We describe a new class of micro-opto-mechanical chemical sensors: A photonic microharp chemical sensor is an array of closely spaced microbridges, each differing slightly in length and coated with a different sorbent polymer. They are optically interrogated using microcavity interferometry and photothermal actuation, and are coupled directly to an optical fiber. Simultaneous measurements of the fundamental flexural resonant frequency of each microbridge allow the real-time detection and discrimination of a variety of vapor-phase analytes, including DMMP at concentrations as low as 17 ppb.
© 2008 Optical Society of America
Small, portable, reusabale chemical sensors that are both sensitive and selective would represent an important tool for applications ranging from remote sensing to counterterrorism to warfighter safety. One sensing technology that has been extensively studied to satisfy these criteria is arrays of sorbent polymer coatings as the sensing element of an “electronic nose” [1, 2, 3]. Recent research has focused on arrays of ultra-sensitive microcantilevers  whose sorptioninduced bending or resonant frequency change is read-out electronically  or optically [6, 7]. For example, mass detection at the level of 6 femtograms , chemical vapor detection at a level of 30 parts-per-trillion , and detection of single DNA base pairs  have recently been demonstrated. Optical read-out approaches in particular have the potential for extremely high sensitivity  and remote interrogation [12, 13]. However, optical read-out based on the beam deflection method [4, 6] are difficult to miniaturize and/or package due to the physical separation required between the optical detector and the microcantilever sensor.
In this work, we demonstrate a new approach to optically detect analyte adsorption onto an array of micromechanical structures: A single optical spot is focused onto a microharp , which is comprised of an array of closely spaced polymer-coated microbridges. Each microbridge differs in length (and thus resonant frequency) by a few percent and is coated with a different sorbent polymer, as shown in Fig. 1. This microharp forms one reflector of a Fabry-Perot microcavity, such that small motion of a microbridge results in large changes to the reflectivity of a laser tuned in wavelength near a mode of the Fabry-Perot cavity  (see Fig. 1(a)). These optical displacement measurements, when coupled with photothermal actuation , are used to simultaneously monitor the fundamental flexural resonant frequency of each microbridge. A decreases in the mechanical resonant frequency of a microbridge is therefore an indication of analyte adsorption onto that particular microbridge.
This frequency-domain multiplexed interferometric optical read-out of microharp chemical sensors presents a number of advantages over previously reported microcantilever sensor arrays [6, 7, 5], which have been characterized by time-domain multiplexed beam deflection schemes in which individual cantilevers are read-out sequentially in time. These advantages include: (i) Sensitivity - microcavity interferometry has been shown to be limited only by thermal-mechanical (Brownian) noise [17, 18] in its displacement resolution, which can be as low as tens of ; (ii) Scalability - a microharp approach requires a single optical spot for read-out, regardless of the number of individual microbridges; and (iii) Remotability - remote optical interrogation can be carried out at distances in excess of 1 km by placing the sensor in front of an optical retroreflector  or by placing an optical fiber between the user and the sensor. In addition, our sensors require no electronic power, are reusable, and enable discrimination between target analytes and background chemicals.
A sensor coated with a sorbent polymer reversibly absorbs chemical vapors from the air by an amount characterized by the partition coefficient, K, specific to the polymer/analyte combination. In a resonant frequency micromechanical sensor, analyte adsorption is detected by measuring changes to the fundamental mechanical resonant frequency f 0 that arise from massloading. The relationship between the analyte concentration in air, x, and the resonant frequency shift Δf under the assumption of Δf≪f 0 is given by:
where ti is the thickness of the polymer layer or gold microbridge, ρ is the density of the microbridge, NA is Avogadro’s number, M is the molar mass of the analyte molecule, and P is the air pressure. Thus, large partition coefficients, large polymer thicknesses (relative to the microbridge thickness) and the ability to resolve small resonant frequency changes result in sensitive detection of analyte molecules.
