A Cantilever Array-based Artificial Nose

*Corresponding author. Tel.: #41-1-724-8929; fax: #41-1-724-8964.

E-mail address: [email protected] (H.P. Lang)

Ultramicroscopy 82 (2000) 1}9

A cantilever array-based arti"cial nose

M.K. Baller!!", H.P. Lang!!"!*, J. Fritz!!", Ch. Gerber!, J.K. Gimzewski!,U. Drechsler#, H. Rothuizen#, M. Despont#, P. Vettiger#, F.M. Battiston$,

J.P. Ramseyer$, P. Fornaro$, E. Meyer$, H.-J. GuK ntherodt$!IBM Research, Zurich Research Laboratory, Nanoscale Science Dept., Sa( umerstrasse 4, CH-8803 Ru( schlikon, Switzerland

"Institute of Physics, University of Basel, Klingelbergstrasse 82, CH-4056 Basel, Switzerland#IBM Research, Zurich Research Laboratory, Micromechanics Dept., Sa( umerstrasse 4, CH-8803 Ru( schlikon, Switzerland

$Institute of Physics, University of Basel, Klingelbergstrasse 82, CH-4056 Basel, Switzerland

Received 31 May 1999; received in revised form 17 August 1999

Abstract

We present quantitative and qualitative detection of analyte vapors using a microfabricated silicon cantilever array. Toobserve transduction of physical and chemical processes into nanomechanical motion of the cantilever, swelling ofa polymer layer on the cantilever is monitored during exposure to the analyte. This motion is tracked by a beam-de#ection technique using a time multiplexing scheme. The response pattern of eight cantilevers is analyzed via principalcomponent analysis (PCA) and arti"cial neural network (ANN) techniques, which facilitates the application of the deviceas an arti"cial chemical nose. Analytes tested comprise chemical solvents, a homologous series of primary alcohols, andnatural #avors. First di!erential measurements of surface stress change due to protein adsorption on a cantilever arrayare shown using a liquid cell. ! 2000 Elsevier Science B.V. All rights reserved.

PACS: 07.07.Df; 07.79.!v; 61.16.Ch; 07.10.Cm

Keywords: Arti"cial nose; Micromechanics; Chemical sensors; Protein adsorption

1. Introduction

Recently, increasing e!orts have been put intothe development of cantilever-based sensors for thedetection of physical phenomena and chemical re-actions. Miniaturized sensors show fast responses,high sensitivity, and are suitable for mass produc-

tion. Their main application "elds are quality andprocess control, disposable diagnostic biosensingfor medical analysis, fragrance design, oenology,and as sensing devices for gaseous analytes, e.g.process gases or solvent vapors.

The application of single-cantilever sensors todetermine quantities below the detection limits ofequivalent `classicala methods has been demon-strated in the literature: Gimzewski et al. haveshown that catalytic processes can be observedwith picojoule sensitivity in calorimetry [1]. Thecantilever method allows the study of solid}solid

0304-3991/00/$ - see front matter ! 2000 Elsevier Science B.V. All rights reserved.PII: S 0 3 0 4 - 3 9 9 1 ( 9 9 ) 0 0 1 2 3 - 0

Fig. 1. Scanning electron microscopy image of a microfabricated cantilever sensor array prior to deposition of sensor coatings. Thearray was produced from silicon by combined dry and wet etching techniques in the Micromechanics department at the IBM ZurichResearch Laboratory. Cantilever length: 500 !m, thickness: 1 !m, width: 100 !m. Typical spring constant: 0.02 N m"#.

phase transitions of alkanes [2,3] using minutequantities of material in the picogram range.Formation enthalpies of small Sn metallic clustershave been determined in the gas phase usingmicrocantilevers [4]. Cantilever sensors have beenapplied for photothermal spectroscopy [5], surfacestress detection [6}8], infrared radiationdetection [9], mass change detection [10,11],photothermal sensors [12}14] and thermog-ravimetry [15]. Cantilever sensors can be operatedin liquids to observe such processes as proteinadsorption [16].

