Silicon photonics, mainly referred to at the near-infrared (NIR) wavelength, have attracted great research interest in the past decade because they are a potential candidate to meet the increasing demands for high data transmission capacity in communication systems. One key driving force of silicon photonics is the fabrication compatibility with the matured CMOS technology, which has the ability to provide low-cost photonic integrated circuits (PIC) for a high volume of production. Due to the worldwide efforts fueled by researchers in the past years, silicon photonic technology working at NIR wavelength range around 1.31 and 1.55 μm for communications is now moving toward the industry from the academy.

In fact, the NIR is not only the wavelength range where silicon photonics have an impact. As shown in Fig. 1, the silicon material has the low absorption loss up to the wavelength of ∼8.5μm ^[1–3], which extends to the mid-infrared (MIR) region. This range covers the sensing wavelengths of a set of gases ^[4], including CO (∼4.5μm), CO2 (2.65 μm, 4.2–4.3 μm), CH4 (3.2–3.45 μm), etc. Some other toxic gases such as HF and H2S also have the absorption peaks within this range (2.33–2.78 μm for HF, 2.5–2.75 μm for H2S). The real-time monitoring of these gases in the environment and during the industrial process demands low cost and compact sensors. Silicon photonics stands out as a promising candidate for such application because it can provide an on-chip solution, which is cost effective and highly compact. Through the appropriate design of the optical waveguide cross-section, the evanescent field of the optical mode could overlap with the surrounding materials and consequently induce absorption loss of the light. The gas concentration can be then extracted from the output optical power detected at the end of the waveguide. Over the past few years, low-propagation-loss silicon MIR waveguides have been demonstrated by several research groups on the silicon-on-insulator (SOI) platform for the wavelength ranging from 2 to 6 μm ^[5–13]. In general, microring resonators (MRRs) are employed to improve the sensitivity and reduce the footprint. The light transmits in the ring with several rounds, thus essentially increasing the effective interaction lengths between the light and the detected materials. The MIR MRRs on the SOI platform were implemented and characterized in 3–5 μm ^[8,9,12,14]. The Vernier effect, which is constructed by MRRs for further improvement of the sensor performance at the NIR wavelength ^[15,16], has transferred to the MIR range by scaling up the device dimension ^[17,18]. The high transmission in the atmospheric window of 3–5 μm ^[2] also guarantees the application of silicon photonics in MIR free space communications and light detection and ranging (LIDAR) systems. Compared with the visible light and NIR waveguides, the large MIR waveguide dimension helps to improve the phase array performance used in the LIDAR system ^[19–23] because it alleviates the requirement of the fabrication precision as well as the phase error induced by the sidewall roughness. Moreover, the MIR lasers are safer for humans compared with their counterparts at the visible light range, as the MIR light is unable to penetrate cornea and skin tissues ^[24]. In addition, in the MIR range of 2–6.5 μm, the two- and three-photon absorption in silicon is negligible, permitting higher-power density transmission than at the telecom wavelengths. On the other hand, higher optical power density in the waveguides or cavities benefits nonlinearity phenomenon generation ^[25,26], which is utilized to explore the MIR nonlinearity devices on a silicon platform, such as the wavelength conversion ^[27], parametric oscillator ^[28,29], amplifier ^[30], and frequency comb generator ^[31]. Actually, a lot of molecules have unique identifiable absorption spectra in the MIR range of 8–20 μm ^[2]. It is worth to develop the on-chip photonic biosensors in this “fingerprint region.” However, beyond the wavelength of 8 μm, silicon loses its competitiveness because its absorption loss drastically increases. Given the large transparency window of 2–14 μm, germanium is a suitable material for the MIR wavelength beyond 8 μm. The MIR waveguides are demonstrated in various germanium-based platforms, such as germanium-on-silicon (GOS) ^[32,33], germanium-on-SOI (GOI) ^[34–36], and germanium-on-silicon nitride (GOSN) ^[37]. Other platforms including silicon-on-silicon nitride (SOSN) ^[38], sapphire-on-silicon (SOS) ^[39,40], SiGe alloy-on-silicon (SGOS) ^[41], and aluminum nitride-on-insulator (ANOI) ^[42] are also explored to demonstrate the MIR photonics devices. The fabrication technology of these platforms is compatible with that of silicon photonics, while most of them use the silicon as cladding or core layers for the waveguides. Thus, in the following discussion, we consider they are silicon photonics technology in a broad sense.

