Skip to main content
NCBI home page
Science Advances logoLink to Science Advances
. 2025 Jan 22;11(4):eado6886. doi: 10.1126/sciadv.ado6886

Miniaturized spectral sensing with a tunable optoelectronic interface

Xiaoqi Cui 1,*,†, Fedor Nigmatulin 1,†, Lei Wang 2,†, Igor Reduto 3, Andreas C Liapis 1, Mingde Du 1, Md Gius Uddin 1, Shafi Abde Mayeen 1, Faisal Ahmed 1, Yi Zhang 1, Hoon Hahn Yoon 4, Harri Lipsanen 1, Seppo Honkanen 3, Timo Aalto 5, Zongyin Yang 6, Tawfique Hasan 7, Weiwei Cai 2,*, Zhipei Sun 1,*
PMCID: PMC11753376  PMID: 39841830

Abstract

Reconstructive optoelectronic spectroscopy has generated substantial interest in the miniaturization of traditional spectroscopic tools, such as spectrometers. However, most state-of-the-art demonstrations face fundamental limits of rank deficiency in the photoresponse matrix. In this work, we demonstrate a miniaturized spectral sensing system using an electrically tunable compact optoelectronic interface, which generates distinguishable photoresponses from various input spectra, enabling accurate spectral identification with a device footprint of 5 micrometers by 5 micrometers. We report narrow-band spectral sensing with peak accuracies of ∼0.19 nanometers in free space and ∼2.45 nanometers on-chip. In addition, we implement broadband complex spectral sensing for material identification, applicable to organic dyes, metals, semiconductors, and dielectrics. This work advances high-performance, miniaturized optical spectroscopy for both free-space and on-chip applications, offering cost-effective solutions, broad applicability, and scalable manufacturing.

An electrically tunable compact optoelectronic interface functions as a spectral sensor, achieving material identification.

INTRODUCTION

Optical spectroscopy stands as a fundamental and universally embraced method for discerning the physical, chemical, and structural characteristics of materials. Over time, characterizing materials within a limited cost, unit size, and power consumption has become increasingly crucial for a broad range of handheld, portable, and integrated applications, leading to the continuous miniaturization of optical spectroscopy (1). However, following the operating principles of tabletop systems that use bulky optical dispersive components or filter arrays, conventional miniaturizing approaches that use smaller optical components have encountered a bottleneck due to imperfections and limitations in advanced manufacturing techniques (2–4).

Recently, there has been a substantial interest surrounding the miniaturization of high-performance reconstructive spectrometers [e.g., bandgap-gradient nanowires (5) and electrically tunable heterojunctions (6–12)], resulting in notable size reductions (2, 3, 13–15). In these methods, input complex spectra (a two-dimensional signal of wavelength and intensity) are converted into corresponding electrical signals (usually with the same dimensions) for spectral reconstruction, from a precalibrated higher-dimensional electrical signal matrix (e.g., nanowires use a unit-wavelength-photoresponse matrix and heterojunctions use a gate-wavelength-photoresponse matrix). For high-performance reconstruction from a three-dimensional matrix to a two-dimensional matrix, the higher-dimensional matrix should be as numerically full rank as possible (3, 16). However, the wavelength-dependent photoresponse matrix typically show insufficient uncorrelation because the external quantum efficiency (EQE) is normally independent of the input wavelength (or photon energy) in these miniaturized devices, and the device tunability is limited by many factors (such as the material bandgap, device damage threshold, and linear dynamic range). This feature results in a severely rank-deficient matrix that could be used for a high-efficient wavelength meter (three-to-one-dimensional reconstruction) rather than a spectrometer in the strict sense, although different algorithms (e.g., the enrollment of artificial intelligence and deep learning) might make a slight difference in the performance.

Here, we demonstrate miniaturized spectral sensing with an electrically tunable compact optoelectronic interface, capable of generating distinguishable electrical signals from various input spectra (Fig. 1A). Leveraging both bias-voltage (VDS) and gate-voltage (VGS) tunability, a high-performance four-to-one-dimensional spectral sensing is performed to decode featured electrical signals from a precalibrated database. We report narrow-band spectral sensing with a high peak accuracy of ∼0.19 nm (free space) and ∼2.45 nm (on-chip). Furthermore, we implement broadband complex spectral sensing for material identification, ranging from organic dye, metallic, and semiconducting to dielectric materials (Fig. 1B). The spectral sensing and identifying process facilitate a number of applications beyond material identification, such as composition analysis [through photoluminescence (PL) peak sensing] and computing (by encoding input spectra). Our work paves the way for high-performance complex spectral sensing, enabling large-scale and cost-effective miniaturized optical spectroscopic applications.

