Spectroscopy information is widely used to identify different matter and has a lot of applications, such as sensors^[1,2], multispectral detection^[3–6], and remote sensing^[7,8]. For conventional multispectral detection systems^[9,10], their wavelength division device and detector are separated and require mechanical or electrical scanning to obtain different spectral information. Most of the commercialized detectors have no ability to obtain spectral information by themselves. Wavelength division devices, such as a grating, prism, and Michelson interferometer, enable the detection system to identify the light wavelength precisely. However, the spectrum acquisition systems based on them are complex with large footprint and lead to heavy load and looseness^[11,12]. They are unsuitable for on-site or portable application areas. To minimize the volume of spectrum acquisition systems, a multi-color detector has been developed by integrating different waveband detection structures vertically to get two or more broadband spectral signals simultaneously^[13,14]. However, the multi-color detector is a longitudinal laminated structure with a certain degree of optical and electronic crosstalk. The spectral resolution is too low due to its broadband response. In addition, some filter-free detectors with a tunable narrowband response spectrum have been developed, such as organic photodetectors^[15–17]. These filter-free strategies have a small footprint, but it is not easy to increase the number of spectral channels to tens or even hundreds. Therefore, a miniature wavelength division device integratable with detectors is the key to reducing the footprint of multispectral and hyperspectral detection systems. To this end, ultra-compact wavelength division devices based on nanophotonic principles, such as metasurface, photonic crystal, and plasmonic structure, have been widely investigated^[18–25]. For example, researchers bond the photonic crystal slab on the charge-coupled device (CCD) or complementary metal oxide semiconductor (CMOS) detector to form a chip-level spectrometer ^[26–31] in the visible. It remarkably reduces the footprint of the spectrum acquisition system. However, the minimum size of metasurface structure is less than 1 µm. It needs highly precise and high-cost fabrication processes like e-beam lithography, which increases the fabrication difficulty and cost enormously. Additionally, no pixel-level wavelength division devices monolithically integrated with a detector in the short-wave infrared (SWIR) have been reported, to the best of our knowledge. Only a dual-waveband SWIR InGaAs focal plane array (FPA) with broad bands of 46 nm and 54 nm has been developed by integrating two filters ^[32].

In this work, we propose a new SWIR multispectral detector chip by monolithically integrating a Fabry–Perot (FP) microcavities array on an InGaAs FPA, which can acquire multiple narrowband spectral signals itself with an ultra-compact footprint. Meanwhile, it has no moving parts and is portable and stable.

We proposed and developed a new type of wavelength division device based on the integrated narrow bandpass filters^[25,30]. FP cavities using a distributed Bragg reflector (DBR) with all-dielectric materials have low loss, high efficiency and are with narrowband, which is very suitable for narrowband optical detection. By changing the cavity layer thickness of the FP cavity, the transmission spectral peak can shift as designed. Here, we fabricated the FP microcavities array directly on the InGaAs FPA. Each detector pixel aligns with an FP microcavity by UV lithography. Therefore, the wavelength division structure monolithically integrates with the detector chip. The spectral acquiring system’s footprint can be remarkably reduced to chip-level, and the stability is correspondingly increased. Only the light coupling with the transmission mode of FP microcavity can be received by the corresponding pixels, enabling different detector pixels to respond to different wavelengths of light. The current techniques can fabricate the FP cavities on a micro-scale by combining lithography and etching processes, with minimum size equal to one detector pixel. The FP cavities can be designed and integrated as an array with different narrow passbands on the pixel level^[12,21]. Furthermore, different from the metasurface and grating, which work with diffraction and scattering principles, the light interference in the FP cavity results in a superior narrowband spectrum.