With regard to the functionality, several MIR passive devices on different platforms have been demonstrated, including the fiber-chip grating couplers ^[5,43–46], multimode interferometers (MMIs) ^[6,33,47,48], Mach–Zehnder interferometers (MZIs) ^[49,50], MRRs ^{[8,9,12,14,17,18,51–53]}, photonic crystals (PhCs) ^[54–56], de/multiplexers ^[10,57–60], polarization splitter and rotator (PSR) ^[61], and polarization beam splitter (PBS) ^[62]. These pioneering works open the way to MIR photonic integration. Besides the passive devices, the active devices, i.e., modulators, lasers, and photodetectors (PDs), are needed to compose a complete photonic system. In the reported work, the free carrier dispersion effect (FCDE) and Pockels electro-optic effect are employed to implement the modulators at short MIR wavelengths ^[61–67]. However, the device performance is far away from its counterparts working at NIR wavelengths. Meanwhile, because of the inherent limitation of the materials used in the aforementioned platforms, the MIR lasing and detection remain as challenges for the silicon photonics. The most promising solution for MIR laser sources is to integrate the III–V quantum well (QW) lasers and quantum cascade lasers (QCLs) ^[68–76] to the silicon photonics platforms. While at the detection side, the detection structures based on GeSn alloy or III–V material are monolithically integrated to the silicon substrate to build up the MIR PDs ^[77–80]. The detection spectral range covers the short- and middle-wave infrared.

Inspired by the success of silicon photonics at NIR, a silicon platform is considered as the most attractive solution for MIR photonics because it has potential to provide the hybrid integration system with low cost by the CMOS pilot lines. In the next sections of this paper, we will review the reported work and introduce IME’s efforts on MIR photonics as follows. Section 2 presents and summarizes the researchers’ elaboration upon the MIR silicon photonics devices. The material limitation and choice of integration platform are discussed in this section as well. Section 3 introduces IME’s research and development efforts of MIR platforms and device library in SOI and SiN platforms. Section 4 deploys the CO2 gas sensor as an example to demonstrate the application of the integrated MIR photonic circuit. Finally, we provide our perspective on potential future research topics and engineering challenges.

In pursuit of the optical waveguide with low propagation loss, various platforms and geometric structures were demonstrated. The reported works are summarized in Table 1. As can be seen, the SOI-based waveguides realize low loss below 2dB/cm with the wavelength up to 4 μm by proper design of the cross-section ^{[5–8,11,12]}. With regards to the fabrication compatibility, SOI is the most preferred platform for short MIR range (λ<4μm). According to the low-loss MIR waveguides demonstrated in Refs. ^[6,13], the SOI wafer with 340 nm top silicon and 2 μm buried oxide (BOX) is suitable for working around the wavelength of 2 μm, while the one with 400 nm top silicon and 3 μm BOX is applicable for the wavelength around 3.75 μm. Because these wafers are the commercially available products, they may become the standard platforms for the two interesting wavelengths. Beyond 4 μm, the SiO2 becomes so lossy that other platforms are employed. Chang et al. first demonstrated the low-loss strip waveguide for the single TM mode at 5.8 μm on the GOS platform ^[32]. However, silicon has a large thermal-optic coefficient that causes the device performance sensitive to the temperature fluctuation. Therefore, GOI is adopted as the platform to fabricate the MIR waveguides because the BOX in the SOI substrate can relieve this problem as an isolation layer. Younis et al. recently reported the GOI waveguides at 3.682 μm, with a loss of ∼8dB/cm ^[34]. In their following work, the rapid thermal annealing process was used to reduce the loss ^[35]. Apart from a germanium-based platform, due to the wide transparent range of 1.2–6.7 μm, SiN is a promising material for MIR. The GOSN and SOSN platforms implemented by the wafer bonding provide more choices for the MIR waveguides ^[37,38]. However, the measured loss is not as low as expected. The authors state that the loss can be further reduced by improving the sidewall quality with advanced processes ^[37]. Researchers also demonstrated the MIR waveguides with SOS platform because of the low loss in sapphire below 5 μm ^[39,40]. But there are not obvious merits compared with other platforms, in aspects of the low loss range and the fabrication compatibility. Another attractive platform is the SiGe alloy on silicon substrate. Though controlling the ratio of silicon and germanium in the vertical direction, Brun et al. realized the low-loss SGOS waveguide at 4.5 and 7.4 μm ^[41]. From the work summarized in Table 1, one can find that the research tendency is toward longer wavelengths, and the germanium MIR waveguide is becoming a hot spot.

