Strategies for high performance and scalable on-chip spectrometers

As we know that the great challenge coming from shrinkage of spectrometer is the limited OPL resulting in poor R and SR, which are two very important metrics for a spectrometer. DPH grating is a new approach for designing devices with the programmed properties to control light scattering many times to gain longer OPL [28–30]. This technology belongs to the class of computer generated holograms and can encode any specific optical transfer function by manipulating the light in the spectral and spatial domains. DPH involves millions of lines specifically located and oriented in order to direct the output light into the designed focal points according to the wavelength as illustrated in figure 3(a). Spectrometers were fabricated on dielectric material such as silicon nitride or silicon dioxide or hafnium dioxide planar waveguides using electron beam lithography (EBL) and dry etching, a typical linewidth of grating's grooves was 100–140 nm linewidth and 50–150 nm depth for SiO₂Ge_x waveguide core, and about 80 nm linewidth and 10–20 nm depth for devices with HfO₂ core as shown in figure 3(b). Each DPH works in one specific spectral bandwidth, whereas the integrated circuitry includes several functionalities (taper, splitter, and waveguide) to efficiently manipulate the light and guide it to the holograms. Integration multiple DPHs on the same device can broaden the measurement spectral range. In this way a high-resolution, broadband spectrometer-on-chip can be realized with performance that is similar to conventional spectrometers. In 2014 [30], the newly developed spectrometer achieved a high spectral resolution of 0.15 nm in the red and near infrared range, over a 148 nm spectral range (630–694 nm, 766–850 nm) with 926 channels, while maintaining the total device footprint below 2 cm². In addition to the bandwidth and linewidth, the dispersion curve of a DPH spectrometer can be customized to assume any specific profile, allowing resolution changes over different bands of interest for specific applications. While the spectral range is mainly limited by the number of channels. Another disadvantage is the scattering loss (is around −15 dB) and the Cross-talk (defined as the ratio between the light intensity in one channel and the intensity of the interfering light in lateral channels ∼10 dB) along light propagation. Despite the loss, an input spectral intensity as low as 50 pW nm⁻¹ can be detected with an integration time of 10 ms. For practical application, high quality of structure with small roughness is highly desired to reduce the optical loss. And the requirement will become stricter for application in light with shorter wavelength for the finer nanostructure of DHP grating should be fabricated.

Zoom In Zoom Out Reset image size

Narrow-band pass filters are another important monochromatic dispersion component, modern optics engineering lay the foundation for complex components fabrication, such as plasmonic metal array and dielectric mini-filters array. A high-resolution miniature spectrometer with 128-channel integrated mini filter array has been demonstrated in 2007 by Lu et al [13] schematically shown in figure 4(a) with ultra-narrow transmission spectra. These filters were fabricated by using the combinatorial deposition technique to deposit film layers to form Bragg-cavity with selected wavelength, and work as a dispersive component whose passbands range from 722.0 to 880.0 nm with a bandwidth (or spectral resolution) from 1.7 to 3.8 nm. The miniature spectrometer is smaller than 1 cm³. Total pixel number is 795 × 596 with each pixel size of 6.5 μm × 6.25 μm, is just put in front of a charge-coupled device (CCD) camera, without any moving parts. Incident light propagate through the ultra-narrow band-pass filter to form monochromatic light and detected by the CCD underneath at different location, then the signals are collected by computer to form a continuous spectrum. This kind of miniature spectrometer has the advantages of very low payload, high resolution. But the complex fabrication process and small amount of incident light on each filter make the sensitivity a great challenge.