Fig. 1. Spectral sensing with an electrically tunable optoelectronic interface.

Fig. 1.

Open in a new tab

(A) Our compact optoelectronic interface generates distinguishable electrical responses from diverse input optical spectra. The sensing and identifying of these spectra are applicable for material identification, composition analysis, and potential computation. (B) Application examples of complex spectral sensing for material identification. The electrical response of the unknown material is encoded from its optical spectroscopic signal through our optoelectronic interface. Then, the electrical signal is analyzed alongside a database of prelearnt materials using a computational algorithm, enabling the identification of the unknown material. Mat., material; CNT, carbon nanotube; Semi., semiconductor.

RESULTS

The configuration of our free-space device is shown in Fig. 2A (device image in inset). Fabrication details are included in the Methods section. We adopt a van der Waals (vdW) metal (NbTe2) as one of the conductive electrodes for a vdW semiconductor InSe. The NbTe2 has a work function of ~4.5 eV (17), which is close to the electron affinity of InSe (~4.45 eV) (18, 19). This unique property of the NbTe2 electrode facilitates efficient tunability of the vdW Schottky junction, constructing a tunable vertical Schottky junction and a tunable lateral homojunction (20, 21). Besides, InSe has a direct bandgap of ~1.26 eV (22, 23) with fast carrier mobility (24) and high photoresponsivity (18, 25), which enables the device to be working in the full visible range. Benefited from the partial shielding effect and the aligned Fermi level, the vertical InSe/NbTe2 vdW Schottky junction can be effectively tuned by VDS, and the lateral InSe homojunction can be configured by VGS. This configuration endows the photoresponses with two dimensions of tunability. The output curves (Fig. 2, B and C) agree with our previous results (21), validating the operation of our configuration (note S2, Supplementary Materials).

Fig. 2. Narrow-band spectral sensing.

Fig. 2.

Open in a new tab

(A) Schematic and the optical image (inset) of the free-space InSe/NbTe2 device. Scale bar, 30 μm. (B) Output curves of the device with 200-nW light illumination at different VGS. The incident light wavelength is 550 nm. (C) Output curves of the device with 200-nW light illumination of different incident light wavelengths. (D) Four signature frames of the four-dimensional photoresponse mapping at VGS = −18, −6, 6, and 18 V in a spectral range from 500 to 720 nm. (E) Identification results of our device (solid) and the references (dashed). a.u. arbitrary units. (F) Identification accuracy of our device with a minimum error of ~0.01 nm for single test wavelength; the smallest RMSD is ~0.19 nm for all test wavelengths. (G) Schematic and the optical image (inset) of the waveguide-integrated device. Scale bar, 10 μm. (H) Signature photoresponse mapping at VGS = 0 V. (I) Identified results (solid curves) and the references (dashed curves) from 500 to 840 nm. The identification accuracy of the integrated device with a minimum error of ~0.1 nm for single test wavelength; the smallest RMSD is ~2.45 nm for all test wavelengths.

We first demonstrate narrow-band spectral sensing by using monochromatic light as the input (see Methods). As shown in Fig. 2D, we display four wavelength-dependent photoresponse mappings at VGS = −18, −6, 6, and 18 V. Note that our device has two dimensions of tunability (VDS and VGS) in contrast to one dimension in the previous demonstrations [e.g., position in nanowire (5) and VGS (6) or VDS (8) in vdW heterojunctions]. Thus, the photoresponse mapping shows an extra degree of freedom and represents as a four-dimensional matrix (wavelength-responsivity-VDS-VGS). This enables a four-to-one-dimensional spectral sensing (note S4, Supplementary Materials) that provides accurate identification of the peak position for unknown spectra (Fig. 2E). In our algorithm, the identification is conducted at each VGS value, yielding a series of results. Four filter functions are applied sequentially to select effective results, which are then averaged to produce the final output (note S4, Supplementary Materials). Benefiting from the electrical signal processing and a filter function in our algorithm, the identified testing wavelengths show a minimal error of ~0.01 nm (at a single testing wavelength of ~520 nm in Fig. 2F). We further define the identification accuracy as the root mean square deviation (RMSD) of all testing cases across the full operation range under the same set of filters (note S4, Supplementary Materials). The identification accuracy is demonstrated to be ~0.19 nm from 500 to 720 nm. It is worth mentioning that we achieve the above performance with a device footprint of only 5 μm by 5 μm (inset of Fig. 2A), which is among the smallest spectroscopic devices so far, and it can be further miniaturized to a few hundred nanometers by optimizing advanced processes (e.g., involving high-resolution lithography and patterning) thanks to its simple configuration. However, we note that the device performance will face challenges when further miniaturized to the nanometer scale, where light absorption, EQE, and gating behaviors are strongly influenced by short channel effects, and surface plasmons at the metal contacts.