The edge coupling scheme requires a more complicated alignment system. One typical setup is to use a CCD camera to capture the input optical mode of the fiber and judge whether it is launched into the waveguide, while an alternative is to employ a visible laser to visualize the light coupling into the waveguide. However, currently, the lack of the lensed MIR fiber limits the light coupling efficiency between the fiber and the chip. In contrary, the fiber-to-chip grating couplers provide an easier way for the light coupling. Hattasan et al. reported an SOI grating coupler operating in a short-wave infrared, which consists of a 220 nm thick crystalline silicon layer with a 160 nm poly-silicon overlay above ^[5]. The etched depth of the grating is 240 nm (80 nm into the crystal silicon layer). The coupling loss of 3.8dB/facet at the peak wavelength of 2.1 μm with a 3 dB bandwidth of 90 nm is obtained. A suspended structure on SOI platforms is considered as one of the solutions to reduce the propagation loss caused by the BOX absorption. The focusing grating coupler for such suspended waveguides was demonstrated by Cheng et al. on the SOI wafer with a 340 nm top silicon layer ^[43]. The measured coupling efficiency at the wavelength of 2.75 μm is 24.7%, corresponding to the coupling loss of 6.07dB/grating. They further designed the “fishbonelike” grating couplers to improve the coupling bandwidth ^[44]. The uniform grating coupler with SOS platform also was experimentally demonstrated by the same group at the wavelength of 2.75 μm with the coupling efficiencies of 32.6% (4.4dB/facet) and 11.6% (9.3dB/facet) for TE mode and TM mode, respectively ^[45]. For longer wavelength at 3.8 μm, Alonso-Ramos et al. first demonstrated a single-etch GOS grating coupler with an inversely tapered access stage ^[46]. Though the coupling efficiency of 7.9% (11dB/facet) is the highest one among the reported MIR GOS grating couplers, the coupling loss is still quite large. Among these jobs, the coupling loss of ∼3–4dB/facet around the working wavelength of 2–3 μm is close to those at the NIR range, which is efficient for the device characterization in the laboratory. But, for the practical application in products, the coupling loss needs to be further reduced. In this case, the grating couplers are not the best choice. Other coupling schemes, such as the suspended edge coupler, may be an alternative solution.

The MMIs are usually utilized as the light splitters to construct the optical hybrid and modulators/switches. It is an important and basic building block in the photonic circuit. A long-wave infrared 1×2 MMI splitter has been demonstrated by Wei et al. in 2010 ^[47]. The device adopted the undercut structure and was characterized at the wavelength of 10.06 μm. Though the optical power at two output ports keeps an acceptable difference (∼0.76dB), the insertion loss is too large (∼23dB) for the device application and integration. In 2012, Milošević et al. reported the 1×2 MMI splitters based on submicrometer SOI strip waveguides, with the insertion loss of 3.6±0.2dB/MMI measured at 3.74 μm ^[48]. Followed by this work, the same group implemented the 1×2 MMI splitters and 2×2 MMI couplers with the GOS rib waveguides working at 3.8 μm. Their measured insertion loss is 0.21±0.02 and 0.37±0.07dB/MMI, respectively. More recently, Rouifed et al. reported a 1×2 MMI splitter operating at 2 μm on the 340 nm SOI platform with an ultra-low insertion loss of 0.29±0.01dB ^[6]. The MZIs have been demonstrated with the SOI and GOS platforms for the wavelength of ∼3.8 and ∼5.3μm ^[49,50]. The design approach of these commonly used components is easily transferred from those at the NIR. However, there is still a large room for improvement of their performance.

The versatile applications of MRRs in optical filters, routers, (de)multiplexers, and modulators make them the most thoroughly studied passive photonic devices. The efforts made on the MIR MRRs working at the MIR range are summarized in Table 2. The working wavelength of the reported SOI-based MIR MRRs range from 2.75 to 5.2 μm. An ultra-high-quality (Q)-factor of ∼106 at the wavelength of 3.5–3.8 μm is achieved, with the extinction ratio (ER) of ∼6.6dB ^[8]. The ER increases with the stronger coupling between the bus and ring waveguides but at the sacrifice of the decreasing of Q-factors. As shown in Table 2, the SOI MRR demonstrated in Ref. ^[17] realizes a large ER around 25 dB, with a low Q-factor of 2900. The MRRs implemented with the platforms of GOS and SOS also encounter the same problems ^[18,51–53]. The MIR MRRs with high Q-factor and large ER are demanded for the optical signal processing. Although the MRRs-based MIR filters have been demonstrated at the working wavelength of 2.75–5.6 μm, there is no existing work demonstrating both high Q-factor and large ER. The performance of the MIR MRRs is still needed to be improved. Another type of microresonators frequently reported are PhC cavities. The typical L3 PhC cavity operating at 4–5 μm was demonstrated by Shankar et al. ^[54]. The Q-factor of 13,600 was obtained but without ER presented. In Shankar’s work, because there is no coupling in and out waveguides for the demonstrated L3 PhC cavity, it is not suitable for the integration with other devices. As an improvement, Zou et al. demonstrated the MIR L21 PhC cavity with a coupling bus waveguide beside ^[56], which gives the possibility to the device as an all-pass filter. Nonetheless, as seen in Table 2, the Q-factor is decreased to 3500 due to the additional cavity loss induced by the bus waveguide coupling. Generally, compared with PhC cavities, MRRs are easier to design and have less stringent requirements of the fabrication process. Thus, MRRs are more practical for the photonic circuit integration.

