorption spectrum recovery.

The measured intensity and phase spectra of the THz-WGMR are depicted in Fig. 2(c). The black circles are the experimental results, and the red curves are the simulated results according to Eqs. (4) and (5). The resonant mode is located at 531.92 GHz, in which the ER and the FWHM are 9.5 dB and 27 MHz, respectively; therefore, the Q factor of the resonant mode reaches approximately 19,700. To the best of our knowledge, the obtained Q factor is the highest value in the integrated THz-WGMR reported so far, benefiting from the optimization of the design and manufacture. According to the phase spectrum, the phase shift at 531.92 GHz is 0.3π, demonstrating that the THz-WGMR works in the under-coupling state, which means the coupling loss resulting from the waveguide is weak. Figure 1(e) demonstrates that the electric fields are mainly located inside the waveguide, reducing the scattering loss from the roughness of the side wall. Therefore, it promotes the fulfillment of the extremely high Q factor. The Q factor can be further improved by introducing a wider waveguide to decrease the scattering loss, together with the Euler-bent structure to decrease the bend loss [47,48].

The intensity transmission spectrum of the THz-WGMR loaded with lactose is shown in Fig. 3(a). There are 19 resonant modes in the frequency range of 500–550 GHz, and the FSR is 2.58 GHz, which satisfies the Nyquist sampling theorem; therefore the absorption spectrum of the lactose can be depicted fully. There is an absorption envelope of around 532 GHz when loaded with lactose. As a comparison, the measuring results for glucose show no such absorption envelope. From the experimental results, the longer the interaction length of the lactose, the greater the decrease in ER around 532 GHz. When the length approaches 4 mm, the ER reaches saturation. The experiment demonstrates the response of the WGMR for the lactose and proves the effectiveness of the WGMR for THz absorption spectroscopy measurement. The apparent change in ER proves the excellent sensitivity of this method. Noteworthily, with the fantastic response ability of the ER for the substance, the absorption envelope of the lactose around 532 GHz is apparent to identify. The absorption peak of the glucose is located at 1.4 THz, far from 532 GHz; therefore, the absorption coefficients are approximately equal in the measured frequency range. Furthermore, the ERs of all modes decrease overall with the increase of glucose. It is reasonable because the increase of glucose leads to a small increase in the absorption loss and, hence, a decrease in the single-pass transmission coefficient a, which causes the ER to decrease consequently.

To quantitatively describe the detection results, the transmissivities of resonant modes at 532 GHz and 534 GHz, near the absorption peak, are picked to record the response of the WGMR for the different substances, and the results are depicted in Figs. 3(c) and 3(d), respectively. For the lactose, the transmissivity increases from −9.5dB rapidly around 0 mm, and saturates when approaching 4 mm. Thanks to the high-Q characteristics of the WGMR, the sensitivity around 0 mm is as high as 8.42 dB/mm. However, for the glucose, the transmissivity is almost unchanged. For the length of 4 mm, the change of the transmissivity for the lactose is 7.8 dB, which is far greater than 1.3 dB from the glucose. The significantly different change in the transmissivity demonstrates the ability of the WGMR to identify the absorption spectrum, and the absorption spectrum of the lactose is distinguished clearly. Noteworthily, the original transmissivity is −9.5dB, which determines the maximum response of the WGMR. The maximum response could be further improved by optimizing the coupling state of the WGMR to decrease the original mode transmissivity.

To demonstrate the high sensitivity of the WGMR directly, the straight waveguide is also performed for fingerprint spectrum measurement. As illustrated in Fig. 4(a), to form an observable envelope, the interaction length of the lactose should exceed 4 mm at least. When the interaction length reaches 20 mm, there is an apparent absorption envelope around 532 GHz. Consistent with the analysis, there is no absorption envelope for glucose in Fig. 4(b). Figures 4(c) and 4(d) demonstrate the transmissivity at 532 GHz of the straight waveguide for lactose and glucose sensing, respectively, and the decreases of the transmissivity are 1.4 dB and 0.1 dB, respectively. The experimental results demonstrate that the straight waveguide could recover the fingerprint spectrum successfully but requires sufficient substances, restricted by the low sensing sensitivity. The value of 1.4 dB is much smaller than that of the WGMR, which is 7.8 dB for measuring lactose, demonstrating the high sensitivity of the WGMR.

This proposed sensing method based on the WGMR may be a competitive alternative for weak absorption peak measurement and promote the development of the THz substance identification technology.

We compare the reported works and our work; as illustrated in Table 1, our work fulfills significant response for tiny substances, and the FSR is narrow enough to recover the fingerprint absorption spectrum.Table 1.

Comparison between the Reported Sensing Schemes and This Work

The metasurface and WGMR possess relatively high sensitivity for different reasons. WGMR possesses a strong electric field due to a high Q. The metasurface possesses a strong electric field due to a small mode volume. The metal line and waveguide possess wide-band characteristics. The working band of the WGMR is 500–550 GHz limited by the metal antennas’ bandwidth. To broaden the working band, a wider band of the measuring system should be introduced. Second, a wider band WGMR can be designed. By specific design in the coupling between the waveguide and the resonator, the number of supported modes should increase. The significant improvement in our work lies in taking advantage of the high sensitivity of the WGMR and optimized FSR to recover the spectrum. In application, the proposed integrated WGMR and corresponding measurement scheme should promote the accurate identification of the fingerprint absorption spectrum in tiny substances and contribute to qualitative and quantitative identification.

In this paper, an enhanced measurement scheme in vibrational absorption spectroscopy based on integrated THz-WGMR is manifested. The fingerprint absorption spectra of the lactose and glucose are significantly distinguished, and the responses are 7.8 dB and 1.3 dB, respectively. Compared with a straight waveguide, the state-of-the-art sensitivity of the WGMR is demonstrated. The responses of straight waveguide for 4 mm lactose and glucose are only 1.4 dB and 0.1 dB, respectively. We believe the proposed sensing method will improve the sensitivity of vibrational absorption spectroscopy measurement and promote THz applications in biomedical substance identification.