Near-field resonant photon sorting applied: dual-band metasurface quantum well infrared photodetectors for gas sensing

1 Introduction

Non-dispersive infrared (NDIR) sensing, widely applied in industrial diagnostics and process control, measures gas concentrations based on mid-infrared absorption lines unique to molecules [1], [2]. In conventional NDIR, broadband emission traverses through the gas specimen and two wavelengths, a target gas absorption line and a non-absorbing reference, are picked out by bulky, spatially separated narrowband filters and passed to two wideband detectors for transmission measurement [1], [2]. The target gas concentration is determined from the detectors' signal ratio to cancel the drift in incoming intensity. In general, relatively slow thermal detectors are applied, measuring concentration from the change in temperature after converting absorbed light into heat. However, NDIR for transient phenomena demanding fast (~1 ms) and strong response, such as in-situ combustion engine diagnostics [3], [, 4], requires photon detectors based on electron transitions in semiconductors.

Metal-dielectric-metal (MDM) metasurface plasmon cavities, precisely controllable narrowband absorbers [5], [6] (??/?_peak = 0.1, where ?? is the full-width half maximum and ?_peak is the peak wavelength of the cavity), offer a simpler NDIR approach [7] without external filters; vertically stacked MDM cavities can exhibit near-perfect absorption of normally incident light via magnetic coupling [5]. Using an MDM detector resonant at 4.28 5m, Lochbaum et al. [8] recently demonstrated filterless NDIR sensing of CO₂ gas. However their system was based on an incomplete NDIR configuration; only a single wavelength was used, thus, subject to intensity drift. Furthermore, the MDM cavities were used only as narrowband absorbers atop a wideband thermal detector, therefore the response time was limited.

Instead of a single type of MDM cavity, metasurfaces combining multiple types of cavities can realize more complex interactions with light. Resonant photon sorting in metasurfaces made of paired MDM cavities offers a route to recover dual-band functionality necessary for NDIR [9], [10], [11], [12]. When two cavities with distinct fundamental resonances are alternately arranged in a subwavelength period, incident photons with different wavelengths are laterally sorted into corresponding cavities within a near field. Because each deep subwavelength cavity has an absorption cross section much larger than its geometrical area [13], [14], [15], [16], [17], each cavity realizes nearly perfect absorption of light at its resonance wavelength, without significant interference from its counterpart. Sorting efficiencies, the fraction of the absorption in the resonant cavity over the total absorption in the two cavities, as high as 95% have been observed [9], [12], [18].

By incorporating an infrared semiconductor detector as the dielectric layer of these MDM cavities [9], [10], two fast response detectors for different wavelengths, integrated in the footprint of one, can be fabricated. Because enhanced intensity in the cavities enables us to minimize the semiconductor thickness, high responsivities (signal) with reduced dark currents (noise) can be realized [19], [20], leading to enhanced signal-to-noise ratio, detectivity. Thus by employing resonant photon sorting, NDIR detection with no filters, with complete dual-band functionality, can be realized in a single device. Although a preliminary photon sorting detector based on paired trench cavities has been reported [18], neither clear dual-band operation nor sufficient responsivities for real-world application could be shown.

Here, we demonstrate a dual-band photodetector for NDIR gas sensing based on resonant photon sorting consisting of two electrically isolated, interdigitated stripe MDM cavities sandwiching quantum well infrared photodetectors (QWIPs), metasurface QWIPs [21], [22], [23], [24], [25], with different resonance wavelengths. Metasurface QWIPs are based on intersubband transitions (ISBT) in quantum wells. An intense vertical electric field, essential for ISBT absorption [26], [27], is provided by the MDM cavity through rotation of the incident horizontal electric field to vertical, and intensity enhancement [14], [28], [29]. High responsivities [23], fast response times [24], [30], and detectivities approaching theoretical limits [21], [24], [25], [31], sufficient to replace toxic HgCdTe presently dominant at wavelengths beyond 5 5m, have been demonstrated.