The (de)multiplexers are widely used in the wavelength-division multiplexing (WDM) system to increase the data transmission bandwidth, the function of which is the optical signal multiplexing and routing between different channels. In particular, the (de)multiplexers are used in the spectrometers as the core elements. The arrayed waveguide gratings (AWGs) and planar concave gratings (PCGs) are the representative (de)multiplexers. The MIR AWGs based on SOI platform operating at 3.8 μm were reported by Muneeb et al. in 2013 ^[10]. The device has six output ports with insertion loss and crosstalk of −1 and −25dB, respectively. The performance is close to the SOI AWGs at telecommunication wavelengths. The eight-port PCGs also was reported in the paper, with the insertion loss and crosstalk of 1.6 and −19.97dB, respectively. In the same year, Malik et al. reported the MIR AWGs and PCGs fabricated with the GOS platform ^[57,58]. The devices are working at a wavelength of around 5 μm. The channel crosstalk of both devices below −20dB is obtained for the TE mode. In 2015, Barritault et al. reported their AWGs based on the SGOS platform. The insertion loss of 5 dB and crosstalk <−20dB are achieved at the wavelength of around 4.5 μm by the optimized design ^[59]. Thus far, the AWGs and PCGs exhibit acceptable performance at the short-wavelength infrared. Besides these two (de)multiplexers, the angled MMI (AMMI) is another (de)multiplexing device. Such a five-channel GOS-based MIR AMMI with an insertion loss of 3 dB, crosstalk of −10dB, and channel spacing of 20 nm was demonstrated by Penades et al. in 2015 ^[60]. Though this device has a novel structure, its performance is currently not comparable with that of the AWGs and PCGs.

Silicon photonics face polarization problems caused by birefringence, which results in the control of polarizations in silicon photonics devices, are of great importance. The PSR and PBS are two key components used to manipulate the polarization. A large amount of work has been done and reported on these two kinds of devices. However, most of the demonstrated devices are working at the NIR range. The PSR and PBS, intentionally designed to operate at the short-wave infrared range, were reported by Hu et al. in 2016 ^[61,62]. The proposed PSR utilize a partially etched grating-assisted coupler on the SOI platform to realize the polarization splitting and rotation in the contra direction. The device shows a conversion efficiency of 96.83% with the conversion loss of −0.97dB at the wavelength of 2.5 μm, while maintaining the low polarization crosstalk of −21.48dB. The PBS is proposed by using a hybrid plasmonic Y-branch on the SOI platform. The operating bandwidth is 285 nm with the center operating wavelength of 2 μm. In fact, the operating principle and design method of the MIR PSR and PBS are the same with those working at telecom wavelengths. However, to the best of our knowledge, there is no experimental demonstration of MIR, PSR, and PBS thus far. One possible reason is that the setup for polarization characterization in MIR is not as straightforward as NIR.

A modulator plays a key role in PICs. To date, only a few works on MIR modulators have been reported. The MIR silicon modulators operating at the wavelength of 2–3 μm based on the FCDE have been reported in ^[63–65]. However, the modulation speed is neither measured nor comparable with modulators working at the telecom wavelengths ^[66,67]. The FCDE accompanies with large absorption loss, while the wavelength is beyond 2.5 μm, which hinders the application of this scheme to longer wavelength modulation. In order to realize the modulators working at longer wavelengths, the Pockels electro-optic effect in lithium-niobate (LiNbO3) and AlN are employed. Chiles and Fathpour implemented the modulators working at 3.39 μm with the silicon-on-LiNbO3-based MZI ^[68]. Unfortunately both the modulation frequency and the efficiency are quite low. Liu et al. demonstrated the DC modulation at λ=2.5μm with the AlN waveguides ^[69]. The effective index change of 2×10−5 is obtained by applying 15 V voltage to the device. More recently, a higher-frequency (∼15GHz) all-optical modulation in porous silicon was realized by Park et al. ^[70]. However, in terms of the construction of the NIR photonic systems, the all-optical modulation is found to be seldom used. Besides these modulation schemes, Nedeljkovic et al. fabricated and characterized the thermo-optic modulators on SOI that work at λ=3.8μm ^[71]. The inherent slow response, in the order of 10 kHz, of the thermo-optic effect limits such device for working in a free space communication system.