To demonstrate the applicability of our configuration for on-chip spectral sensing, we integrate our design with a Si3N4 waveguide (Fig. 2G). The fabrication details are included in the Methods section. We choose graphite as the vdW metal contact to show the flexibility of our configuration. Different from the NbTe2 electrode, the work function of graphite is explicitly higher than the electron affinity of InSe, so the device has less tunability compared with the NbTe2 electrode, as indicated by the signature wavelength-dependent photoresponse mapping in Fig. 2H. With this photoresponse mapping, we can sense and identify unknown input monochromatic light ranging from 500 to 840 nm. The spectral identification results and the reference spectra are attached in Fig. 2I, which has a minimum error of ~0.1 nm (single wavelength), and the identification accuracy is demonstrated to be ~2.45 nm.

Furthermore, we demonstrate a spectroscopic application of material identification (Fig. 1B) by implementing broadband complex spectral sensing with the free-space device (note S5, Supplementary Materials). We begin with learning the photoresponses of different materials [tris(chloropropyl)phosphate, GaSe, Au cluster, carbon nanotube, and SiN], ranging from organic dye, metal, semiconductor, to dielectric materials (Fig. 3A). A photoresponse database is created from the material-encoded spectroscopic signal (excited by a 532-nm laser then filtered by a 533-nm notch filter) of each learnt material. The 532-nm laser excites a broadband complex spectrum (fig. S7) that contains the material information, and the notch filter extracts this spectroscopic signal by removing the high-intensity single frequency excitation (method details in note S5, Supplementary Materials). The learning results are depicted in Fig. 3B. Subsequently, we test unknown materials from the aforementioned set of materials with the same method and device. The outcome electrical signals are then used for computation with results from the learning step. Using our algorithm (flowchart in note S5, Supplementary Materials), a weight coefficient vector of the input unknown material is calculated (different columns in Fig. 3C). In this weight coefficient vector, where each element represents the output probability for each material class, the index of the largest element indicates the identified result of the unknown material, i.e., the material with the same index in the photoresponse mapping. With this method, all input unknown materials are accurately identified (Fig. 3C). In addition to the identification accuracy, the average peak probability across all the tested materials is also calculated with the value of 0.827. Such a value means that the probability of the unknown signal being identified as a specific material is significantly larger than that of other materials, i.e., the identification result is highly unlikely to be confused with other materials in the dataset. These results (including the identification accuracy of 1 and the average peak probability of 0.827) demonstrate the excellent distinguishable ability of our device for various input spectra and the potential for the differentiation of materials with similar optical properties. Furthermore, we conduct a proof-of-concept material identification using five different organic dyes (Fig. 3D) that exhibit similar spectroscopic signals (Fig. 3E). The identification results (Fig. 3F) indicate that our device can distinguish between materials with a PL peak difference of ~2.4 nm. These results indicate that complex spectral sensing and identification render the precise spectrum reconstruction (which is mandatory in traditional spectroscopy tools) unnecessary (note S6, Supplementary Materials). Therefore, such a compact optoelectronic interface, capable of generating distinguishable photoresponse with complex spectra input, can significantly miniaturize traditional spectroscopic tools by eliminating bulky dispersive components (e.g., gratings) and provide a cost-effective solution in computational power for reconstructive spectrometers.

Fig. 3. Complex spectral sensing for material identification.

Fig. 3.

Open in a new tab

(A) Schematic of the complex spectrum sensing and identification with the free-space device. (B) Photoresponse mapping of the studied materials. (C) Results of the material identification with weight coefficient vectors. The largest coefficient indicates the identified material index in the photoresponse mapping. (D) Pictures of five organic dyes under white light illumination (top) and UV light illumination (bottom), including Rhodamine-110, Rhodamine-123, Rhodamine-6G, Rhodamine-B, and Rhodamine-101. (E) PL spectra of the five organic dyes, excited by a 403-nm laser and filtered by a 450-nm long-pass filter. (F) Material identification results with weight coefficient vectors.