2. Fabrication and characterization
The microharps are fabricated on sapphire substrates using a combination of photolithography, electroplating, and surface micromachining. Electroplated gold posts support the array of gold microbridges and surround a thin metallic layer (approximately 20 nm thick) on the top of the substrate that serves as the bottom mirror of the microcavity. Though the top of the microbridges are rough due to the electroplating process, the bottom of the microbridges (which serves as the top mirror for the microcavity) are optically flat. The microbridges are typically suspended about 9 µm above the substrate, and range in length, L, from 180 µm to 220 µm across the microharp shown in Fig. 1. Within a microharp, the microbridge width, w, (typically 20 – 30 µm) and thickness, t, (typically 1 – 2 µm) are not varied. The relatively large gap between the microbridges and the substrate (9 µm) is crucial to minimize the effects of air damping on their motion , thereby increasing the mechanical quality (Q) factor and decreasing the minimum resolvable frequency shift .
The polymers are deposited onto the microbridges using a micropainting technique. Briefly, each polymer is first dissolved in an appropriate solvent. Then, a microneedle is dipped into the solution and touched onto the top of the microbridge, leaving a film with a thickness determined by the polymer concentration and number of touches. For the data presented in this work, three polymers are used: HCSA2 (HC), a hyperbranched carbosilane polymer developed at the Naval Research Laboratory  that is functionalized with hydrogen bond acidic hexafluoroisopropanol groups and targets hydrogen-phosphonate ester or nitroaromatic species such as dimethyl methylphosphonate (DMMP); polyethyleneimine (PEI), a commercial polymer that is strongly hydrophilic; and CS3Ph2 (CS), a linear carbosilane developed at NRL that has phenyl functional groups and is used for the detection of hydrocarbons. The micropainting technique has been shown to have a spatial resolution of about 10 microns, an essential feature for the selective coating of a single microbridge in the microharp without contaminating adjacent microbridges. A coated microharp is shown in Fig 1(b). By coating only one microbridge with each polymer, analyte adsorption into that polymer is mapped to changes in the mechanical resonant frequency of the associated microbridge. It is worth noting that coating nonuniformities do not affect the optical properties, since the coated side of the microbridge is not part of the Fabry-Perot microcavity. For nonuniform coatings, Eq. 1 can be expressed in a more general form as (again assuming Δf≪f 0):
where m eff is the effective dynamic mass for the fundamental flexural mode, w and L are the bridge width and length, respectively, s(x,y,z) is the polymer shape function, and ϕ(x) is the normalized mode shape of the fundamental flexural mode. For an unstrained doubly-clamped microbridge, m eff=ρtw∫L/2 -L/2 ϕ 2(x)dx such that .
Two lasers are used to read-out the resonant frequency of each microbridge in the microharp. The first (the read-out laser), a tunable diode laser with a wavelength between 1440 nm and 1640 nm, is used to measure vertical displacement of a microbridge using microcavity interferometry . By tuning the wavelength of this laser to the side of a Fabry-Perot mode (see the inset of Fig. 3), changes in the microbridge distance from the substrate are mapped into changes in the optical reflectivity. While this laser alone could be used to track the resonance of each microbridge using the resonant displacement noise induced by thermal-mechanical fluctuations , this approach is inherently noisy. Instead, a second laser (the photothermal actuation laser ) at 1310 nm is used to resonantly drive the microbridges, allowing for a higher signal-to-noise determination of f 0. This, in turn, enables the measurement of smaller adsorption induced resonant frequency changes, Δf. This second laser is amplitude modulated by a network analyzer, photothermally driving the microbridges into resonance. The two lasers are combined in a single mode optical fiber (see Fig. 2) and optically excite the microharp. A circulator sends the reflected read-out light from the microharp to an amplified photodetector and the network analyzer. A long-pass filter in front of the photodetector ensures that any signal detected at the second laser‘s modulation frequency is due to microbridge motion, not directly detected light from the actuation laser.
A mechanical vibration spectrum of a microharp measured in this manner is shown in Fig. 3. The abscissa corresponds to the frequency at which the photothermal actuation laser is modulated and the ordinate is the signal at that frequency detected by the network analyzer and photodetector. Four clear peaks indicate the fundamental flexural resonance of each of the four microbridges in the microharp. The spacing of the resonances is a function of the specific length of each individual microbridge. Since each microbridge differs in length, slight bowing of the microbridges implies that the microcavity length, and thus Fabry-Perot mode wavelength, also differs between microbridges, typically by a few nanometers. This is the reason that the wavelength for optimal read-out of microbridge displacement varies by a few nanometers within the microharp, as indicated by the different spectra in Fig. 3.