Silicon and germanium are inefficient material for optical light generation due to their indirect band gap. Similar to the proposed optical system prototype in the telecom wavelengths, III–V materials are utilized to fabricate the MIR lasers. The GaInAsSb/AlGaAsSb QW lasers have shown the excellent characteristics in 2–3 μm ^[72]. Using the quinary material (AlGaInAsSb) instead of the quaternary AlGasASb, the QW laser working wavelength could be extended to 3.4 μm ^[73,74]. The most promising solutions for longer MIR laser sources are the QCL. Multiwatt output power, continuous wave (CW), room temperature QCL operating in the wavelength ranging from 3 to 20 μm have been demonstrated during the past two decades ^[75–77]. More recently, some pioneer work of the MIR laser integration with the waveguides also have been reported ^[78–80]. In both the on-chip sensors and the LIDAR systems, the MIR PDs are the necessary components to receive the optical signals. At the short-wave infrared range, the GeSn alloy attracts a lot of interest to build up the detectors due to its widely tunable bandgap. The GeSn MIR PDs fabricated on silicon substrate were reported in Refs. ^[81,82]. The devices can work at the wavelength up to 2.3 μm with a responsivity of ∼0.1A/W. Wang et al. demonstrated the PbTe-based PDs at the wavelengths of 3.5 and 3.7 μm where the entire detector structure was directly deposited onto a silicon substrate ^[83,84]. Wu et al. recently reported the monolithically integrated InAs/GaAs quantum-dot MIR PDs on silicon substrate ^[85]. The detection range of the device is up to ∼8μm at 80 K. This work proves that the direct integration of III–V PDs to the silicon photonics circuit is feasible. However, the low cost and high yield approach to integrate these III–V components, including the lasers and PDs, with silicon photonic platform remains a challenge. The integration method demonstrated at telecom wavelengths may provide an example for the MIR range ^[86].

IME has been spearheading the R&D on silicon photonics over the past decade with various silicon photonics devices, products, and platform building. It is natural for us now to extend our work to the MIR range. The preliminary effort is to develop the appropriate platforms for the MIR passive device design and fabrication. In this section, we will introduce the basic MIR photonic components demonstrated in IME on the SOI and SNOI platform. The devices, including the strip/rib waveguides, bending waveguides, directional couplers (DCs), and MRRs, were characterized at the short-wave infrared of ∼3.7 and ∼2μm, respectively. The research efforts of ANOI based devices are in progress, and it is expected to have some exciting demonstrations in the near future.

The SOI device fabrication starts from a commercially available 8-in. SOI wafer with 220 nm top silicon layer and 3 μm BOX layer. Blanket silicon epitaxy is performed to increase the top silicon thickness to 400 nm. SiO2 is deposited as the hard mask for fine structure definition. The device is patterned by deep ultraviolet (DUV) photolithography followed by a two-step silicon reactive ion etching (RIE) to define the waveguide devices. The 3 μm cladding SiO2 is deposited using plasma-enhanced chemical vapor deposition (PECVD). Finally, a deep trench is formed for butt fiber coupling.

The optical and scanning electron microscope (SEM) images of the fabricated SOI devices are shown in Figs. 2(a)–2(i). To ensure a good optical mode confinement in the silicon core and maintain the single-mode condition at 3.70 μm for strip waveguide and 3.75 μm for rib waveguide, the SOI strip waveguide is designed with a cross-section of w×h=1.2μm×0.4μm, while the rib waveguide is 1.2 μm in width and has 0.24 μm etch depth. The SEM image of them are shown in Figs. 2(e) and 2(f), respectively. The light is coupled from a ZrF4 fiber to a waveguide taper tip with 0.2 μm width [Fig. 2(d)] and adiabatically tapered to the waveguide through 200 μm tapering length. Figure 3 presents the loss of an SOI waveguide measured by the cutback method. The measured propagation loss at 3.75 μm of the strip waveguide and rib waveguide is plotted in Figs. 3(a) and 3(c). The linear fitting results indicate the loss of the two type of waveguides is 2.65±0.08 and 1.75±0.22dB/cm, respectively, which is comparable with the reported best data. We also study the bending loss of bends with different bending radius. As shown in Fig. 3(b), the bending loss gradually decreases from 0.054 to 0.008 dB/90°, as the bending radius increases from 5 to 25 μm. The reduction of bending loss could be attributed to smaller mode mismatch, as the light travels from a straight to bend waveguide and lower radiation loss for a bend with a larger bending radius.