DISCUSSION

For spectral sensing and identification, the physical limit is defined as the smallest distinguishable separation between electrical signals. Here, we propose a few key parameters that affect the physical limits, including the noise level of the detected electrical signal, the learning step between two learnt materials, and the uncorrelation between photoresponse curves (Fig. 4A).

Fig. 4. Physical limits of miniaturized spectral sensing.

Fig. 4.

Open in a new tab

(A) Relation between identification accuracy, noise level, learning step in the learning process and the uncorrelation of the photoresponse mapping. (B) Impact of noise level on the identified results with different SNRs. (C) Identification accuracy versus SNR values at different peak wavelengths. Our experimental results are marked by the red stars. (D) Electrical curve and its interpolation with a much larger sampling step of our device (top) and the reference device (bottom). (E) Identified results of our device and the reference device. (F) Future perspectives of our miniaturized spectral sensing with improved uncorrelation. The gray strip indicates traditional tabletop tools.

The presence of noise, quantified as the signal-to-noise ratio (SNR), can cause measured electrical curves to deviate noticeably from learnt curves (Fig. 4B), thereby reducing the device performance. We show that the SNR needs to be maintained above 37 dB (for our test wavelengths of 550 and 600 nm, shadow area in Fig. 4C; see note S7 for details) to facilitate an identification accuracy better than 5% of the device operation range. The noise inevitably emerges in real experiments and originates from various sources, such as input fluctuations, contamination, and defects of the device, and even the inherent noise from the measuring tools (e.g., the ground vibrations and the power source). Notably, the noise tends not to increase linearly with the rise in photocurrent, suggesting that the influence of noise can be alleviated by hiring high-photoresponsivity devices.

The learning step is an important parameter in narrow-band spectral sensing, in which the input signals continuously change (peak wavelengths). The learning step defines the gap between two adjacent learnt items in the photoresponse mapping (e.g., the separation between two central wavelengths of the studied monochromatic light). We show that a small learning step is the prerequisite for high accuracy, but it becomes unnecessary when analytical methods (i.e., interpolation, machine learning, or other advanced methods leveraging prior information) are applicable (Fig. 4, D and E; see note S7 for details). When the photoresponse mapping is low-frequency dominated, it requires much less effort in the learning process to achieve equivalent high accuracy. This feature exhibits practical significance in compound analysis, such as characterizing the fluctuation of precursor ratio when synthesizing quantum dots by identifying the PL peak position, which continuous shifts with the ratio change (see note S7 in the Supplementary Materials).

Mathematically, the high uncorrelation of the photoresponse mapping determines the identification accuracy, but the increasing uncorrelation typically necessitates massive device calibration with much smaller learning steps. The higher uncorrelation alleviates the degree of ill-posedness to enhance immunity to noise, thus achieving more accurate identification (16). However, the high uncorrelation implies that the device should exhibit a fast-changing response with only a slight variation in the input optical signal, which is essentially unattainable in reality. In practical terms, there are a few strategies worth investigating for improving the uncorrelation. For example, introducing negative photoresponses such as negative differential resistance (26, 27) and tunable quantum tunneling (e.g., band-to-band tunneling, direct tunneling, and Fowler-Nordheim tunneling) (8, 28–31) will induce an abrupt change that helps increase variability of the photoresponse curves. Figure 4F illustrates the potential of the identification accuracy with improved uncorrelation. The uncorrelation coefficient is defined by considering both the average correlation coefficient of the columns and the learning step (see note S7, Supplementary Materials).

In summary, we demonstrate miniaturized spectral sensing for both narrow-band and broadband complex spectra, with an electrically tunable compact optoelectronic interface. We achieve an ~0.19-nm peak wavelength identification accuracy in free space, with a device footprint downscale to a mere 5 μm by 5 μm. The on-chip integrated spectral sensing is illustrated with an ~2.45-nm accuracy. We further implement the complex spectral sensing and identification for a cost-effective material identification. Our configuration is universal and applicable to various vdW semiconductors, allowing for flexible device design, compatible fabrication, and mass production. The physical limits are delved by investigating the interplay between noise level, learning step, and uncorrelation in relation to identification accuracy. This work paves the way for miniaturized spectroscopic applications in both free-space and on-chip scenarios, offering large-scale repeatability and compatible manufacturing.

