Towards Integrated Mid-Infrared Gas Sensors

Optical gas sensors play an increasingly important role in many applications. Sensing techniques based on mid-infrared absorption spectroscopy offer excellent stability, selectivity and sensitivity, for numerous possibilities expected for sensors integrated into mobile and wearable devices. Here we review recent progress towards the miniaturization and integration of optical gas sensors, with a focus on low-cost and low-power consumption devices.


Introduction
Gas sensors are used in a variety of scientific, industrial and commercial applications [1]. Among various sensing techniques [2], sensors based on the interaction of light with gas molecules [3], can offer high sensitivity [4,5], and long-term operation stability [5]. In addition, they have longer lifetimes and shorter response times [3,6], compared to other techniques [2], making them suitable for real-time [7], and in situ [8] detection. Most optical gas sensors rely on absorption spectroscopy [3], where a gas is detected by measuring the light absorbed (due to its interaction with the gas) as a function of wavelength [9]. Many important organic and inorganic molecules [9] have characteristic absorption lines in the mid-infrared (MIR) spectral region (λ ∼ 2-20 µm) (Figure 1) [10], corresponding to fundamental vibrational and rotational energy transitions [9]. The MIR fundamental transitions have stronger line strengths than their overtones, typically used in the visible and near-IR regions [9,10]. In addition, spectra are less congested, allowing selective spectroscopic detection of many molecules [9,10]. This molecular "fingerprinting" capability makes MIR gas sensors highly desirable for an increasing number of applications involving chemical analysis, such as industrial process control [11][12][13], environmental monitoring [14,15], and medical diagnosis [16].

Optical Gas Sensor Topologies
Most optical gas sensors rely on the Beer-Lambert's law [9], where a gas is detected according to the relation I(λ) = I 0 (λ)e −α(λ)cl : where I(λ) and I 0 (λ) [W/m 2 ] are the detected and emitted optical intensities at the wavelength λ, respectively; α(λ) [L/gm] is the gas absorption coefficient; c [g/L] is the gas concentration; and l [m] is the light-gas interaction pathlength. A typical sensor (depicted in Figure 2), is comprised of: (i) an emitter to generate I 0 (λ), (ii) an optical path, l (gas cell), to guide light to interact with the gas, (iii) an optical filter to select the range of wavelengths (λ) characteristic to the gas target, and (iv) a detector to detect the absorbed light, I(λ). A common technique relies on nondispersive sensing, where unfiltered light is used to interact with the gas [3,6]. Nondispersive gas sensors allow selective detection (with λ), by filtering the detected light based on the characteristic absorption spectra, α(λ), of the molecular species [9]. Sensors configured with IR emitters and detectors, are traditionally known as nondispersive IR (NDIR) sensors [3], although variations for other spectral regions, or for configurations with acoustic instead of optical detectors [3], share the same operating principle based on the Beer-Lambert's law [9].
Various topologies have been implemented to fabricate optical gas sensors ( Figure 3), with the most commonly used based on gas cells formed between face-to-face configured emitters and optical detectors [48,53,54] (Figure 3a-c,e). Strategies to miniaturize the gas cell include: the use of enhancement layers, such as photonic crystals [55], optical cavities [56], multi-pass cells [57], or gas enrichment layers [58], to increase the light-gas interaction ( Figure 3c); planar configurations of emitters and detectors [44,59] (Figure 3f); or use of waveguides for evanescent-field interaction [60][61][62] ( Figure 3g). The absorbed light, ∼ I(λ), is typically detected via an optical detector such as a photodiode [47], thermopile [63], or pyroelectric [64], or an acoustic detector such as a microphone [44]. The sensor response signal, ∼ I(λ), is typically extracted by means of a lock-in detection technique, from a known frequency used to modulate the emitter [19,65]. A reference detector is often used to compensate for changes in the emitted light [40,48,53,54,58,59,64]. Additional sensors can be used to compensate for environmental parameters such as temperature, pressure or humidity [44,51,59,63,66]. In this section, we review the performance of current topologies with a focus on miniaturized devices, based on both acoustic and optical detection.   [48,58] or without [54] filters. (c) The gas cell can be reduced by increasing the light-gas interaction, e.g., by using photonic crystals [55], optical cavities [56], multi-pass cells [57], or gas enrichment layers [58]. (d) Photoacoustic cell. Acoustic waves created by light-gas interaction are detected by a microphone. It can be resonant [67] or non-resonant [68]. (e) Open cell configuration, using either dual optical detection [64] or microphones sealed with target gases [69].
(f) Cell with emitter and detector in planar configuration. Multiple optical detectors with filters in the range of interest can be used [59], or microphones sealed with target gases [44]. (g) Waveguide sensors based on evanescent field interaction [60][61][62] require optical coupling with both emitters and detectors.