Figure 1(a) shows the schematic diagram of a monolithic integrated multispectral InGaAs detector. The detector chip comprises two parts: one is the 64×64 pixels InGaAs FPA, whose origin spectral response range is from 900 to 1700 nm; the other is the monolithic integrated FP microcavities array, which is directly fabricated on the backside of the InGaAs FPA. The film stacks of the FP microcavity can be described as (LH)32.37L(HL)3, which can be designed in detail according to our previous work^[33]. H and L represent high and low refractive index materials with optical thickness of a quarter of the designed wavelength λ0. The theoretical physical thicknesses of H and L are λ0/(4nH) and λ0/(4nL). We selected Si and SiO2 as H and L layers with refractive indices of 2.53 and 1.46, respectively. The experimental physical thickness of each SiO2 (L) and Si (H) layer is 291 nm and 136 nm, respectively. The initial cavity layer is designed as 2.37L to enable the right spectrum peak to locate at l666 nm, and other spectral channels blue shift with the reduction of cavity layer thickness. The number of DBR stacks (LH) or the thickness of the cavity layer can tune the bandwidth and spectral resolution of the filter: the larger number of DBR stacks (LH), the higher the spectral resolution of the structure. The reason for choosing Si and SiO2 as high and low refractive index materials is to get large refractive index difference and have wide spectral tunable range. Other materials like TiO2, Ta2O5, can also be selected as high refractive index material in the SWIR range. SiO, ZnS, etc., can also be selected as low refractive index material in the SWIR range. In these materials, the refractive index difference of Si and SiO2 is the largest, but the absorption of Si is also relatively large, leading to low transmittance. The absorption is low with high transmittance when the refractive index difference is small with other materials. Hence, different choices and combinations of materials can be made according to different application requirements. By changing the cavity thickness of each FP cavity in different pixel areas, the transmission peak can be tuned to different wavelengths. The simulation spectra of this structure are shown in Fig. 1(b). Here, we designed 16 response spectra of the detector by integrating different thicknesses of each FP cavity for demonstration.

The fabrication process is illustrated in Fig. 2. The Si and SiO2 layers are deposited on the backside of the InGaAs detector by using an electron beam evaporation system (Leybold ARES1110) with vacuum of 1.6×10−4mbar (1 mbar = 100 Pa) and temperature of 100°C, where the evaporation speed for SiO2 is 48 nm/min, and for Si is 12.6 nm/min. The bottom DBR stacks and cavity layer of (LH)32.37L are deposited firstly. Then, the cavity layer is processed with combinatorial etching processes. Sixteen cavities with different thicknesses are fabricated with only four times of UV lithography (MJB4, SUSS Mask aligner05080104) and inductively coupled plasma (ICP, Oxford PlasmaPro System 100) etching process. The two gases used for etching were argon (Ar) and trifluoromethane (CHF3) with a flow rate ratio of 1:3. The radio frequency (RF) power and ICP power were set as 50 W and 80 W, respectively. In addition to the combinatorial etching process, the FP microcavities array can also be realized by using combinatorial deposition processes^[34]. Lastly, the top DBR stacks of (HL)3 are deposited on the structure. The FP microcavities array is monolithically integrated onto the InGaAs detector chip. A representative cross-sectional scanning electron microscopy (SEM) image of the monolithic integrated FP microcavity is shown in Fig. 3(a) to verify the accuracy of our experiment. The incident light is filtered by the FP microcavity and absorbed by the InGaAs detector pixel, then converted to the electrical signal, and output by the readout circuits (ROIC). The picture of the fabricated 16-channel multispectral detector chip with ROIC is shown in Fig. 3(b). Meanwhile, its detector pixels on the former side and the FP microcavities array on the backside are correspondingly aligned during the lithography process. Figure 3(c) is the image of the 64×64 pixels InGaAs detector FPA. Figure 3(d) is the monolithically integrated FP cavities array, and 16 different optical channels are separated by the diaphragms (Au with 100 nm thickness) of yellow areas. During the combinatorial etching processes, the diaphragms work as a buffer area between channels to prevent crosstalk etching.