Our dual-band metasurface QWIPs show a maximum unpolarized responsivity of 2.1 A/W at a peak wavelength of 6.9 5m (external quantum efficiency: 38%), comparable to previously demonstrated single-band stripe metasurface QWIPs [21]. Tailoring one resonance to an absorption peak of NO₂ (6.25 5m) and the other to a non-absorbing reference wavelength (7.15 5m), we determine NO₂ gas concentration with 10 ppm accuracy and 1 ms response times, suitable for transient gas sensing applications demanding fast and accurate detection such as combustion engine diagnostics. Through systematic experiments and calculations, resonant photon sorting behavior, nearly perfect absorption at the resonant cavity, and suppressed absorption in the non-resonant cavity are confirmed and graphically visualized. Our dual-band metasurface QWIPs represent a high-performance, flexible approach to photodetection, consolidating the functions of two separate detectors and filters into a single device. Both the cavity and QWIP resonance can be fully controlled, enabling fully engineered detection across the mid-infrared spectrum.

2 Results

The QWIP layer used in this study consists of a single 4 nm GaAs quantum well layer sandwiched by two 50 nm Al_0.3Ga_0.7As barrier layers and two 48 nm GaAs contact layers [21], [25], [32], a total thickness T = 200 nm, which exhibits a photoconductive responsivity peak at 6.7 5m [21], [25]. Stripe cavities consist of a t_m1 = 100 nm thick Au stripe and a thick Au substrate (650 nm, but set at t_m2 = 200 nm for calculations) sandwiching the QWIP layer. Each stripe cavity is called a QWIP stripe from here on. Stripes with widths of L₁ and L₂ (L₁ < L₂) are alternated with a period P/2 (Figure 1A). The QWIP layer between the QWIP stripes is removed [21], [25] for electrical isolation between neighboring stripes. The resonance wavelengths ?₁ and ?₂, determined by L₁ and L₂, respectively, should fall within the absorption range of the quantum well (6.0-7.4 5m). The resonance occurs only for x-polarized incidence.

Figure 1:

Design and fabrication of dual-band metasurface quantum well infrared photodetectors (QWIPs). A) Cross section showing critical dimensions of individual QWIP stripes. E_x and k indicate the electric field and wave vector of incident light. Quantum well (QW) position is indicated by lines in the semiconductor (blue) layer. B) Simulated spectra showing absorption peaks of QWIP stripes with L₁ and L₂ and target wavelengths 6.25 5m (blue) and 7.15 5m (orange). The red curve is NO₂ absorbance at 100 ppm-m in 1 atm N₂ at 25 0C (right axis). C) Colorized SEM image of QWIP stripes in a dual-band device. Gold indicates the Au substrate and top stripe. Blue indicates the QWIP layer.

For demonstration of NO₂ sensing afterward, resonance wavelengths of ?₁ = 6.25 5m and ?₂ = 7.15 5m are chosen to match the NO₂ absorption peak and provide a non-absorbing reference wavelength, respectively. Near-perfect absorption (~99%) at these wavelengths is predicted in cavities with L₁ = 0.87 5m and L₂ = 1.02 5m for a period P = 3.1 5m (Figure 1B), based on calculations assuming the QWIP layer to be a homogeneous dielectric with a refractive index n = 3.05 + 0.10i taken from preliminary experiments [21], [25]. All calculations in this work are based on rigorous coupled wave analysis (RCWA). Several dual-band devices with QWIP stripes around these L₁ and L₂ values were fabricated following previously reported methods [21], [25] (Figure 1C).

Two interdigitated single-band detectors, called channel 1 (Ch. 1) and channel 2 (Ch. 2), make up a dual-band metasurface QWIP, sometimes simply called a device. In each device, QWIP stripes with the same L value are connected to an electrode pad at one end for independent biasing and current collection. When a bias voltage is applied across the QWIP layer and light within absorption range impinges on each detector, photocurrent is generated (Figure 2A). Based on the principles of resonant photon sorting [9], [10], [11], [12], light with wavelengths of ?₁ (blue) and ?₂ (red) should be selectively absorbed in the corresponding resonant cavity (L₁ or L₂).