METHODS

Device fabrication

Free-space device: The free-space device is fabricated with the same method used in our recently published articles (20, 21). Thick flakes of NbTe2 and InSe are exfoliated from bulk materials and transferred to a Si wafer covered with 300-nm-thick SiO2, assisted by polydimethylsiloxane stamps. The Ti/Au (5 nm/100 nm) is patterned through electron beam lithography (EBPG 5000, Vistec Electron Beam, Germany), electron beam evaporation (MASA IM-9912), and lift-off in acetone. After the annealing at 180°C in high vacuum (~10−5 mbar) for 2 hours (AML-AWB wafer bonding machine), a 2-nm thick aluminum layer is deposited by electron beam evaporation, followed by atomic layer deposition (ALD) (Beneq TFS-500) of 20-nm-thick AlOx. Last, the second annealing process with the identical condition is conducted. Waveguide-integrated device: The InSe and graphite are exfoliated from bulk materials and transferred on top of a Si3N4 waveguide with ready electrodes. The angle of the flake is controlled during the transfer to ensure that the vdW junction is fully located on the top of the waveguide and partially overlapped. A 50-nm-thick AlOx layer is deposited on the top of the junction via ALD, serving as an insulating layer. The ionic gel is used for applying top gate electric field.

Electrical and optoelectronic measurements

The device is bonded to a printed circuit board, then two source meters Keithley 2401 and Keithley 2400 are used to apply VDS and VGS, and the drain-source current (IDS) is measured by Keithley 2401. The data are collected in a customized LabVIEW program that controls the source meters. The optoelectronic measurement is implemented in a homemade setup (6) that can tune the wavelength and power of the incident light by a tunable filter (SuperK Varia, NKT Photonics) from a supercontinuum laser (SuperK Extreme, NKT Photonics). Because of the limit of the setup, the spectral range studied is set to 500 to 840 nm. A 50x objective [numerical aperture (NA) = 0.42] and a 20x objective (NA = 0.4) are used for light coupling and imaging. Reference spectra are measured by a commercialized spectrometer (6). The monochromatic light has a bandwidth of 10 nm with the peak position considered as the incident light wavelength. All the measurement is accomplished within a narrow input power range, which ensures that the device is working in its linear dynamic range.

Waveguide-integrated device measurement

The device is bonded to a printed circuit board and loaded to a homemade waveguide coupling system. After passing the tunable filter (SuperK Varia, NKT Photonics), the light is coupled into a broadband fiber and delivered to the coupling system. A fiber lens (TPMJ 3S, OZ Optics) is then used to inject light into the waveguide. Electrical measurement is the same as described above.

Material identification

The experiment uses an adapted micro-Raman setup (Witec Alpha300 R). A 532-nm laser is inserted onto the sample. The signal is collected by a 50x lens and coupled to a fiber after passing a 533-nm notch filter and an attenuator and then input into the free-space device. Electrical data are collected as described above.

Acknowledgments

We acknowledge the provision of facilities and technical support from the Otaniemi research infrastructure (OtaNano-Micronova Nanofabrication Centre and OtaNano-Nanomicroscopy Centre). X.C. acknowledges the Vilho, Yrjö and Kalle Väisälä Foundation of the Finnish Academy of Science and Letters.

Funding: This work was supported by the Academy of Finland with grant nos. 314810, 336144, 336818, 352780, 352930, 360411, 359009, 365686, 367808, and 353364 (to Z.S.), no. 333982 (to A.C.L.), no. 340932 (to M.D.), and no. 348920 (to H.H.Y.); the Academy of Finland Flagship Programme (PREIN) with grant no. 320167 (to Z.S.) and 320166 (to I.R.); the EU H2020-MSCA-RISE-872049 with grant IPN-Bio (to Z.S.); the Jane and Aatos Erkko foundation and the Technology Industries of Finland centennial foundation with grant no. Future Makers 2022 (to Z.S.); the National Natural Science Foundation of China (NSFC) with grant no. 52061135108 (to W.C.); the Industrial Strategic Technology Development Program funded by the Korean government Ministry of Trade, Industry, and Energy (MOTIE) with grant no. RS-2024-00431676 (to H.H.Y.); the Electronics and Telecommunications Research Institute (ETRI) Internal Project funded by the Korean government Ministry of Science and ICT (MSIT) with grant no. 24YR1300 (to H.H.Y.); and the ERC with grant no. 834742 (to Z.S.).

Author contributions: X.C. conceived the ideas during the discussion with Z.S. and W.C. X.C. designed the experiments and carried out the characterizations/measurements. M.D., X.C., and F.N. fabricated the free-space devices. X.C. and I.R. fabricated the waveguide. M.G.U., S.M.A., and F.A. helped with the electrical and optoelectrical measurements. X.C., F.N., L.W., W.C., and H.H.Y. developed the identification algorithm. X.C., L.W., F.N., and W.C. carried out the analysis. A.C.L., Y.Z., H.H.Y., H.L., S.H., T.A., Z.Y., T.H., W.C., and Z.S. commented on the experimental results and helped with the data analysis. X.C., L.W., and F.N. designed the graphics and wrote the manuscript. Z.S. supervised the research. All authors participated in the scientific discussion extensively and contributed to the manuscript writing.