Based on the good propagation and bending loss result, we fabricated an SOI strip [Fig. 2(i)] and rib [Fig. 2(g)] DCs with gap g=500nm and varying coupling lengths Lc. The coupled mode theory (CMT) was applied to analyze the DCs enabled by evanescent wave coupling in previous studies. The coupling in the S bend region, which brings two far-apart waveguides close to allow for evanescent wave coupling, was considered negligible. Though this assumption might be valid in NIR, in MIR the optical mode is less confined, which will cause significant coupling in the S bend region. Due to this fact, we slightly modify the equation obtained previously by the CMT by adding an initial phase Ø0 to take coupling in the S bend region into account ^[13]: TI=t2=cos2(π2LcLπ+Ø0),(1)XI=K2=sin2(π2LcLπ+Ø0),(2)Y=arctan⁡(X/T)=(π2Lπ)Lc+Ø0.(3)In Eqs. (1) and (2), T is the power transmitted through the waveguide. X is the power coupled evanescently to the neighboring waveguide. I=X+T is the total power in the DCs. t is the power transmission coefficient. K is the power coupling coefficient. Lc is the coupling length of the straight coupling region. Lπ is the straight length to achieve π phase shift. Ø0 is the initial phase introduced by S bends. Equation (3) is derived from Eqs. (1) and (2). By linear fitting Y with respect to Lc, Lπ and Ø0 can be extracted. The characterization results of SOI strip and rib DCs are shown in Figs. 4(a) and 4(c). The 50% power splitting ratio is achieved at the coupling lengths of 21 and 8 μm, respectively. The sine squared fitting suggested by Eqs. (1) and (2) results in an adjusted R-square value of 0.997, showing the consistency between the data and our proposed model. Linear fitting results of Y with respect to Lc suggested by Eq. (3) are shown in Figs. 4(b) and 4(d) as insets. The extracted slope and intercept, which presents π2Lπ and Ø0, respectively, are substituted into Eq. (2) to obtain the power coupling coefficient K. Figure 4(d) reveals a K value of ∼0.4 at Lc=0, corresponding to the 16% coupling caused by the S bend region. This significant amount of coupling demonstrates that including Ø0 term in our proposed model is significant.

We also fabricate SOI strip racetrack MRRs with different coupling lengths and gaps. Figure 5(a) shows the transmission spectrum of the MRRs with the bending radius of 5 μm and the gap of 500 nm. The coupling length between the ring and the bus waveguide is 5 μm. The red solid dot is the detected optical power by the power meter. Limited by the laser scanning step of 0.5 nm, the CMT and transfer matrix method are used to fit the measured data. The blue line is the fitted transmission spectrum. The Q-factor of ∼5900, ER of ∼6.84, and free spectral range (FSR) of ∼85nm are extracted from the zoom in the spectrum shown in Fig. 5(b). The theoretical spectra for both MRRs fit the experimental data very well, where the group index is obtained by the Lumerical FDTD simulation. The dispersion relation of the effective index is expressed as neff(λ)=(−0.4839×106)*λ+3.9666. The coupling coefficient and round-trip loss are 0.225 and 0.981, respectively. The transmission spectrum of a rib racetrack MRRs is shown in Figs. 5(c) and 5(d), from which we see that the Q-factor of ∼4500, ER of ∼5.29, and FSR of ∼67.84nm. By the optimized physical parameters, the MRRs can be designed working close or even at the critical coupling condition to realize a high Q and a large ER simultaneously.

The SNOI device fabrication starts from a commercially available 8-inch silicon wafer. A 4 μm thick SiO2 is first deposited by PECVD as the insulating layer. 400 nm silicon nitride (Si3N4) is then deposited by PECVD as the device layer. The following fabrication steps for the device formation are similar to those of SOI devices, as mentioned above.

The SNOI strip waveguide is designed with a cross-section of w×h=1μm×0.6μm for application at 2 μm [Fig. 6(a) inset]. The light is coupled from a ZrF4 fiber to a waveguide taper tip with 0.3 μm width and adiabatically tapered to the waveguide through 200 μm tapering length. The loss characterization result of SNOI waveguide is shown in Fig. 6. The propagation loss of our SNOI waveguide is 2.52±0.10dB/cm at 2 μm. Figure 6(b) demonstrates the bending loss of an SNOI bend with various bending radius. For a 10 μm radius bend, the output power is below −40dBm, which is the threshold of the PD. Hence, the 10 μm radius bend is too lossy for light transmission. For 10 and 50 μm radius bend, the bending loss is 0.147±0.006 and 0.029±0.002dB/cm, respectively. Thus, for SNOI devices for MIR application, the bend radius should be large enough in order for less propagation loss.