Figure 2:

A) Schematic of device operating principle. One device consists of two detectors (Ch. 1, Ch. 2) generating independent photocurrents (I₁, I₂), composed of electrically isolated interdigitated quantum well infrared photodetector (QWIP) stripes with different dimensions (L₁, L₂) selectively absorbing at their corresponding resonance wavelengths (?₁, ?₂). B) Experimental and simulated unpolarized responsivity of QWIP stripes from three dual-band devices. Brackets indicate which detectors correspond to which output signal in a device.

Responsivities R for unpolarized incidence for three devices-Device 1, 2, and 3 with (L₁, L₂) = (0.85, 1.04), (0.89, 1.07), and (0.92, 1.11) in 5m, respectively-evaluated at 78 K are compared with simulated results in Figure 2B. Simulated R was determined as

where A_ISBT is the ISBT absorption of the quantum well calculated for x-polarization [21], [25], g (=2.6) the photoconductive gain previously determined for unpaired etched QWIP stripes [21], ? the wavelength, e electron charge, h Planck's constant, and c the speed of light. The factor of two indicates that the stripe cavity has responsivity only for x-polarized incidence. The term AISBTWg/2 shows the external quantum efficiency for unpolarized incidence [26], [27]. Simulated R in this work was based on previously determined experimental R of the QWIP layer itself evaluated by Brewster-angle incidence [21], [25].

For all three devices, the solid lines in Figure 2A and values in Table 1 indicate that the QWIP stripes with L₂ have maximum experimental peak responsivities (R_peak) as high as 2.1 A/W (L₂ = 1.04 5m), comparable to the maximum reported value for single-band stripe metasurface QWIPs (R = 2.3 A/W, background-limited detectivity DBG*=3.3W1010cmHz/W) [21], and R_peak gradually decreases with increasing L. The QWIP stripes with L₁ have much lower R_peak than those with L₂, and exhibit increasing R_peak with increasing L. Fairly good agreement between experiment and simulation (dashed lines, Figure 2A) near 7 5m, and overestimated R in the simulation compared to experiment near 6 5m is consistent with what Ref. 21 reported on etched stripe metasurface QWIPs. This characteristic relationship between experiment and simulation was attributed to changes in the R profile of the QWIP layer due to the metasurface fabrication and insensitive regions at the etched surfaces of the stripes [21].

In each device, when one QWIP stripe exhibits a peak responsivity R_peak at its resonance wavelength ?₁ or ?₂ (bold R_?1, R_?2, Table 1), its counterpart shows greatly reduced R at the same wavelength (non-bold R_?1, R_?2, Table 1), leading to a high contrast in R at these wavelengths (Figure 2A, Table 1). Simulated spectra show similar trends. Resonant absorption in each QWIP stripe proceeds with minimal interference from its counterpart.

Numerical simulation clarifies how photons are sorted into resonant QWIP stripes. |E_z|² in Device 2 for x-polarized incidence at simulated A_ISBT peak wavelengths ?₁ = 6.30 5m and ?₂ = 7.15 5m (Table 1) is shown in Figure 3A, B. Strong |E_z|² is observed in only the resonant QWIP stripes, manifesting the selective coupling of incident light into the corresponding cavities (Figure 3A, B).

Figure 3:

Photon sorting over two periods P for Device 2 (L₁ = 0.89 5m, L₂ = 1.07 5m, and P = 3.1 5m) at ? = 6.30 5m and ? = 7.15 5m, respectively. A, B) Calculated |E_z|² in Device 2. E_x and k indicate the electric field and wave vector of incident light. C, D) Power flow in Device 2. Stream lines are spaced every 100 nm at z = 1100 nm. Solid black lines and black arrows describe the sorting efficiency of each cavity over one period P (white arrows). Dashed lines and red arrows define the absorption by the quantum well.

Power flow [13], [14], [16], [25] clearly visualizes how photons with different wavelengths are routed into the corresponding cavities (Figure 3C, D). Stream lines of the time-averaged Poynting vector for x-polarized incidence show that the vertically incident light starts to deflect around z = 800 nm and is finally collected into the resonant cavities almost perfectly (Figure 3C, D). The adjacent QWIP stripe exhibits negligible absorption. The solid black curves and the black arrows at the top of Figure 3C, D illustrate each cavity's sorting efficiency [9], [12], [18] over one period P. These graphical sorting efficiencies of 95% at both ?₁ = 6.30 5m and ?₂ = 7.15 5m are consistent with the numerically obtained absorption in individual cavities.