Competing interests: The authors declare that they have no competing interests.

Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials.

Supplementary Materials

This PDF file includes:

Supplementary Notes S1 to S8

Figs. S1 to S13

References

sciadv.ado6886_sm.pdf (3.1MB, pdf)

REFERENCES AND NOTES

  • 1.Crocombe R. A., Portable spectroscopy. Appl. Spectrosc. 72, 1701–1751 (2018). [DOI] [PubMed] [Google Scholar]
  • 2.Yuan S., Ma C., Fetaya E., Mueller T., Naveh D., Zhang F., Xia F., Geometric deep optical sensing. Science 379, eade1220 (2023). [DOI] [PubMed] [Google Scholar]
  • 3.Yang Z., Albrow-Owen T., Cai W., Hasan T., Miniaturization of optical spectrometers. Science 371, eabe0722 (2021). [DOI] [PubMed] [Google Scholar]
  • 4.Wang J., Zheng B., Wang X., Strategies for high performance and scalable on-chip spectrometers. JPhys. Photonics 3, 012006 (2021). [Google Scholar]
  • 5.Yang Z., Albrow-Owen T., Cui H., Alexander-Webber J., Gu F., Wang X., Wu T.-C., Zhuge M., Williams C., Wang P., Zayats A. V., Cai W., Dai L., Hofmann S., Overend M., Tong L., Yang Q., Sun Z., Hasan T., Single-nanowire spectrometers. Science 365, 1017–1020 (2019). [DOI] [PubMed] [Google Scholar]
  • 6.Yoon H. H., Fernandez H. A., Nigmatulin F., Cai W., Yang Z., Cui H., Ahmed F., Cui X., Uddin M. G., Minot E. D., Lipsanen H., Kim K., Hakonen P., Hasan T., Sun Z., Miniaturized spectrometers with a tunable van der Waals junction. Science 378, 296–299 (2022). [DOI] [PubMed] [Google Scholar]
  • 7.Wu G., Abid M., Zerara M., Cho J., Choi M., Coileáin C. Ó., Hung K.-M., Chang C.-R., Shvets I. V., Wu H.-C., Miniaturized spectrometer with intrinsic long-term image memory. Nat. Commun. 15, 676 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Uddin M. G., Das S., Shafi A. M., Wang L., Cui X., Nigmatulin F., Ahmed F., Liapis A. C., Cai W., Yang Z., Lipsanen H., Hasan T., Yoon H. H., Sun Z., Broadband miniaturized spectrometers with a van der Waals tunnel diode. Nat. Commun. 15, 571 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Deng W., Zheng Z., Li J., Zhou R., Chen X., Zhang D., Lu Y., Wang C., You C., Li S., Sun L., Wu Y., Li X., An B., Liu Z., Wang Q. J., Duan X., Zhang Y., Electrically tunable two-dimensional heterojunctions for miniaturized near-infrared spectrometers. Nat. Commun. 13, 4627 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Wang J., Pan B., Wang Z., Zhang J., Zhou Z., Yao L., Wu Y., Ren W., Wang J., Ji H., Yu J., Chen B., Single-pixel p-graded-n junction spectrometers. Nat. Commun. 15, 1773 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Yu Y., Zhong M., Xiong T., Yang J., Hu P., Long H., Zhou Z., Xin K., Liu Y.-Y., Yang J., Qiao J., Liu D., Wei Z., Spectrometer-less remote sensing image classification based on gate-tunable van der Waals heterostructures. Adv. Sci. 11, 2309781 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Darweesh R., Yadav R. K., Adler E., Poplinger M., Levi A., Lee J.-J., Leshem A., Ramasubramaniam A., Xia F., Naveh D., Nonlinear self-calibrated spectrometer with single GeSe-InSe heterojunction device. Sci. Adv. 10, eadn6028 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Yuan S., Naveh D., Watanabe K., Taniguchi T., Xia F., A wavelength-scale black phosphorus spectrometer. Nat. Photonics 15, 601–607 (2021). [Google Scholar]
  • 14.Cui X., Zhang Y., Liapis A. C., Sun Z., Reconstructive spectrometers taper down in price. Light Sci. Appl. 12, 142 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Fan Y., Huang W., Zhu F., Liu X., Jin C., Guo C., An Y., Kivshar Y., Qiu C.-W., Li W., Dispersion-assisted high-dimensional photodetector. Nature 630, 77–83 (2024). [DOI] [PubMed] [Google Scholar]