3. Chemical sensing
The detection of chemical vapors is carried out by continuously monitoring the central resonant frequency of each microbridge within the microharp while it is exposed to various concentrations of analytes. The first microbridge is intentionally left uncoated to serve as a reference to compensate for frequency drift due to temperature or pressure changes. To facilitate comparison between the microbridges, the change in resonant frequency of the ith microbridge (Δf (i)) is normalized by its resonant frequency at t=0 (f (i) 0). Thus, the time-dependent microharp response can then be characterized by three parameters: Δf (2)/f (2) 0-Δf (1)/f (1) 0 (the “HC” response in red in Figs. 4(a)–(c), labeled B2-B1); Δf (3)/f (3) 0-Δf (1)/f (1) 0 (the “PEI” response in green in Figs. 4(a)–(c), labeled B3-B1); and Δf (4)/f (4) 0-Δf (1)/f (1) 0 (the “CS” response in blue in Figs. 4(a)–(c), labeled B4-B1). These three parameters are plotted vs. time in Fig. 4 for time-varying exposure to DMMP (Fig. 4(a)), toluene (Fig. 4(b)), and water vapor (Fig. 4(c)). The data clearly show a unique response for each analyte. Upon exposure to DMMP vapor in the parts-per-billion (ppb) range, only the microbridge coated with HC shows a response (Fig. 4(a). Upon exposure to water vapor in the low parts-per-million (ppm) range, only the microbridge coated with PEI shows a response (Fig. 4(c)). Upon exposure to toluene in the high ppm range, both the HC-coated microbridge and the CS-coated microbridge show a response, due to the comparable partition coefficients of these polymers for toluene adsorption. Nevertheless, the use of multiple coatings enables the clear discrimination between DMMP adsorption and toluene adsorption, even though the toluene concentration is 3000× larger in these exposures.
Analyte concentrations are determined from known saturated vapor pressure and dilution parameters, which are periodically verified by relative humidity and thermal desorber - gas chromatograph - mass spectrometer (TD-GC-MS) measurements. These concentrations can then be used with Eq. 1 to estimate an average coating thickness across each microbridge. Our estimates are in approximate agreement with our measured thicknesses (in the range of hundreds of nanometers), though precise comparison is difficult due to the nonuniformities in the coating thickness.
Upon exposure to an analyte vapor, it takes a finite amount of time for the microbridge to reach a new equilibrium resonant frequency. These response times are limited by the polymer thickness and the diffusion rate of the particular analyte into the polymer. The exposure time of 120 seconds is approximately equal to the response time of toluene into CS and significantly longer than the response time of toluene into HC. However, this exposure time is clearly shorter than the full response time of DMMP into HC and water vapor into PEI. Additional measurements (not shown) reveal that the response time of both PEI/water and HC/DMMP is in the range of 200 s to 400 s, depending on the polymer thickness. The inset of Fig. 5 also shows clear detection of DMMP at a concentration of approximately 17 ppb during a 360 second exposure time, with a one-sigma (Allan variance ) noise-limited detection level of 1.4 ppb. The slow baseline drift in this data results from background resonant frequency drifts in the microbridges not related to analyte sorption.
In order to show the ability of the microharp to distinguish between these analytes, we performed two-dimensional principal component analysis of the response of each analyte over a range of concentrations. For each concentration of analyte, the relative differential frequency shift was found for each polymer (B2-B1, B3-B1, and B4-B1) after a 120 second exposure. This three-dimensional vector was then transformed such that the two principal components of the transformed parameter space are the x-axis and y-axis of Fig. 5. Each analyte is represented by the angle of a vector, with the concentration represented by the vector amplitude. The data show clear distinction between the three analytes tested, even at DMMP concentrations as low as 34 ppb.