However, this does not simply show the light absorbed in the quantum well. Most stream lines terminate in the Au stripe and Au substrate without any contribution to the photocurrent. The dashed black curves and red arrows at the top of Figure 3C, D define the fraction of total absorbed incidence collected in the quantum well over one period P. These fractions agree with the calculated A_ISBT corresponding to the simulated spectra of Figure 2B (22 and 26% for L₁ and L₂, respectively, in Device 2), but are lower than the total sorting efficiency of the cavity, similar to previous reports for other metasurface QWIPs [21], [25]. Nevertheless, the single quantum well design of our QWIP layer [25] allows us to realize high g and quantum efficiencies, leading to the high R_peak values predicted from Equation (1).

The observed contrast in R between QWIP stripes and photon sorting behavior from power flow suggest our dual-band device can operate effectively as an NDIR sensor. As mentioned previously, we chose NO₂ as our target molecule for demonstration. Emission control of NO₂, the most toxic component of NO_x, remains a critical topic for automotive engineering research [33], [34], [35]. A typical combustion engine emits NO₂ on the order of 10¹-10³ ppm within a cycle of 50-80 ms, demanding ppm accuracy and ms response time to investigate the generation process of the NO₂ gas [34], [35].

Presently, state-of-the-art NO_x analysis with 5 ppm accuracy [36] and 2 ms response times [34] for engine development is conducted by chemiluminesence [34], using a chemical reaction of NO and O₃. However, the present technique requires extraction of gas from the exhaust port to a separate reaction chamber, and also a NO₂-to-NO converter before the chamber.

Our aim here is to demonstrate similar performance using NDIR NO₂ sensing with our dual-band detector, which integrates the functions of two bandpass filters and two infrared photodetectors into a single, much smaller, monolithic device. Compared to chemiluminesence systems, NDIR can use the gas channel from the exhaust port as the optical path for directly measuring NO₂ without any need for a special reaction chamber and a gas converter.

In Device 2, Ch. 1 corresponds to the NO₂ absorption wavelength at 6.25 5m, while Ch. 2 is used as the reference at 7.15 5m (Figure 4A). Based on calibration experiments for known NO₂ concentrations, the NO₂ concentration can be determined from the ratio of photocurrent I_P in Ch. 1/Ch. 2 (Supplementary information S1, Figure S1A) with a standard deviation of 18 ppm (Figure S1B). To clearly demonstrate ms response in our detectors, we introduce a pulse of NO₂ (Figure 4B, Methods) and simultaneously observe changes in Ch. 1 and Ch. 2 response to track NO₂ concentration. Figure 4C shows a sharp decrease in I_P in Ch. 1 from the introduced NO₂ followed by a more gradual return to the original signal, while I_P in Ch. 2 remains nearly constant. Fitting to the rise of this curve gives a time constant of 3.2 ms (Figure 4D).

Figure 4:

Dual-channel sensing using a dual-band metasurface quantum well infrared photodetector (QWIP) (Device 2). A) Experimentally determined resonance of Ch. 1 (L₁ = 0.89 5m) is matched to NO₂ absorption peak at ? = 6.25 5m. Ch. 2 (L₂ = 1.07 5m) is the non-absorbing reference. NO₂ absorbance at 100 ppm-m in 1 atm N₂ at 25 0C (right axis). B) Experimental setup for fast response NO₂ concentration measurement. C) Response to rapid concentration changes in each channel of Device 2. Photocurrent I_P in Ch. 2 (top) and Ch. 1 (middle) measured over multiple NO₂ pulses. The signal ratio (bottom) gives NO₂ concentration. D) Fitting the rise associated with a single pulse to determine the time constant.

This speed is only limited by the gas supply system, not the detector. Using another experimental setup with a much faster gas supply, we confirmed a 0.83 ms response time with the same detectors (Supplementary information S2, Figure S2), clearly indicating that our device performance already approaches state-of-the-art benchmarks.