Path to Miniaturization and Integration
Optical gas sensors provide excellent stability, selectivity, and sensitivity [3,6], being among the most reliable methods for measuring CO 2 levels in exhale human breath [16,19,20], and therefore are well suited for next generation medical and consumer electronics end-use applications. However, integration technologies that are efficient, are low-cost and can enable low-power consumption, remain the central challenges of applied modern MIR technologies [45,86]. Although significant effort is being dedicated towards the miniaturization of MIR devices [15,47], progress towards chip-scale, low-cost formats, most needed in a variety of applications, is still in its infancy [17,32]. In this section, we review current progress towards the miniaturization and integration of optical gas sensors, and discuss current major challenges.

MIR Emitters
The high-cost and limited tuning range as well as high-power consumption of current MIR sources [45] (the core of an optical gas sensor), make the use of optical gas sensors with low-cost, battery-operated systems an ongoing problem, and even more so with wireless systems [15,30]. For example, despite the success of QCLs in the MIR [37,38], their high-cost (∼$1000) and high-power consumption have limited their application to consumer electronics. MIR LEDs can offer lower power consumption with overall high efficiencies [39,40], however, their operation above ∼5 µm is challenging [45] and comes at significantly increased costs (∼$100). Nevertheless, renewed scientific interest in the miniaturization of low-cost optical gas sensors [43,54,63], is being fueled by advances in silicon micromachining [36,87]. Recently, membrane microhotplates based on MEMS technology [88][89][90] (Figure 4), came up as compact, integrated thermal light sources [42,43,91]. MEMS heaters are proven to be energy efficient [90], allow for rapid modulation owing to their low thermal mass [19,90], and are compatible with standard CMOS foundry processes [19,90]. They are typically used with CMOS compatible thermal detectors (e.g., thermopiles [19,43,54,63,92], bolometers [73], or pyroelectric detectors [59,64,77]), as they allow broadband MIR detection at room temperature [93] with minimum manufacturing costs [36]. However, standard CMOS materials exhibit inherently low MIR emissivity/absorptivity, especially for wavelengths <8 µm, which makes additional post-CMOS/MEMS blackening layers and filter elements necessary [36], often needed to fulfil applications such as spectroscopy.  We have developed various CMOS microhotplates based on tungsten metallization, as well as several thermal engineering techniques to enhance and tailor their MIR proprieties, Figure 4. Tungsten is an interconnect metal found in high temperature CMOS processes, and can enable stable MIR emitters [90] with excellent device reproducibility and the possibility of a wide range of on-chip circuitry, at very low cost [36]. We have engineered highly efficient plasmonic metal structures to enhance the microhotplate MIR emission via excitation of surface plasmon resonances [43], which can be broadly tuned by varying the structure unit cell geometry. CMOS integrated MIR emitters, with drive and temperature control, can feature membrane diameters as small as 600 µm, and have ∼50 mW DC power consumption (∼1 mW optical output power), when operated at 550 • C, with good emission for λ > 8 µm (Figure 4a). We have also showed that the radiation properties of carbon nanotubes (CNTs) can significantly enhance both emissivity [95] and absorptivity [94] of MIR devices, due to their blackbody-like behaviour (nearly unity) (Figure 4b).
We have developed a filter-free technique for the detection of CO 2 based on CMOS plasmonic emitters and detectors [54] (Figure 5), that could be applied for spectroscopic detection across the entire MIR spectrum. The detector signal is computed differentially between the plasmonic and non-plasmonic cells as shown in Figure 5b. Note that except for the plasmonic layer, the two integrated detector cells are identical. The differential signal has a peak around 4.26 µm in the CO 2 detection range and low absorptivity at other wavelengths. More recently, arrays of wavelength-dependent detectors, based on similar plasmonic/metamaterial thermal engineering concepts have been proposed [106][107][108], to enable spectroscopic detection across various bands in the MIR.