The SNOI strip MRRs are fabricated as well [Fig. 2(k)]. Figure 7 shows the characterization result of our SNOI strip ring resonator with the radius of 20 μm and gap of 550 nm. As seen from the zoom-in spectrum in Fig. 7(b), a higher ER of 11.36 dB is obtained, but at the price of a low Q-factor around 3200. The trade-off between the Q and ER can be achieved through appropriate design of the MRRs dimension.

As mentioned in Section 1, the MIR photonics has the potential to realize the portable, cost-effective, and mass-produced integrated systems for applications such as on-chip sensors, free space communications, LIDAR, and thermal imaging. In this section, we will discuss the CO2 gas sensors as an example to illustrate the applications of the integrated MIR photonics. An on-chip gas sensor is usually composed of three parts: the laser diode, analyte interaction component (absorption of the optical evanescent field), and PD. The hybrid integration technology for MIR laser and PDs are discussed in Section 2.G, whereas here we focus on the analyte interaction part.

The detection limit and sensitivity are the most important performance metrics in the gas sensor. In order to enhance these metrics, choosing the high absorption wavelength and increasing the effective interaction optical path are the commonly used methods. In the MIR region, the absorption coefficient of the pure concentration CO2 is larger than 80cm−1 at the MIR wavelength around 4.3 μm ^[87]. This strong absorption gives the ability to achieve the high sensitivity sensing of CO2 with the MIR photonics. On the other hand, the approaches to increase the effective interaction optical path can be classified into two categories. One is to design the true optical length structures such as the spiral waveguides ^[88]. The other is to use the resonator, such as MRRs, to guide the light with multiple passes ^[89].

A concept of the MIR photonics-based CO2 gas sensor is shown in Fig. 8(a). A polydimethylsiloxane (PDMS) gas chamber with the inlet/outlet is adhered to the optical spiral waveguide area to form the analyte interaction region. The waveguide cross-section is designed with a confinement factor Γ. The optical transmission power can be evaluated by the Beer–Lambert Law ^[88]: where αg is the absorption coefficient of 100% CO2 (partial pressure 1 atm), C is the volume concentration of the CO2, and L is the waveguide length, which interacts to the analyte gas. Assuming Γ is 0.6 and the absorption coefficient is linearly proportional to the gas concentration (80cm−1 for 100% CO2). The calculated power variation as the function of the CO2 concentration is shown in Fig. 8(b). The detected CO2 concentration is around 500 ppm for a 50% optical power variation in a 30 or 40 cm spiral waveguide, while for the same power variation, the detection concentration is 1200 ppm for a 10 cm long spiral waveguide. Assuming the optical power launched into the spiral waveguide is 0 dBm, the optical propagation loss in the waveguide is ∼1.5dB/cm, and the detection limit of the PDs is −90dBm, the spiral waveguide of 40 cm is feasible to undergo a 30 dB optical power variation (100%–0.1%) for gas sensor. As shown in Fig. 8(b), for the 40 cm long spiral waveguide, the detection concentration reaches 110 ppm with a 20% (∼0.97dB) power variation. The sensitivity near 200 ppm is calculated as ∼8.4×10−3dB/ppm from Fig. 8(b). It increases as the CO2 concentration becomes lower, which indicates that the sensor has the detection limit below 20 ppm, as such CO2 concentration change would induce a readable optical power variation of ∼0.17dB at the detector. If the propagation loss can be further reduced and a longer spiral waveguide is used, the detection limit can be improved to be below 10 ppm.