The sensor presented is based on a microfab-ricated array of silicon cantilevers. Each of thecantilevers is functionalized by coating it witha speci"c sensor layer to transduce a physicalprocess or a chemical reaction into a nanomechani-cal response. Such arrays allow use of somecantilevers as reference sensors (di!erentialmeasurements) [17}19]. Here, we mainly takeadvantage of the swelling e!ect of polymers uponexposure to gaseous analytes. The kinetics of theswelling process is related to the vapor pressureand the solubility characteristics of the analyte inthe polymers. In addition, e!ects related to heattransfer or mass change can be evaluated foranalyte detection.

2. Experimental

2.1. Sample chamber setup

Sensor parallelism is a key prerequisite for ap-plications requiring a high degree of complexityand selectivity [20}24]. Our sensor array chip(Fig. 1) comprises eight linearly arranged canti-lever-type sensors (length: 500 !m, width: 100 !m,thickness: 1 !m, pitch: 250 !m), which can be sensi-tized individually for various detection purposes.This array is housed in a sample chamber made ofaluminum (volume: 6 cm$) with analyte inlet andoutlet as well as windows for cantilever de#ectionreadout (Fig. 2). Cantilever de#ection is measuredvia incident light beams from eight individual lightsources having the same pitch as the cantilevers(vertical cavity surface emitting lasers (VCSELs)optical power: up to 2 mW, wavelength: 760 nm,single mode emitters, CSEM ZuK rich, Switzerland).The identical pitch of light sources and cantileversfacilitates the optical readout of sensors via beamde#ection. Optics consist of a doublet of achro-matic lenses providing a 1 : 1 projection. The lightis re#ected o! the cantilever apex and collectedby a linear position-sensitive detector (PSD). Tosimplify device construction, a time-multiplexing

2 M.K. Baller et al. / Ultramicroscopy 82 (2000) 1}9

Fig. 2. Schematic setup of sample chamber with analyte inletand outlet, cantilever de#ection readout electronics and control.The acquisition and control unit sequentially switches the lightsources on and o! via a time-multiplexing scheme (40 ms foreach cantilever). The laser light (!"760 nm) is directed viafocusing optics onto the apex of the cantilevers of the array, thenre#ected and collected by a linear position sensitive detector(PSD). The photocurrents from opposing electrodes are con-verted into voltages. The voltage di!erence is adjusted by ano!set to allow optimal ampli"cation and fed into the dataacquisition and control unit. Data processing is performed o!-line on a personal computer.

Fig. 3. Schematic view of a cantilever array with polymer coat-ings used for the eight cantilevers of the sensor array.PVP"polyvinylpyridine, PU"polyurethane, PS"polysty-rene, PMMA"polymethylmethacrylate. Schematic cross sec-tion of a coated cantilever.

scheme [17,18] is applied by switching on and o!individual light sources to illuminate only one can-tilever apex at a time and to provide an electricalsignal at the PSD for each cantilever de#ectionthree times per second. The y position of the lightspots on the PSD depends directly on the di!erenceof photocurrents from opposing electrodes. Thephotocurrents are converted into voltages usingtransimpedance ampli"ers, and the voltage di!er-ence is transmitted to a data acquisition and con-trol unit. This so-called nanotechnology olfactorysensors (NOSE) system [25] consists of modularunits, which can be expanded by further sensorarrays.

2.2. Cantilever sensor coatings

Cantilever sensor arrays were coated with 30 nmof gold by electron beam evaporation. Polymerswere spray-coated onto one side of the cantileversto form a homogeneous layer of +5 !m thickness(Fig. 3). Commercially available polymers were dis-

solved in toluene or ethanol (concentrations for allsolutions: 5 mg/ml).