Zoom In Zoom Out Reset image size

Most current microspectrometers rely on interference filters and interferometric optics that limit their photon efficiency, resolution and spectral range. Bao and Bawendi have developed an efficient, cost-effective microspectrometer that overcomes many of these limitations by replacing interferometric optics with a 2D absorptive filter array consisting of 195 different semiconductor colloidal quantum dots array with transmission features that cover a broad spectral range [22], figure 4(b) gives the typical transmission spectral of colloidal quantum dots (CQDs) and the device picture. With it the authors reconstructed spectra with R of 2–3 nm and broad BW of 300 nm, where R is related to the variation of the CQDs size and BW is determined by the optical properties of intrinsic semiconductor material. The dynamic spectral range can be extended to a quiet a broad range by using the other kind of CQDs such as ZnS or InP, respectively [48]. This approach gives a simple and easy way for a miniature spectrometer with high performance, especially potential higher R could be obtained by well controlled CQDs fabrication. But its operation spectral range depends on the semiconductor's property, as a result, a limited measurable spectral range. As well as basically, the delta transmission spectra contribute to the spectra reconstruction, this approach still does not improve the weak light–matter interaction strength caused by size shrinkage.

Photonic crystal nanocavity or random structure can function as monochromatic dispersion component. A compact spectrometer with an array of high-quality-factor PC nanocavities [49], coupled via a planar 2D waveguide enables spectral analysis of incident light with resolution as high as the bandwidth of the cavity mode –0.3 nm at 840 nm. Due to limited OPL, the BW is only about 20 nm with 100 cavities. Unfortunately, it is very sensitive to fabrication errors. In addition to these carefully designed systems, disorder and scattering system have also been explored for spectroscopy applications benefiting from its longer OPL [41, 49]. Early in 2003, Xu et al used the spatial-spectral transmission patterns induced on low coherent field by disordered PCs to construct multimodal spectrometers which are designed to measure the spectral density averaged over multiple spatial modes [50]. In 2010, Dogariu et al showed that random scattering materials have sufficient diversity in spectral transmission to allow for precise measurements of the spectrally dependent polarization state of an optical field [51]. In 2013, Cao et al used multiple scattering in on-chip random structure of semicircular (with radius of 25 μm) array of randomly positioned air holes [46] to build a compact spectrometer, which greatly increase the OPL and achieved fine spectral resolution of 0.75 nm and 25 nm bandwidth around 1500 nm with a small footprint. As shown in figure 4 (c). The incident light is guided by a waveguide then diffuses through the random array via multiple scattering and eventually reaches 25 defect waveguides which will couple the signals to the detectors array (not integrated). The distribution of intensities over the detectors is used to identify the input spectrum. But, losses coming from scattering and the out-of-plane leakage limit its sensitivity. By replacing the detector array with superconducting nanowire SCDs, most recently Pernice et al employed the similar random PC (with radius of 100 μm) structured spectrometer equipped with a SCD and realized high sensitivity down to powers of −111.5 dBm in the telecom band [32, 52]. They reconstructed near infrared spectrum with super fine linewidth of 4 nm and 30 pm (around 1500 nm and around 800 nm, respectively) and high sensitivity in 1.3 K ambient.

The working principle of such disordered PC devices is based on the analysis of the speckle pattern which is formed by light after propagation through a disordered medium. The speckle pattern provides a unique 'fingerprint', which is usually stored beforehand in a calibrated transmission matrix. This matrix contains the spectral-spatial-mapping of the spectrometer. By measuring this 'fingerprint', any probe signal can be reconstructed by inverse multiplication with the transmission matrix. There are two main advantages of this smallest spectrometer: fine spectral resolution can be achieved due to the greatly enhanced OPL; the other is the spectrometer can work in a broad dynamic spectral range determined by the calibrating spectra. However, this kind of spectrometer will be damaged if the probe light is out of the calibrating light. Meanwhile the limited number of output channels result in narrow BW for each measurement.

A summary of some remarkable monochromatic dispersive micro spectrometers is presented in table 1, with respect to device footprint, bandwidth, resolution, sensitivity and so forth.

