Device architecture and design. Figure 1a illustrates a schematic of the proposed MIR PIC-based sensor on the Ge-OI platform, comprising a slot waveguide for analyte sensing (passive sensing part) and a waveguide-integrated PD (detector part), monolithically integrated onto a single chip. Our on-chip photonic sensor utilizes light–analyte interaction within the sensing waveguide through absorption spectroscopy based on the Beer-Lambert law. An air-clad slot waveguide, supporting hollow-core guiding, has been employed to induce stronger light absorption with enhanced field confinement compared to conventional strip or rib waveguides, thus aiming to improve the sensitivity factor or to reduce the physical length of the sensing waveguide11,34. The residual light is then directly coupled from the sensing waveguide into the waveguide-integrated PD. As noted earlier, the operational principle of our proposed detector is the bolometric effect combined with FCA in Ge. For the bolometric material that converts light-induced temperature variations into changes in electrical resistance, we employed a TiO2/Ti/TiO2 tri-layer film, whose temperature-dependent electrical properties can be finely tailored by engineering the thickness of each layer31–33. The temperature change in bolometric detectors, in response to periodically varying incident light, can be described by23
where ΔT represents the temperature change, η is the absorption efficiency for the given wavelength, ω and Φ0 are the angular frequency and the amplitude of the periodic radiation, respectively, Gth is the thermal conductance between the detector and the surrounding environment, and Cth is the thermal capacitance of the detector. As inferred from Eq. (1), increasing η, while diminishing Gth and Cth, is critical for enhancing ΔT for a given incident optical power, which directly correlates with the bolometric detector’s responsivity. In order to boost η within our PD, FCA in Ge should be elevated, which greatly depends on the type of free carriers and the doping concentration for particular wavelengths. To take full advantage of FCA in Ge, we selected heavily-doped p-type Ge (p+ Ge) as the MIR-absorbing medium (details can be found in Supplementary Note 1). For the reduction of Gth and Cth, optimizing device geometries is crucial. Here, the optimization process, including the geometrical parameters of the bolometer region – specifically, a length (LB) of 4 µm and a width (WB) of 8 µm – was conducted by considering heating efficiency, back-reflection, and in-house fabrication capabilities. The systematic process of optimizing geometries with numerical simulations is detailed in Supplementary Note 2. Figure 1b shows the simulated steady-state temperature distribution for the device designed with the final parameters. The input waveguide, having a width (Win) of 2 µm, was designed to support only the fundamental transverse-electric (TE) mode. The incoming light was set to an optical power of 1 mW at a wavelength (λ) of 4.18 µm. As depicted in Fig. 1b, there is a significant temperature rise confined within the bolometer region. This localized heat generation is achieved by FCA within the p+ Ge, which demonstrates the viability of an FCA-based thermalization process acting as a compact and efficient MIR absorber, even in the absence of resonance structures.
Figures 1c and 1d show the optical microscope and cross-sectional transmission electron microscopy (TEM) images, respectively, of the fabricated device on the Ge-OI photonic platform featuring a 500 nm-thick top Ge, a 2 µm-thick Y2O3 BOX, and a Si substrate. Here, the proposed waveguide-integrated PD incorporates a boron-doped p+ Ge (bolometer region), a SiO2/Al2O3 (20/25 nm) insulating layer stack, a bolometric material of TiO2/Ti/TiO2 (25/2/25 nm) tri-layer film, and a Ti/W (100/150 nm) metal electrode. The thickness of each layer in the bolometric material was carefully optimized (discussed in Supplementary Fig. S6). Additional characterizations, such as X-ray photoelectron spectroscopy (XPS) and X-ray diffraction (XRD), are detailed in Supplementary Figs. S8 and S9, respectively. Energy-dispersive X-ray spectroscopy (EDS) elemental mapping (see Supplementary Fig. S10) confirms the successful deposition of each layer. Secondary-ion mass spectrometry (SIMS) depth profile analysis in Fig. 1e quantitatively reveals the impurity dopant concentration within the p+ Ge region. To make full use of FCA in Ge, ion implantation was performed with a high dopant dose of 5×1015 cm− 2, and the implant energy was carefully adjusted to 110 keV. This optimization contributes to exposing a larger fraction of the modal field to the peak doping-concentration region of the absorbing medium, aligning with the mode-field maximum and the projected range (Rp) of the implanted ions in the p+ Ge.