2.3. Data acquisition and analysis

An automated gas handling system [25] is usedfor programmed exposure of the sensor array tovarious gaseous analytes. Gas #ows are adjusted inthe range of 2}100 ml/min by software-driven com-mercial digital mass #ow controllers (Bronkhorst,The Netherlands). By means of two mass-#owcontrollers it is possible to mix carrier gas (drynitrogen, purity 99.999990%) with well-de"nedamounts of analyte. Two-hundred microliter ofeach analyte was "lled into 4-ml vials closed bya septum. Vapor from the head space was injectedinto the chamber via the gas handling system(Fig. 2).

The sensitized cantilever array is exposed to theanalyte for 10 s. Then, the cantilever response ismeasured for 2 min while being purged with nitro-gen. To reduce the amount of data the normalizedsignal amplitude is evaluated for each cantilever at

M.K. Baller et al. / Ultramicroscopy 82 (2000) 1}9 3

Fig. 4. (a) Reaction traces from eight polymer-coated cantileversupon injection of ethanol for 10 s. Signal magnitudes at "vepoints in time t

#}t

%are extracted. These data points su$ciently

characterize the analyte desorption process. (b) Normalizingthese "ve response magnitudes from all eight cantilevers (S1}S8)yields a "ngerprint pattern of the analyte suitable for furtheranalysis via principal component analysis and neural networktechniques.

"ve equidistant points in time (Fig. 4). This "nger-print information ("ve points for eight canti-levers"40 parameters) re#ects the analytedesorption characteristics, which were found to re-sult in a better separation of the analytes comparedto the adsorption characteristics and the maximumamplitudes of the sensor response. The "ngerprintis analyzed by a commercial principal componentanalysis (PCA) software package (MVSP, multi-variate statistical package, version 3.0, KovachComputing Services, UK). PCA extracts most-dominant deviations in responses for variousanalytes. The largest di!erences in signal ampli-tudes of the "ngerprint patterns are plotted ina two-dimensional graph. The axes refer to projec-

tions of the multidimensional datasets into twodimensions (principal components). This procedureis aimed at maximum distinction performancebetween analytes, i.e. several measurementsof the same analyte should yield a cluster inprincipal component space, whereas measurementsof di!erent analytes should yield well-separatedclusters.

For more complex measurements, e.g. to analyzemulticomponent mixtures of gaseous analytes suchas natural #avors, a di!erent strategy involvingarti"cial neural networks (ANN) is pursued.Whereas PCA extracts most-dominant di!erencesin the "ngerprint pattern, neural network analysisconsiders all components of the "ngerprint. Usingseveral training sets for each analyte, the ANNsoftware package (e.g. Neuroshell Classi"er, Ver-sion 2.0, Ward Systems Group Inc., USA) can`learna the corresponding "ngerprint pattern toidentify the analytes.

2.4. Stress measurements in liquids

To measure stress-related cantilever bending inliquid environment we use a setup based on a modi-"ed commercial atomic force microscope head andliquid cell (Digital Instruments, Santa Barbara, CA,USA), which can be modi"ed for current #ow ex-periments. The volume of the liquid cell (+800 !l)can be injected or exchanged with a micropipette.For simplicity we use a cantilever array of twogold-coated Si cantilevers (cantilever length:500 !m, width: 100 !m, thickness: 1 !m, pitch:500 !m, thickness of evaporated gold layer: 30 nm).This system can easily be upgraded to operationwith multiple cantilever arrays. Utilizing a similarreadout technique based on VCSEL light sourcesand PSD as described above, cantilever bendingcan be observed individually for each cantilever.

The gold surface of one cantilever was blockedfor protein adsorption by 11-(pentaethyleneglycol)undecan-1-thiol (PEG5 thiol, courtesy ofFadhil Kamounah, University of Copenhagen,Denmark). The other was functionalized foradsorption by hydrophobic hexadecanthiol (HDT,Fluka Chemicals, Buchs, Switzerland, furtherpuri"ed by column chromatography). Adsorption


Fig. 5. Principal component analysis (PCA) of primary alcoholsin a two-dimensional representation in principal componentspace. Analytes form clearly separated clusters ordered accord-ing to their molecular weight.