Acoustic vs. Optic Detection
Compared to traditional nondispersive gas sensors, PA gas sensors present several advantages. They do not require optical detectors and are wavelength independent. The absorbed light, ∼ I(λ) ∼ c, is measured directly (i.e., not relative to a background), meaning PA is highly accurate, with very little instability [3,79]. Other advantages include smaller (sub-cm [51,69,80]) optical pathlengths, l, and more robust setups [3,44]. Among these, non-resonant PA gas sensors are more stable, feature lower modulating frequencies and require smaller volumes and pathlengths, hence are less susceptible to noise [15,44,50,51,69,80,82,84]. Resonant PA sensors can, however, offer higher sensitivities, but their stability is affected by environmental parameters, such as temperature and pressure [7,66,67,[70][71][72]81]. Despite these benefits, only recently efforts towards non-resonant PA gas sensor miniaturization have been reported, e.g., based on thermal emitters [51,69,80], and LEDs [15,44,50] in combination with microphones, or LEDs and QTFs [109]. Among these, sensors based on highly-sensitive MEMS microphones (employed to detect pressure pulses modulated at audio frequencies) are easy to integrate [110], and offer sensitivities down to ∼tens ppm [15,67,80], with overall small power consumption [50], and form factor [44,50]. PA is unique since it is a direct monitor of a sample nonradiative relaxation channels and, hence, complements absorption spectroscopic techniques [9]. Although PA spectra can be recorded by measuring the sound at different wavelengths of light [79], it requires tunable or multiple MIR sources centred at specific wavelengths, which are not available at low-cost and/or low-power consumption [45]. On the other hand, spectroscopic detection techniques based on arrays of plasmonic detectors [54,[106][107][108] can be achieved at very low cost and minimum power consumption.

Electronics and Signal Processing
Optical gas sensors require signal amplification and processing techniques to increase their signal-to-noise ratio. A common technique relies on lock-in amplification [19,65], where the sensor response signal is recovered from noise by extracting it at a specific reference frequency, typically used to modulate the emitter (e.g., light pulses). Bench-top lock-in amplifiers are widely used in optical gas sensor setups [60,61,66,68,70,71,81,82,84], however, they are not suitable for use with portable sensing devices, and even less so with integrated circuits (ICs). Miniaturized, IC-based lock-in amplifiers can be used to implement optical gas sensors, with relatively good performance [19]. Other techniques include the use of digital signal processors (DSPs) based on microcontrollers (MCUs) [43,75,80] or field programmable gate arrays (FPGAs) [7]. MCU-implemented digital lock-in amplifiers [43,75,80], or fast Fourier transform (FFT)-based techniques [44,69], are increasingly used. An integrated optical gas sensor concept is shown in Figure 6.

Outlook
Optical gas sensors based on MIR absorption spectroscopy offer excellent stability, selectivity, and sensitivity for a growing number of potential applications. Experimental setups that combine low-cost and low-power consumption emitters and detectors, are attractive prospects for battery-operated mobile devices and networks. CMOS-based technologies increase device fabrication flexibility, in addition to having economic advantages. Among various configurations, sensors based on dome-like gas cells [44,47,50,59], allow for planar integration of emitters and detectors, and could be used to further reduce their size. The main challenge is the relatively small pathlengths, l, which can be addressed by further device optimization, e.g., based on multi-pass designs [57,74,111], or use of photonic crystals [55,112] or optical cavities [56] for enhanced absorption. More recently, photoacoustic gas sensors, based on low-cost commercially available MEMS microphones have emerged as simple, compact, and highly reliable devices [15,50].
Most current MIR light sources are expensive and suffer from high-power consumption [45]. MEMS microhotplates have clear advantages in terms of costs, and could potentially (given their full CMOS compatibility) enable sensors with costs below $1. In addition, they have been demonstrated at various wavelengths in the MIR [43,54], with power-consumption possibly below 1 mW. In principle, optical gas sensors based on plasmonic microhotplates could operate across the entire MIR range with relatively high performance [68]. The integration of nanostructures and nanomaterials in MEMS silicon technology could, in principle, produce novel broadband MIR tunable sources. The recent demonstration of a filter-free gas sensor shows the possibility of using this approach for a broad spectral range [54]. These integration technologies could be applied to various gas sensor designs, based on both optical and acoustic detection.