For further optimized NO₂ sensing, we can increase the relatively low R of Ch. 1 (Figure 4A) by redesigning the quantum well for maximum sensitivity at 6.25 5m. In addition, since the peak wavelength in both channels can be precisely controlled by the design of the MDM cavity and quantum well, our dual-band can be applied for other gases.

3 Discussion

Our dual-band device consolidates the function of two single-band detectors in the footprint of one; each QWIP stripe in a dual-band device operates as if their counterpart is not present. Here, we examine the influence of the counterpart stripe more straightforwardly and quantitatively by comparing paired QWIP stripes from Devices 1 to 3 with unpaired QWIP stripes having the same (or very close) L values but no counterpart from our previous study [21] on single-band devices. Unpaired single-band devices were designed (P = 2.4 5m) for optimized R [21].

Figure 5 shows experimental total absorption (A_TOT) for paired QWIP stripes from Devices 1 to 3 (black) and corresponding unpaired QWIP stripes (colored). Maximum values of A_TOT, corresponding to resonant absorption in one of the paired QWIP stripes show near-perfect correspondence in position, magnitude and overall profile with unpaired QWIP stripes with the same L. At the resonance wavelengths of paired QWIP stripes, each resonant stripe acts like an unpaired QWIP stripe, absorbing almost all light, as if their counterpart was not present.

Figure 5:

Comparison of paired quantum well infrared photodetector (QWIP) stripes from dual-band devices (black) and unpaired QWIP stripes with identical L (colored). A) One period of each structure shown schematically for comparison. B-D) Experimental total absorption (A_TOT) taken from s-polarized reflection (Incidence angle ? = 260 in the y-z plane) of Devices 1, 2 and 3 along with corresponding unpaired QWIP stripes. L of the unpaired QWIP stripes corresponds with either L₁ or L₂ in the paired QWIP stripe. An unpaired QWIP stripe with L =0.93 5m is shown to be compared with the paired QWIP stripe with L₁ = 0.92 5m due to lack of an unpaired QWIP stripe at the exact dimensions. All A_TOT is calibrated, thus independent of the device area.

Highly selective resonant absorption in paired QWIP stripes is further confirmed from measured R. In Figure 6A, experimental R spectra of paired (solid) and unpaired (dotted) QWIP stripes again show nearly identical peak position, magnitude, and overall profile. In dual-band devices, each QWIP stripe maintains its original performance as a single-band device without essential degradation.

Figure 6:

A) Experimental unpolarized R of paired quantum well infrared photodetector (QWIP) stripes from dual-band devices (solid) and unpaired QWIP stripes with identical L (dotted). One period of each structure is shown schematically for comparison. An unpaired QWIP stripe with L =0.93 5m is shown to be compared with the paired QWIP stripe with L₁ = 0.92 5m due to lack of an unpaired QWIP stripe at the exact dimensions. B) Close-up of A) for respective devices. Non-resonant R in paired QWIPs around their counterpart's resonance wavelength (black vertical lines). Black filled circles indicate values of experimental non-resonant R in paired QWIP stripes. Black unfilled circles indicate values of non-resonant R corresponding unpaired QWIP stripes (dotted lines). All experimental R is calibrated, thus independent of the device area.

Away from resonance, a critical difference between the paired and unpaired QWIP stripes, also characteristic of photon sorting, can be observed (Figure 6B). In all three dual-band devices, around its counterpart's resonance wavelength (black vertical lines), R in a non-resonant QWIP stripe is significantly suppressed (filled circles) compared to an unpaired QWIP stripe with the same L (unfilled circles)-behavior also observed in previous reports on photon sorting [9], [12], [18]. Instead, these photons are absorbed in the resonant QWIP stripe. This can be interpreted that the persistent absorption of the resonant QWIP stripe and the high contrast in R between the paired QWIP stripes are supported by suppressing the absorption of the non-resonant counterpart. Combined, the enhanced resonant absorption and suppressed non-resonant absorption lead paired QWIP stripes to behave as if their counterpart was not present, allowing us to consolidate two single-band detectors in the footprint of one.