Fig. 6. PCA of methanol, ethanol, and a 1 : 1 mixture of bothalcohols showing distinct clusters.

of bovine serum albumine (BSA, 1 mg/ml,Sigma}Aldrich Chemicals, Buchs, Switzerland,in phosphate-bu!ered saline, PBS, pH 7.4, FlukaChemicals, Buchs, Switzerland) on HDT and theblock of BSA adsorption by PEG5 were checked byellipsometry (data not shown).

3. Results and discussion

3.1. Homologous series of primary alcohols

To demonstrate the analyte recognition capabil-ity of our NOSE device, a homologous series ofprimary alcohols from methanol to heptanol(Fluka Chemicals, puriss., p.a. grade, Buchs, Swit-zerland) was investigated. Octanol and higher alco-hols are not considered due to the relatively lowvapor pressures and thus long desorption times. Asan example the cantilever responses for ethanol aredisplayed in Fig. 4a.

All analytes were measured "ve times in randomorder to exclude systematic in#uences. The resultswere found to be reproducible. This set of"ngerprints was analyzed using PCA based onthe procedure described in Fig. 4. The resultis shown in Fig. 5. The "ve measurements of eachanalyte show distinct clusters in principal compon-ent space and allow unambiguous identi"cation ofanalytes.

3.2. Mixture of alcohols

We extended the previous investigations to studyhow a mixture of alcohols is characterized by PCAmethods. For this purpose a mixture of 100 !l ofmethanol and 100 !l of ethanol was put into a vialfor measurement. The principal components evalu-ated based on seven "ngerprint data sets showedclear clustering (Fig. 6). Special care had to be takenconcerning the amount of vapor extracted in thehead space of the vial in order not to in#uence thecomposition of the vapor mixture due to the di!er-ence in vapor pressures of methanol and ethanol.Using PCA it is possible to distinguish amongdi!erent mixtures of analytes provided that thesemixtures have been previously characterized byPCA. It is di$cult to determine the mixing ratio ofindividual analytes directly from the cluster posi-tions of mixture constituents in the PCA plot, be-cause analyte desorption kinetics of analytemixtures do not depend on the mixing ratio ina clearly predictable way.

3.3. Solvents

Polar and unpolar solvents such as water,ethanol, methanol, 1-propanol, 2-propanol,acetone, dichloro-methane, toluene, and heptaneare widely used in chemical and technological pro-duction processes. To demonstrate the separationselectivity of our NOSE device we investigated


Fig. 7. PCA of various commonly used solvents.

Table 1Neural network analysis of natural #avors (Note: BA"bitter almond, CH"cherry, OR"orange, RU"arti"cial rum, VA"vanilla,LE"lemon. `Actuala signi"es the actual species of analyte, `classi"ed asameans the assignment by the neural network software basedon a discrimination probability of at least 0.5. Numbers indicate identi"cation probabilities. All analytes have been identi"ed correctly(bold numbers). For most analytes the identi"cation probability is close to 1, except for RU, which was identi"ed with a probability of0.728 only. For each analyte, seven training sets and one test set were used. The neural network was composed of "ve input and 15hidden neurons.)

Test Actual Classi"ed as BA CH OR RU VA LE

1 BA BA 0.993 0.001 0.000 0.000 0.000 0.0052 CH CH 0.001 0.923 0.069 0.005 0.002 10.0003 OR OR 0.000 0.026 0.974 0.000 0.000 0.0004 RU RU 0.000 0.272 0.001 0.662 0.054 0.0115 VA VA 0.000 0.000 0.000 0.000 0.994 0.0066 LE LE 0.000 0.000 0.000 0.020 0.000 0.980

vapor samples of the above-mentioned solventsusing 100 !l of analyte in vials as described above.Fig. 7 clearly shows clustering for the analytestested, i.e. successful identi"cation and selectivity.Further applications may be settled in the "eld ofquality and purity control of solvents in productionlines.