The successful operation of our all-optical microharp chemical sensors required advances in the state-of-the-art in a number of areas. First, by moving the microbridges away from the substrate (by distances up to 10 microns), mechanical squeeze film air damping  is minimized, which allows Q-factors in the range of 150 - 1000, depending on the beam geometry. These values are higher than typical resonant frequency micromechanical sensors operated in air at room temperature, and enable the measurement of small resonant frequency shifts. Second, by using micropainting to deposit polymer coatings instead of more traditional ink jetting techniques, microbridges were able to be positioned close enough to each other to allow for simultaneous interrogation by a single optical spot. This technique has important implications for scalability in future systems with more than four microbridges. Mechanical resonant frequency measurements using microbridges coated on the side opposite the Fabry-Perot microcavity make the sensor insensitive to coating nonuniformities. Third, the use of microcavity interferometry and photothermal actuation to read-out the resonant frequencies  within the microharp allows compact fiber-optic interrogation with no electrical power. Last, the deposition of functionalized polymers with some of the highest reported partition coefficients for the organophosphonates (DMMP)  has enabled noise-limited detection levels as low as 1.4 ppb for those compounds.
We have demonstrated detection of the nerve agent simulant DMMP at concentrations as low as 17 ppb that is distinguishable from hydrocarbon and water vapor adsorption, even though these background vapors are present at much higher concentrations. By placing additional coated microbridges in the microharp, further increases in selectivity can be achieved without impacting the read-out complexity. In addition, smaller microbridges characterized by higher resonant frequencies, higher mechanical Q-factors, and greater mass-loading sensitivity, will result in even more sensitive chemical vapor detection. While we recognize that even such an improved future generation photonic microharp sensor will suffer from the same limitations as previously demonstrated polymer sensor arrays regarding analyte identification in mixtures , many important applications exist that can tolerate nonzero false positive rates. Such applications depend on a sensor that is sensitive, selective, compact, networkable, and inexpensive to manufacture. The read-out simplicity inherent in the frequency-domain multiplexing of photonic microharp chemical sensors is an important step towards meeting these criteria.
References and links
1. R. A. McGill, M. H. Abraham, and J. W. Grate, “Choosing polymer-coatings for chemical sensors,” Chemtech 24(9), 27–37 (1994).
2. A. J. Ricco, R. M. Crooks, and G. C. Osbourn, “Surface acoustic wave chemical sensor arrays: New chemically sensitive interfaces combined with novel cluster analysis to detect volatile organic compounds and mixtures,” Acc. Chem. Res. 31, 289–296 (1998). [CrossRef]
3. S. L. Rose-Pehrsson, J.W. Grate, D. S. Ballantine, and P. C. Jurs, “Detection of hazardous vapors including mixtures using pattern-recognition analysis of responses from surface acoustic-wave devices,” Anal. Chem. 60(24), 2801–2811 (1988). [CrossRef] [PubMed]
4. T. Thundat, E. A. Wachter, S. L. Sharp, and R. Warmack, “Detection of mercury-vapor using resonating microcantilevers,” Appl. Phys. Lett. 66, 1695–1697 (1995). [CrossRef]
5. N. Abedinov, C. Popov, Z. Yordanov, T. Ivanov, T. Gotszalk, P. Grabiec, W. Kulisch, I.W. Rangelow, D. Filenko, and Y. Shirshov, “Chemical recognition based on micromachined silicon cantilever array,” J. Vac. Sci. Technol. B. 21(6), 2931–2936 (2003). [CrossRef]