Thermo-electrical characterization. We first investigated the temperature-dependent electrical properties. Figure 2a shows the current-voltage (I-V) curves from a voltage sweep ranging from − 3.0 V to + 3.0 V, with a 0.01-V interval, measured from 293 K to 363 K (1-K step), limited by our Peltier-driven stage. A nearly-linear characteristics with Ohmic behavior between the bolometric material and the electrode stack was obviously observed. Figure 2b plots the temperature-dependent current values at 3-V derived from the I-V curves, demonstrating a significant relationship with the temperature. Figure 2c presents the resistance-temperature (R-T) characteristics under constant-voltage mode, which can be modeled following equation to estimate the activation energy (ΔE):
$$R(T)={R_{\text{o}}}\exp \left( {\frac{{\Delta E}}{{{k_B}T}}} \right)$$
2
where R(T) denotes the temperature-dependent electrical resistance, T is the absolute temperature, Ro is a constant, and kB is the Boltzmann constant. From Eq. (2), ΔE is extracted from the slope of the Arrhenius plot (ln(R) vs. 1000/T) shown in Fig. 2d, which was determined to be 0.315 eV within the measured temperature range. The temperature-dependent current modeling over a high-temperature range, based on the Arrhenius relationship by Eq. (2), is described in the inset of Fig. 2b, revealing a rapid increase in electrical current with rising temperature.
To quantify the temperature dependence of the electrical resistance, temperature-coefficient of resistance (TCR), a crucial performance indicator for bolometric detectors, is introduced, which is defined as the derivative of resistance with respect to temperature,
$${\text{TCR}}=\left( {\frac{1}{R}} \right)\left( {\frac{{dR}}{{dT}}} \right)= - \frac{{\Delta E}}{{{k_B}{T^2}}}$$
3
Photoresponse measurement. We now turn to explore the MIR photoresponse. The incident optical power coupled into the bolometer region was precisely calibrated, accounting for insertion losses from passive components with the assistance of an identical reference waveguide pattern without the detector part (see Supplementary Fig. S7). The un-illuminated I-V curve (dark state), plotted in the inset of Fig. 3a, revealed a dark current (Idark) of 127.5 nA at 3-V bias. Figure 3a presents the optical power (Pin) dependence on the photocurrent (Iph = Ilight – Idark) and the corresponding responsivity (R = Iph / Pin) at 4.18 µm, under a 1 kHz chopping frequency. We highlight that our device achieved an R of 12.19 mA/W (from linear fitting at Pin > 0.3 mW), exceeding the previous state-of-the-art value in waveguide-integrated thermal-type PDs beyond 3 µm, which is 10 mA/W at 5.2 µm in graphene-based PD on ChG-on-CaF2 waveguide using the PTE effect28. A slight nonlinearity is observed at lower Pin ranges, potentially attributed to variations in thermo-electrical properties and changes in both Gth and Cth with temperature. Additionally, we estimated the noise-equivalent power (NEP) by taking the ratio of the noise spectral density in dark state (see Supplementary Note 5) to the R of 12.19 mA at 4.18 µm, calculated as 3.4×10− 9 W/Hz0.5 at 1 kHz. Here, this far exceeds that of previously reported waveguide-integrated MIR PD using the bolometric effect (10.4 µW/Hz0.5 at 3.8 µm)27 and is comparable to the PTE-based PD (1.1 nW/Hz0.5 at 5.2 µm)28.