Both the limited channel counts and moving parts are the challenges for FTIR spectrometers scaling down. In 2018, Hu et al proposed and experimentally demonstrated a novel dFT spectrometer architecture that resolved the performance and scalability challenges [60] by digitally controlling exponentially increase in channel counts. The device consists of two arms of MZI with number j optical switches on each arm that direct light to waveguides of unique path lengths illustrated in figure 5(a). And thus the channel counts increase by 2^j . This approach claims three key advantages over state-of-the-art techniques. First, both the resolution and spectral channel count scale exponentially with j. The unique exponential scaling laws inherent to the dFT architecture enable high-resolution spectra to be acquired with minimal chip space and control electronics. Second, direct modification of the waveguide path offers over two orders of magnitude larger OPL modulation per unit waveguide length compared to thermo-optic- or electro-optic-based index modulation, enabling superior spectral resolution within a compact device. Third, the device benefits from the multiplex advantage and requires only a single-element photodetector rather than a linear detector array, which reduces cost and system complexity. They used this spectrometer with 64-channels (device picture is given in figure 5(b)) to restore a broadband spectrum from 1550 to 1570 nm with super fine resolution of 0.025 nm. The dynamic spectral range can be expanded by increasing the number of the switches as well as the length of MZI.

This work pioneers dFT spectroscopy as a high performance, scalable solution for on-chip optical spectrum analysis. Moreover, its proven compatibility with industry-standard foundry processes enables scalable manufacturing and drastic cost reduction. They further developed an elastic-D1 machine learning regularization technique to achieve significant noise suppression and resolution enhancement. The powerful combination of dFT spectroscopy and machine learning techniques will empower future applications of spectroscopy such as chemical and biological sensors-on-a-chip, space-borne spectroscopy, optical network monitoring, and radio-frequency spectrum analysis.

SWIFT working without any moving parts is promising for integration by analyzing light by standing wave, which was first proposed by Gabriel Lippmann in 1891, unfortunately, this principle was too complex to use, even though there has been great progress in more than one century [61]. In 2007, Coarer et al taking advantage of the development of photonics and near-field optics, proposed a new kind of stationary-wave integrated Fourier-transform spectrometry (SWIFTS) [17, 62], in which direct sampling of evanescent standing waves was achieved using a collection of optical nanoprobes of gold nanowires (named as evanescent field samplers (EFSs)) with 4 $\mu {\text{m}}$ × 50 nm in size, around 3 $\mu {\text{m}}$ in space, illustrated in figure 5(d). Light is coupled into a single mode waveguide terminated with a mirror. When reflected onto the mirror, the waves become stationary. Miniature localized detectors are placed in the evanescent field of the waveguide mode in order to extract only a small fraction of the guided energy. As different light has different OPL difference between the incident light and reflected light, the position of stationary wave of different light are different, the target spectrum can be recovered by computing the discrete Fourier transform of the stationary-field profile.

This method can be realized mainly due to the advanced waveguide fabrication and development of detecting and imaging the near field intensity, this technology enables high-resolution spectrometers with compact and robust design. At this stage, the prototype 1 mm² device with a spectral resolution of 4 nm over a working spectral range of 96 nm centered on 1500 nm face strong bandwidth limitations due to undersampling of the standing wave. Integrated spectrometers based on the SWIFTS principle have been demonstrated on various material platforms including silicon, silicon nitride, quartz and polymers. However, due to the inevitably extreme undersampling of the standing-wave pattern, the resolvable bandwidth of devices based on SWIFTS is limited to 5–14 nm at a wavelength of 630–1100 nm.

A short summary of some remarkable Fourrier transform based micro spectrometers is presented in table 2.