  • 16.P. C. Hansen, Rank-Deficient and Discrete Ill-Posed Problems: Numerical Aspects of Linear Inversion, SIAM monographs on mathematical modeling and computation (SIAM, 1998). [Google Scholar]
  • 17.Majidi L., Yasaei P., Warburton R. E., Fuladi S., Cavin J., Hu X., Hemmat Z., Cho S. B., Abbasi P., Vörös M., Cheng L., Sayahpour B., Bolotin I. L., Zapol P., Greeley J., Klie R. F., Mishra R., Khalili-Araghi F., Curtiss L. A., Salehi-Khojin A., New class of electrocatalysts based on 2D transition metal dichalcogenides in ionic liquid. Adv. Mater. 31, e1804453 (2019). [DOI] [PubMed] [Google Scholar]
  • 18.Bandurin D. A., Tyurnina A. V., Yu G. L., Mishchenko A., Zólyomi V., Morozov S. V., Kumar R. K., Gorbachev R. V., Kudrynskyi Z. R., Pezzini S., Kovalyuk Z. D., Zeitler U., Novoselov K. S., Patanè A., Eaves L., Grigorieva I. V., Fal’ko V. I., Geim A. K., Cao Y., High electron mobility, quantum Hall effect and anomalous optical response in atomically thin InSe. Nat. Nanotechnol. 12, 223–227 (2017). [DOI] [PubMed] [Google Scholar]
  • 19.Man M. K. L., Margiolakis A., Deckoff-Jones S., Harada T., Wong E. L., Krishna M. B. M., Madéo J., Winchester A., Lei S., Vajtai R., Ajayan P. M., Dani K. M., Imaging the motion of electrons across semiconductor heterojunctions. Nat. Nanotechnol. 12, 36–40 (2017). [DOI] [PubMed] [Google Scholar]
  • 20.Du M., Cui X., Yoon H. H., Das S., Uddin M. G., Du L., Li D., Sun Z., Switchable photoresponse mechanisms implemented in single van der Waals semiconductor/metal heterostructure. ACS Nano 16, 568–576 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Du M., Cui X., Zhang B., Sun Z., Deterministic light-to-voltage conversion with a tunable two-dimensional diode. ACS Photonics 9, 2825–2832 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Mudd G. W., Svatek S. A., Ren T., Patanè A., Makarovsky O., Eaves L., Beton P. H., Kovalyuk Z. D., Lashkarev G. V., Kudrynskyi Z. R., Dmitriev A. I., Tuning the bandgap of exfoliated InSe nanosheets by quantum confinement. Adv. Mater. 25, 5714–5718 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Cui X., Du M., Das S., Yoon H. H., Pelgrin V. Y., Li D., Sun Z., On-chip photonics and optoelectronics with a van der Waals material dielectric platform. Nanoscale 14, 9459–9465 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Sucharitakul S., Goble N. J., Kumar U. R., Sankar R., Bogorad Z. A., Chou F.-C., Chen Y.-T., Gao X. P. A., Intrinsic electron mobility exceeding 103 cm2/(V s) in multilayer InSe FETs. Nano Lett. 15, 3815–3819 (2015). [DOI] [PubMed] [Google Scholar]
  • 25.Liu L., Wu L., Wang A., Liu H., Ma R., Wu K., Chen J., Zhou Z., Tian Y., Yang H., Shen C., Bao L., Qin Z., Pantelides S. T., Gao H.-J., Ferroelectric-gated InSe photodetectors with high on/off ratios and photoresponsivity. Nano Lett. 20, 6666–6673 (2020). [DOI] [PubMed] [Google Scholar]
  • 26.Jariwala D., Howell S. L., Chen K.-S., Kang J., Sangwan V. K., Filippone S. A., Turrisi R., Marks T. J., Lauhon L. J., Hersam M. C., Hybrid, gate-tunable, van der Waals p–n heterojunctions from pentacene and MoS2. Nano Lett. 16, 497–503 (2016). [DOI] [PubMed] [Google Scholar]
  • 27.Shin J.-C., Kim Y. H., Watanabe K., Taniguchi T., Lee C.-H., Lee G.-H., Band structure engineering of WSe2 homo-junction interfaces via thickness control. Adv. Mater. Interfaces 9, 2101763 (2022). [Google Scholar]
  • 28.Sun X., Zhu C., Zhu X., Yi J., Liu Y., Li D., Pan A., Recent advances in two-dimensional heterostructures: From band alignment engineering to advanced optoelectronic applications. Adv. Electron. Mater. 7, 2001174 (2021). [Google Scholar]