In the MRRs, the light with the resonant wavelength propagates for several rounds. This allows the MRRs to provide an effective long path for the optical evanescent field to interact with the analyte gases. Figure 8(c) shows the configuration of MRRs absorption enhanced gas sensor. The gas absorption will induce the change of resonance line shape in transmission spectrum. Through the inspection of the line shape change, i.e., monitoring the change of Q-factor or ER, the concentration change of analyte gas can be measured ^[89]. In terms of the CMT, the Q-factor of the MRR is expressed as ^[90] where Qi is the intrinsic quality factor of the of the MRR, while Qc is the coupling quality factor between the bus and the ring waveguide. Qcav is the combination quality factor of Qi and Qc. The analyte gas absorption loss contributes the quality factor of Qg, which is descripted as where ω, λ, and τp are the angular frequency, vacuum wavelength, and photon life time of the light resonant in the MRR. In terms of Eq. (5), when Qcav is comparable or larger than Qg, the considerable influence of the gas absorption on the total Q will be observed. The relationship between the Q-factor variation and the CO2 concentration is plotted in Fig. 8(d). As can be seen, for the MRR with an original Q-factor of 105, taking the CO2 (concentration of ∼800ppm) absorption into account, the 20% variation of Q-factor is inspected. The same Q variation of the MMR with Q-factor of 104 can only detect the CO2 concentration of ∼8000ppm. The sensitivity near 200 ppm is around 3.43×10−4/ppm for the MRR with Q of 105. Assuming that 2% variation of Q can be observed in the detection analysis, the estimated detection limit of such MRR is around 58 ppm. Obviously, the MMRs with higher Q have the potential capability to detect the CO2 with concentration below 10 ppm. However, it is a challenge to demonstrate the MRR with Q over ∼106 on silicon photonic platform. In fact, the high Q MRR increases the effective interaction length between the optical evanescent fields and the detection gas, which can be calculated by where c is the light speed in the vacuum. The Leff varying with the CO2 concentration is shown in Fig. 8(d) as well. The effective optical transmission length under the 200 ppm CO2 for the MRR with original Q of 105 is around 6.42 cm. Assuming that the MRR radius is in the order of several 10s μm, the footprint is <0.031mm2. In contrast, realizing the comparable performance, the spiral waveguide composed of multiple bend waveguides with radius ranging from 10 to 1000 μm is needed, the footprint of which could be as large as ∼100mm2. The footprint of MRR is thus 1000 times smaller, while the detection limit could be comparable. However, one of the disadvantages of the MRR-based gas sensor is the inevitable use of the wavelength scanning method. To adopt the wavelength scanning method, the sensor must be equipped with a tunable MIR laser in order to obtain optical transmission spectrum change to calculate the analyte gas concentration. As an attempt to eliminate the utilization of a tunable laser, an electrical tracing assisted MRR sensing system could be adopted ^[91]. In pursuit of improving the detection limit, other complicated MIR photonic devices are innovated. One of them is the photoacoustic (PA) gas sensor ^[92]. The detected gas molecules in the PA sensor chamber are excited to a higher-energy state by absorbing the MIR photon (injected from an MIR waveguide) energy. Then it is relaxed to the ground state by releasing the energy through collision with other molecules. An efficient MEMS microphone is used to detect the acoustic wave generated in this process. This type of sensor can achieve a detection limit below 10 ppm yet through a more complicated fabrication process with multiple bonding wafers.

The MIR photonics is becoming a hot research topic. We reviewed in this paper the recent R&D efforts of the silicon-based MIR photonics. Up to date, the platforms of SOI, GOS, GOI, GOSN, and SOS have been explored for the MIR passive device demonstration. The waveguides with different structures are implemented on these platforms, covering the wavelength range from 2 to 7.4 μm. The main achievements of the passive MIR photonic devices, including the grating couplers, MMIs, MZIs, MRRs, PhC cavities, and (de)multiplexers, are summarized in Section 2, as well as the active devices of MIR modulators, laser sources, and PDs. We also reviewed IME’s research and development efforts on the MIR photonic platforms. The basic passive MIR components with an SOI and SNOI platform are fabricated and characterized at the wavelengths of 3.8 and 2 μm. The MIR QCL laser sources, PDs, and relevant integration technology are under design and experiments. The discussion of the CO2 gas sensors illustrated one of the applications of the integrated MIR photonic system. Theoretical analysis indicates that the detection limit of 20 ppm can be achieved for the 40 cm spiral waveguide with the propagation loss less than 1.5dB/cm. For the MRR-based CO2 gas sensor, the Q-factor is required to be larger than 105 to obtain a detection limit below 60 ppm. The detection limit below 10 ppm is possible by utilizing longer spiral waveguide or MRR with higher Q.

Although MIR photonics has attracted increasing research interests, there are still many areas requiring further exploitation, including: 1)

Platform development for low-loss waveguide beyond 7.4 μm. The MIR waveguides with propagation loss <3dB/cm have been demonstrated in different platforms with wavelengths of less than 7.4 μm. However, as of now, there are very limited demonstrations of a silicon photonics platform for low-loss waveguides beyond the 7.4 μm wavelength range, especially for the long-wavelength fingerprint region.

Acknowledgment. We thank Ms Xin Guo and Prof. Hong Wang from School of Electrical & Electronic Engineering, Nanyang Technological University for their great support on the device characterization.