6. L. R. Senesac, P. Dutta, P. G. Datskos, and M. J. Sepaniak, “Analyte species and concentration identification using differentially functionalized microcantilever arrays and artificial neural networks,” Analytica Chimica Acta 558, 94–101 (2006). [CrossRef]
7. H. P. Lang, M. K. Baller, R. Berger, C. Gerber, J. K. Gimzewski, F. M. Battiston, P. Fornaro, J. P. Ramseyer, E. Meyer, and H. J. Güntherodt, “An artificial nose based on a micromechanical cantilever array,” Analytica Chimica Acta 393, 59–65 (1999). [CrossRef]
8. N. V. Lavrik and P. G. Datskos, “Femtogram mass detection using photothermally actuated nanomechanical resonators,” Appl. Phys. Lett. 82, 2697–2699 (2003). [CrossRef]
9. L. A. Pinnaduwage, V. Boiadjiev, J. E. Hawk, and T. Thundat, “Sensitive detection of plastic explosives with self-assembled monolayer-coated microcantilevers,” Appl. Phys. Lett. 83(7), 1471–1473 (2003). [CrossRef]
10. J. Fritz, M. K. Baller, H. P. Lang, H. Rothuizen, P. Vettiger, E. Meyer, H.-J. Güntherodt, C. Gerber, and J. K. Gimzewski, “Translating biomolecular recognition into nanomechanics,” Science 288, 316–318 (2000). [CrossRef] [PubMed]
11. T. H. Stievater, W. S. Rabinovich, H. S. Newman, R. Mahon, D. McGee, and P. G. Goetz, “Measurement of Thermal-Mechanical Noise in Microelectromechanical Systems,” Appl. Phys. Lett. 81, 1779–1781 (2002). [CrossRef]
12. E. A. Wachter, T. Thundat, P. I. Oden, R. J. Warmack, P. G. Datskos, and S. L. Sharp, “Remote optical detection using microcantilevers,” Rev. Sci. Instrum. 67(10), 3434–3439 (1996). [CrossRef]
13. T. H. Stievater, W. S. Rabinovich, R. Mahon, M. S. Ferraro, N. A. Papanicolaou, J. B. Boos, R. Bass, R. A. McGill, and J. Stepnowski, “Remote All-Optical Detection of Chemical Vapors using Micromechanical Resonators,” in Nanoelectronic Devices for Defense and Security (NANO-DDS) Conference (DOD-DTRA, Arlington, VA, USA, 2007).
14. D. W. Carr and H. G. Craighead, “Fabrication of nanoelectromechanical systems in single crystal silicon using silicon on insulator substrates and electron beam lithography,” vol. 15, pp. 2760–2763 (AVS, 1997).
15. T. H. Stievater, W. S. Rabinovich, H. S. Newman, J. L. Ebel, R. Mahon, D. J. McGee, and P. G. Goetz, “Microcavity Interferometry for MEMS Device Characterization,” J. Microelectromech. Syst. 12, 109–116 (2003). [CrossRef]
16. T. H. Stievater, W. S. Rabinovich, M. S. Ferraro, N. A. Papanicolaou, J. B. Boos, R. A. McGill, and J. L. Stepnowski, “All-Optical Micromechanical Chemical Sensors,” Appl. Phys. Lett. 89, 091,125 (2006). [CrossRef]
17. T. B. Gabrielson, “Mechanical-thermal noise in micromachined acoustic and vibration sensors,” IEEE Trans. Electron. Devices 40, 903–909 (1993). [CrossRef]
18. T. H. Stievater, W. S. Rabinovich, N. A. Papanicolaou, R. Bass, and J. B. Boos, “Measured limits of detection based on thermal-mechanical frequency noise in micromechanical sensors,” Appl. Phys. Lett. 90(5), 051114 (pages 3) (2007). URL http://link.aip.org/link/?APL/90/051114/1. [CrossRef]
19. G. C. Gilbreath, W. S. Rabinovich, T. J. Meehan, M. J. Vilcheck, M. Stell, R. Mahon, P. G. Goetz, E. Oh, J. A. Vasquez, K. Cochrell, R. L. Lucke, and S. Mozersky, “Progress in development of multiple-quantum-well retromodulators for free-space data links,” Opt. Eng. 42, 1611–1617 (2003). [CrossRef]
20. T. Veijola, “Compact models for squeezed-film dampers with inertial and rarefied gas effects,” J. Micromech. Microeng. 14, 1109–1118 (2004). [CrossRef]
21. “Linear and hyperbranched hydrogen bond acidic poly(silylene-methylene)s for chemical sensor applications,” in ABSTRACTS OF PAPERS OF THE AMERICAN CHEMICAL SOCIETY, vol. 225 of 316-PMSE Part 2 (2003).