  • 29.Kim S., Myeong G., Shin W., Lim H., Kim B., Jin T., Chang S., Watanabe K., Taniguchi T., Cho S., Thickness-controlled black phosphorus tunnel field-effect transistor for low-power switches. Nat. Nanotechnol. 15, 203–206 (2020). [DOI] [PubMed] [Google Scholar]
  • 30.Xiong X., Huang M., Hu B., Li X., Liu F., Li S., Tian M., Li T., Song J., Wu Y., A transverse tunnelling field-effect transistor made from a van der Waals heterostructure. Nat. Electron. 3, 106–112 (2020). [Google Scholar]
  • 31.Yoon H. H., Ahmed F., Dai Y., Fernandez H. A., Cui X., Bai X., Li D., Du M., Lipsanen H., Sun Z., Tunable quantum tunneling through a graphene/Bi2Se3 heterointerface for the hybrid photodetection mechanism. ACS Appl. Mater. Interfaces 13, 58927–58935 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Kruskal J. B., Three-way arrays: Rank and uniqueness of trilinear decompositions, with application to arithmetic complexity and statistics. Linear Algebra Appl. 18, 95–138 (1977). [Google Scholar]
  • 33.S. Foucart, H. Rauhut, “An invitation to compressive sensing” in A Mathematical Introduction to Compressive Sensing (Springer, 2013). [Google Scholar]
  • 34.Liu Y., Guo J., Zhu E., Liao L., Lee S.-J., Ding M., Shakir I., Gambin V., Huang Y., Duan X., Approaching the Schottky–Mott limit in van der Waals metal–semiconductor junctions. Nature 557, 696–700 (2018). [DOI] [PubMed] [Google Scholar]
  • 35.Hu S., Luo X., Xu J., Zhao Q., Cheng Y., Wang T., Jie W., Castellanos-Gomez A., Gan X., Zhao J., Reconfigurable InSe electronics with van der Waals integration. Adv. Electron. Mater. 8, 2101176 (2022). [Google Scholar]
  • 36.Feng W., Wu J.-B., Li X., Zheng W., Zhou X., Xiao K., Cao W., Yang B., Idrobo J.-C., Basile L., Tian W., Tan P.-H., Hu P., Ultrahigh photo-responsivity and detectivity in multilayer InSe nanosheets phototransistors with broadband response. J. Mater. Chem. C 3, 7022–7028 (2015). [Google Scholar]
  • 37.Li Z., Hu S., Zhang Q., Tian R., Gu L., Zhu Y., Yuan Q., Yi R., Li C., Liu Y., Hao Y., Gan X., Zhao J., Telecom-band waveguide-integrated MoS2 photodetector assisted by hot electrons. ACS Photonics 9, 282–289 (2022). [Google Scholar]
  • 38.Darwan D., Lim L. J., Wang T., Wijaya H., Tan Z.-K., Ultra-confined visible-light-emitting colloidal indium arsenide quantum dots. Nano Lett. 21, 5167–5172 (2021). [DOI] [PubMed] [Google Scholar]
  • 39.Kwak D., Polyushkin D. K., Mueller T., In-sensor computing using a MoS2 photodetector with programmable spectral responsivity. Nat. Commun. 14, 4264 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Wang C.-Y., Liang S.-J., Wang S., Wang P., Li Z., Wang Z., Gao A., Pan C., Liu C., Liu J., Yang H., Liu X., Song W., Wang C., Cheng B., Wang X., Chen K., Wang Z., Watanabe K., Taniguchi T., Yang J. J., Miao F., Gate-tunable van der Waals heterostructure for reconfigurable neural network vision sensor. Sci. Adv. 6, eaba6173 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Lee S., Peng R., Wu C., Li M., Programmable black phosphorus image sensor for broadband optoelectronic edge computing. Nat. Commun. 13, 1485 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Wan T., Shao B., Ma S., Zhou Y., Li Q., Chai Y., In-sensor computing: Materials, devices, and integration technologies. Adv. Mater. 35, e2203830 (2023). [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Notes S1 to S8

Figs. S1 to S13

References

sciadv.ado6886_sm.pdf (3.1MB, pdf)

Articles from Science Advances are provided here courtesy of American Association for the Advancement of Science

RESOURCES