A plasmon has a resonant frequency that depends on the metal used and on the nanostructure of its surface [67]. As early as in 2001, first attempt at exploiting plasmonics for color pixels was using subwavelength grid nanostructures fabricated on top of complementary metal oxide semiconductor (CMOS) for imaging in the visible [68]. High responsivities and transmissivities were shown at red, green and blue wavelengths by tuning the grid periodicity and spacing [69]. This approach was later used to demonstrate plasmonic color filtering on the scale of tens of micrometers [70–73]. It seems like these approaches are very promising, however, they still require complex fabrication steps. To overcome this difficulty, researchers found a simpler route to plasmonic filters in the form of subwavelength hole arrays instead of numerous layers [74, 75]. Tuning the size of the holes and their mutual distance alters the resonant frequency of the array and hence the color that they transmit. In this configuration, most of the light is transmitted and relatively narrowband color filters can be produced [69, 76, 77]. The plasmonics metal array can be used as filter in the miniature spectrometer for its superior small size. In 2011 Kurokawa et al reported this kind of filter-based miniature spectrometer as well as a method for spectral reconstruction using adaptive regularization [14, 78]. The schematic of the prototype low-cost nano-optic filter-based spectrum sensor is shown in figure 6(a). The core technology of this spectrum sensor on-a-chip is based on nano- optical filter array integrated onto a regular CMOS imager. By introducing nanoscale structures on metal films, plasmonic devices can provide a unique way to control polarization and wavelength of light passing through the structures. Four typical transmission spectra out of the 195 filter array are given in figure 6(b). Notably, unlike the ideal filter with ultra-narrow delta-shape transmission curve, the transmission function of these filters are broad and overlapped containing multiple peaks, and long tails as illustrated in the figure. When these non-ideal spectral filters are used, the raw outputs of spectral filters cannot compose a spectrum automatically. Therefore, a spectrum reconstruction process converting the raw data from spectral filters into spectrum is necessary.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Accurate spectrum reconstruction was realized by regularization techniques and constrained optimization method shall be applied [79]. The method of Tikhonov regularization and with the L-curve method for parameter selection for stabilizing the solution of inverse problems is applied to the prototype nano-optic filter-based spectrum sensor for spectrum reconstruction, which was found useful to deliver satisfactory estimate. NanoLambda has been developing a chip scale spectrum sensor based on the new plasmonic filter array technology as well as this spectrum reconstruction method.

A photonic bandgap fiber bundle spectrometer is realized in 2010 with very limited R of 30 nm. Unlike the regular bandgap dispersion, series of broad irregular but coded transmission spectra could be used for spectra reconstruction, which benefits from wavelength multiplexing and advanced spectra reconstruction algorithm, meanwhile overcomes the limitation like the FTIR spectrometers need regular sine or cosine modulation to obtain interferogram for spectra retrievement. A single-shot spectral imaging method was proposed by Yu et al by using complex optical interference in PC slabs array as the dispersive filter [20, 80], which are micrometer-thin dielectric layers with periodic patterning. Light incident from free-space can couple to lateral propagation modes, where the periodic nanostructures allow light to bounce back and forth many times. Unlike microcavities, where the path enhancement only occurs at resonant frequencies, the effect of path enhancement in PC slabs spreads over a broader spectral range and creates a transmission spectrum with rich spectral features, including sharp peaks due to guided resonances, broad background variation by Fabry–Perot resonance, and irregular line shapes due to Fano interference. To work as a spectrometer, arrays of different PC slabs are fabricated on top of a CMOS imaging sensor. Each PC slab has a different periodicity, lattice constant, and hole sizes. These slabs creates a transmission spectrum with unique and rich spectral features. The authors recovered broadband spectrum from 550 to 750 nm with spectral resolution of ∼1.5 nm by regularization method. The spectrometer is CMOS compatible and promising application for lab-on-a-chip and hyperspectral imaging. However, the transmission through PC slabs is angle dependent, thus the spectrometer requires the incidence angle to be the same for the calibration and measurement, the performance degrades if the two angles are different by more than 1 degree, in practical application, it requires a collimating aperture to ensure a consistent incident angle. Alternatively, the PC slabs offer a flexibility to reduce the size of PC slabs to increase the angle tolerance at the cost of poorer spectral resolution.