Throughout this work, low experimental R in the QWIP stripes with L₁ resonant around 6 5m has been noticeable. To increase R at this wavelength range, we considered increasing the number of stripes resonant at shorter wavelengths in one detector. We fabricated dual-band devices with two L₁ QWIP stripes and one L₂ QWIP stripe, with a larger P = 3.75 5m to accommodate the third stripe (Supplementary information S3, Figure S3). However, we could not realize significantly increased R in the QWIP stripes with L₁, but instead observed a red-shift of ?₁ from 6.05 to 6.40 5m. These results are presumed to indicate the near-field coupling between two L₁ QWIP stripes (Figures S4, S5). Because adding an additional stripe did not lead to enhanced R, we should instead redesign our quantum well to improve R around 6 5m.

Finally, we would like to remark on the optical cross section of each QWIP stripe. In the power flow in Figure 3 and Figure S5, the numerically visualized absorption cross section of each stripe is much larger than its geometrical area. This is also evidenced in the experimental responsivity spectra. In Figure 6A, unpaired QWIP stripes exhibited a very similar R spectrum to that of paired QWIP stripe, irrespective of the difference in P (2.4 and 3.1 5m for the unpaired and paired QWIP stripes, respectively). Also in Figure S4B, the QWIP stripe with L₂ in a three-stripe device showed a very similar R with that of a QWIP stripe from a one-stripe device despite an even greater difference in P (2.4 and 3.75 5m for one- and three-stripe devices, respectively).

Responsivities are independent of device areas. Therefore, R for one- (unpaired), two- (paired), and three-stripe devices show the photocurrents for the same area. As P increases from 2.4 to 3.1 5m and 3.75 5m, the number of resonant QWIP stripes per area correspondingly decreases. Nonetheless, devices with fewer resonant stripes maintain a nearly constant R. This means that resonant QWIP stripes in two- and three-stripe detectors with a larger P are generating greater photocurrent per stripe than one-stripe detectors; i.e., they are collecting incident light from wider areas. Experimentally observed consistent R for one-, two-, and three-stripe devices is direct evidence of the numerically visualized optical cross sections.

In addition, these results reveal another essential conclusion-the optical cross section of a cavity is not necessarily unique to the cavity itself, but depends on the environment of the cavity. When another cavity is placed between equivalent cavities, the original cavities can have much larger cross sections.

4 Conclusion

In summary, we have demonstrated dual-band, dual-channel photodetection based on resonant photon sorting in metasurface QWIPs made of paired stripe cavities. Individual QWIP stripes with responsivities up to 2.1 A/W at 6.9 5m at 78 K, comparable with previous high-detectivity, single-band devices have been demonstrated. Our devices operate as dual-band integrated NDIR sensors combining the function of two filters and two detectors in the footprint of a single device. Tailoring one detector to the specific absorption wavelengths of NO₂, concentration measurement with 10 ppm accuracy and 1 ms response time, comparable with state-of-the-art chemiluminescence systems, has been demonstrated. Through electric field distribution, power flow, and comparison with unpaired stripes, we observed how photon sorting behaviors lead to efficient resonant absorption from a wider area and suppressed non-resonant absorption in the paired stripes. These behaviors ensure high R is maintained even for devices with larger numbers of different cavities.

Conventionally, dual-band infrared photodetectors have been achieved by sequential epitaxial growth of semiconductor crystals such as toxic HgCdTe and QWIPs [26], [27], [37] with different compositions. The responsivity spectrum of each band has been determined by the band structure of the individual crystal. In contrast, resonant photon sorting realizes dual-band detection with a high flexibility with a single semiconductor crystal; the responsivity spectra are defined by the metasurface design. By employing photon sorting, the flow of the incident photons are laterally manipulated in the near field region, and finally converted to flow of electrons in a desired channel.

Engineering of both electrons and photons, as in metasurface QWIPs, would suggest a promising frontier in optoelectronic devices. Although intensive engineering of photonic structures was demonstrated in this work, tailoring electronic structures would be another important topic. For instance, by replacing the photoconductive QWIP with a photovoltaic QWIP [31], [38], room temperature operation with sufficient detectivity could be achieved, greatly increasing the functionality of our devices.

5 Methods