The frequency response was analyzed by varying the chopper frequency. As illustrated in Fig. 3b, our device showed stable performance with a nearly flat response up to 1 kHz (the limit of our setup). Although higher bandwidth might be beneficial, it is not a major concern in most spectroscopy applications – unlike in telecommunications and data communications – suggesting that our device is sufficiently robust for MIR lab-on-a-chip systems. We also evaluated the spectral response in the MIR band ranging from 4030 to 4360 nm. During the measurement, the Pin was maintained within the linear fit region (Pin > 0.3 mW). Our device exhibited a broadband photoresponse with an R of ~ 12 mA/W across the entire measurable range without any cutoff wavelengths. Lastly, we assessed the long-term stability with switching behavior, a key parameter for evaluating PDs. Notably, as depicted in Fig. 3d, highly stable and repeatable photocurrent generation was observed without noticeable performance degradation throughout the measurements. Here, we note that the response times were constrained by the open/close time of the beam shutter.
We have comprehensively compared our device’s performance with that of previously reported MIR waveguide-integrated thermal-type PDs, as shown in Fig. 4. Our device exhibits a broadband responsivity of ~ 12 mA/W, the highest among its counterparts, achieving an improvement of over 4,000 times compared to the previous waveguide PD using the bolometric effect27. Furthermore, there have been no demonstrations of waveguide-integrated thermal-type PDs beyond 5.2 µm. Traditional Si-based photonics platforms, such as Si-on-insulator (SOI) with a limit of ~ 4 µm and suspended-Si with ~ 8 µm, inevitably encounter wavelength limitations due to the intrinsic material absorption of SiO2 and Si in the MIR range36. Here, by leveraging the FCA-induced heating process on the Ge-OI platform that provides a broad transparency window, our approach can be widely utilized across much shorter or longer wavelength ranges in the MIR spectrum, offering significant potential for spectroscopic analysis of numerous biochemical molecules (shown in Fig. 4) without wavelength constraints. The responsivity modeling, based on FCA in Ge30 and normalized with our experimental results, is presented in Fig. 4 (red dotted line). Detailed performance characteristics are summarized in Supplementary Note 9.
Sensing demonstration. To demonstrate the label-free light–analyte interaction capabilities of our MIR PIC-based sensing platform, we arranged a 5-mm-long slot waveguide with our waveguide-integrated PD on a single Ge-OI chip, as shown in Fig. 5a. Efficient mode conversion was facilitated by strip-to-slot and slot-to-strip mode converters34 positioned at the entry and exit points of the slot waveguide (detailed in Supplementary Note 8), respectively, as depicted in the optical microscope and scanning electron microscope (SEM) images in Fig. 5b. The slot waveguide, designed for high confinement within an air-clad, featured geometrical parameters of 1.8 µm width (W), 0.2 µm slot gap (G), and 500 nm height (H)34, as shown in the cross-sectional SEM image in Fig. 5c, highlighting the well-defined slot region where the light–analyte interaction occurs.
Here, CO2, a major greenhouse gas contributing to global warming37, was selected as the target analyte with a strong absorption coefficient at 4.23 µm38. Under the continuous-wave (CW) operation at 4.23 µm, changes in CO2 gas concentration were detected by the photocurrent signal from our detector while simultaneously monitoring the actual CO2 levels using a commercial gas sensor placed near the device inside the chamber. Operation conditions were maintained at 3 V and 1 kHz for biasing and chopping frequency, respectively. Figure 5d presents the normalized photocurrent signal depending on the CO2 concentration, which exhibits a downward trend as expected from the absorption spectroscopy technique, achieving a sensitivity of 0.0696%/ppm through linear fitting. Additionally, to assess the repeatability of our optical sensing, we cycled the CO2 valve under nitrogen (N2) gas purging, varying the CO2 concentration between 100 and 250 ppm. As indicated in Fig. 5e, the photocurrent signal varied clearly and repeatably with the CO2 levels, exhibiting no memory effects. It should be noted that the response times were constrained by our experimental setup for both injecting and removing CO2 gas within the chamber.