A monolithic single nanowire miniature spectrometer was realized by Hasan [34] and Wang [35], taking fully advantages of bandgap-graded semiconductor NW. Firstly, bandgap-graded semiconductor nanowire is a promising wavelength selective component for the advanced on-chip spectrometer because of its large and continuously tunable absorption range (from 505 to 710 nm). Secondly, the nanowire naturally supports fabricating linear array photo-detectors in a facile way and has a miniature footprint. The composition-graded CdS_x Se_1−x single nanowires which were first fabricated by Tong and Yang in Zhejiang University [81], are synthesized by vapor–liquid–solid method in a horizontal quartz tube mounted in a furnace using a source moving thermal evaporation technique. Figure 7(a) shows the fluorescent photo of a typical nanowire. The gradually changed luminescence color already implies a changing of bandgap.

Using these wires, Hasan et al fabricated conductive photodetectors array along the nanowire as shown in figure 7(b). By using this nanowire miniature spectrometer, they realized spectra reconstruction with R of 5 nm (with 38 devices, the more the devices the finer of the R), the BW is determined by the bandgap of the NW. They also demonstrated the capacity of spectral imaging from centimeter-scale focal planes down to lensless, single-cell-scale in situ mapping. Furthermore, Wang fabricated Schottky diode as photodetector array along the NW, instead of photoconductors, and demonstrate its very promising application in on-chip measurement. Comparing the Ohmic photo-conducting detectors, the Schottky diode has very low dark current, faster response time and, the most importantly, has very low figure of noise, and thus higher sensitivity larger than 10¹³ Jones which is almost two orders higher than its photoconductor counterpart given in figure 7(c). For the on-chip luminescence measurement, spectrometers with higher sensitivity is highly demanded for weak light detection. Another advantage is that nanowire also support waveguide-mode operation, in this way, it naturally allows detecting spot light source emitted from single quantum dot or molecule; as well as the waveguide mode can also enhance the light–matter interaction during light propagation along the guiding and active nanowire. And it does not change the photo-response dispersion as that operated in far-field mode.

Equipped with this high sensitivity nanowire monolithic spectrometer, on-chip PL measurement was carried out by aligning other long nanowires with the nanowire spectrometer in end-to-end configuration (figure 7(d)). Positioning an excitation laser at the far-end of emission nanowire, PL emits out from the near-end can directly enter the spectrometer by face-to-face coupling. And the spectrometer is operated in the waveguide mode, which allows all the photodiodes to simultaneously obtain photocurrent. Figure 14 shows the PL (at a power lower than 0.1 pW) and retrieved spectra. The reconstruction accuracy obviously relays on the sensitivity of the sensors, highlighting the importance of light–matter interaction in microspectrometer. For the commercial application, this strategy still needs well developed nanowire assembling technique for CMOS compatibility.

As we know silicon photodiode is a good detector for visible light. Vertical silicon nanowire (Si NW) array can act as filter due to its cavity effect, tunable absorption spectra can be achieved by tuning the radius of NW array [36]. Thus recently, Crosier et al employed Si NW photodetector array only as a miniature spectrometer [21, 82], on the one hand, the NW array act as filters for the effect of waveguide enhancing the resonant wavelength absorption, on the other hand, each Si NW is a photodetector which converts light to current recorded by the outer circuit. Given the absorption spectrum of each array and the output photocurrent signals, combining with the reconstruction method, the spectrum of the incident light can be retrieved. This spectrometer consists of 24 pixels of Si NW array that vary in radius from 60 to 175 nm, with height around 2.8 μm, each NW pixel has an extent of ∼150 μm × 150 μm. It demonstrates R with 5–6 nm and cover from 400 to 800 nm, practical measurement should make a trade-off between R and BW due to limited pixels. The paradigm of detector-only spectrometer of combining the detector and filter in one is a breakthrough in this area, but the complex fabrication process and reliability will be a challenge for practical application, especially the slow EBL writing process will limit its performance and device output.

A summary of the main bandgap dispersive micro spectrometers is presented in table 3.