After the monolithic integrated process, the multispectral detector chip is bonded with ROIC [Fig. 3(a)]. The total footprint of the fabricated 16-channel multispectral detector chip is about 2mm2, which is minimum and much smaller than the size of conventional infrared multispectral system. The spectral response of the fabricated 16-channel multispectral detector is measured by a monochromator with a collimator, as shown in Fig. 4(a). Meanwhile, the same type of standard InGaAs FPA is measured by the same system as a comparison. The response spectra of the reference standard detector and the fabricated multispectral detector are shown in Figs. 4(b) and 4(c). The response spectrum of the standard InGaAs FPA is broadband from 900 to 1700 nm. In contrast, the monolithically integrated multispectral InGaAs FPA has 16 distinct response spectra on different pixels. The peaks of the measured 16 different response narrowband spectra are 1666, 1649, 1642, 1628, 1606, 1591, 1584, 1571, 1543, 1529, 1522, 1508, 1486, 1473, 1467, and 1454 nm.

As designed, each channel of our multispectral detector has a unique narrowband response spectrum. The peaks of these 16-channel response spectra are separated and enable each detector channel to detect different specific wavelengths of light. Moreover, the channel number of the multispectral detector chip can be increased exponentially by increasing the times of the combinatorial etching process. Experimentally, the full width at half-maximum (FWHM) reaches 24 nm on average, which can be narrowed by adding the DBR stack to be as narrow as 1 nm^[30] or even narrower, and meets the needs of many multispectral detection fields. The difference between simulation and experimental results is mainly caused by the fabrication processes. The FWHM and transmittance of the FP cavities array are directly affected by the n (refractive index) and k (extinction coefficient) values of the materials. The vacuum and temperature of the coating process can influence the n and k values of the films. The experimental FP microcavities array is directly fabricated onto the InGaAs FPA. Hence, the measured 16-channel response spectra of this multispectral detector chip are the combined results of the FP cavities array spectra [Fig. 1(b)] and InGaAs detector response spectrum [Fig. 4(b)]. For instance, the low response intensity of the last channel (spectrum peak at 1666 nm) comes from the intrinsic low responsivity of the InGaAs detector itself, as shown in Fig. 4(c).

Indeed, the multispectral detector chips with narrowband response spectra are helpful for infrared spectral detection. Increasing the number of DBR layers or adopting material with a higher refractive index difference can narrow the bandwidth of the response spectrum. Furthermore, increasing the maximum etching depth and etching times of the fabrication process will increase the spectral range and the channel number of multispectral detector chips, respectively. The response channels of 1649, 1606, 1571, 1529, 1454 nm correspond to the absorption peak of gases CH₄, H₂S, CO, C₂H₂, CO₂, respectively. Since many gases have characteristic absorption spectra in the SWIR range, such a multispectral detector chip can work as a sensor and multispectral imager. The peaks of response spectra can be designed and controlled to gas absorption wavelengths for the application of the non-dispersive infrared (NDIR) multi-gas sensor. In addition, an ultra-compact multispectral image camera can be realized with the help of a push-broom setup on such a detector chip or construction of a wavelength division structure by using different channels super-pixel arrays.

In conclusion, we have demonstrated a novel SWIR multispectral detector chip by monolithically integrating the narrowband FP microcavities array with the InGaAs FPA. Each FP microcavity and the corresponding detector pixels are integrated and aligned using UV lithography. The multispectral detector chip can distinguish 16 different SWIR wavelengths by itself. The response spectra of each detector pixel can be adjusted flexibly by changing the thickness of the cavity layer. The spectral channels can be increased to 128 by only seven times of combinatorial etching. The monolithically integrated detector chip without moving elements is portable and stable, remarkably reducing the footprint of the multispectral detection system to a 2mm2 size. It is promising for practical applications such as multispectral infrared imaging in smartphones, object identification in automatic drive, and real-time monitoring of multiple gases in the industrial fields.