3.4. Natural yavors

Natural #avors designed to give bakery productsa special taste are complex mixtures of variouscomponents, mainly water and 1,2-propanediol, insome cases ethanol, besides characteristic #avoringredients. Typical examples of such bakery

#avors are bitter almond (BA), cherry (CH), orange(OR), arti"cial rum (RU), vanilla (VA) and lemon(LE). Such analytes are di$cult to separate byPCA, because similar ratios of the major constitu-ents of these mixtures are present in all analytes, inparticular water and 1,2-propanediol. For this rea-son ANN techniques were applied, taking into ac-count all parameters from the "ngerprint data sets.For each analyte, seven independent measurementswere "rst recorded and fed as training sets into anANN. Then, after this `learninga process, the suc-cess of recognition of another independent data setwas determined. As can be seen in Table 1 allanalytes were identi"ed, "ve of which with a prob-ability close to 1. Only the arti"cial rum sample hasa lower identi"cation probability of 0.728, but itsidenti"cation probability could be increased signi"-cantly if more training sets for analytes were used.

3.5. Quantitative measurements

Fig. 8 shows quantitative measurements of 1-propanol concentrations ranging from 500 to1000 ppm in steps of 100 ppm. This data was pro-duced by adjusting the gas #ow through the analytevial in discrete steps using the gas handling system.The sensitivity is +30 ppm !m"# de#ection.

At the moment, the sensitivity-limiting factorsin our system are the #ow rates imposed by thegas handling system and the chamber volume,but not the cantilever de#ection measurement


Fig. 8. Quantitative measurement of various concentrations of1-propanol. For clarity, only four of the eight traces of cantilevermotions are reproduced. Analyte injection time: 10 s, purgingtime (N

&): 6 min.

and the sensitivity of the sensor layer. A reductionof the chamber volume will increase the mass sensi-tivity and reduce response times. Measurementstatistics can be enhanced using an autosamplerinstead of the gas handling system. Taking theseimprovements into account, estimated detectionlevels are expected to be well below 1 ppm.

3.6. Protein adsorption on a cantilever

An advantage of Si cantilever sensors is that theycan be operated in di!erent environments such asair, vacuum, or liquids. Especially operation inliquids introduces the possibility of using function-alized cantilevers as biosensors. The "rst applica-tion of Si cantilevers as resonant mass sensors wasa biomass-monitoring device for living cells [26].Further experiments are reported, where cells havebeen grown on cantilever surfaces and static de#ec-tions were measured due to cellular activity [27].Other applications in liquids include surface stressmeasurements of electrochemical processes [28,29]and of adsorption of proteins on the cantileversurface [16,30]. Another type of experiments withbiomolecules attempts to use Si cantilevers tomonitor molecular recognition, for example of anti-body}antigen interaction or DNA hybridization.The use of cantilevers allows molecular recognitionto be transduced directly into a nanomechanicalresponse.

The major disadvantage of surface stressmeasurements in liquids reported up to now is thatexperiments were performed without a referencecantilever. Using two or more cantilevers of a singlearray in parallel allows one of the cantilevers to beoperated as a reference sensor. This is especiallyimportant for measuring the static de#ection ofcantilevers in liquids, because small changes of pH,ion concentration, refractive index, temperature, oreven the quality and treatment of the cantileversurface can in#uence dramatically the de#ectionsignal. For example, a change of the refractive in-dex can mimic a de#ection signal several tens ofnanometers in magnitude. These artifacts can becancelled out by measuring the di!erential signalbetween two sensors, an inert reference sensor anda speci"cally sensitized sensor.

Such a di!erential measurement is shown inFig. 9. The liquid cell with two sensors in parallelwas #ushed with bu!er, then with BSA, and againwith bu!er while recording the de#ection signal ofthe two cantilevers. The raw data (Fig. 9a) showsturbulences due to injection of the liquid, thermalequilibration after injection for 10 min or more,and wave-like changes of the de#ection signal dueto a change of room temperature. In contrast, thedi!erential signal (Fig. 9b) compensates for suchthermal e!ects and most of the turbulences. With-out thermal stabilization, it shows only a slight driftof about 5 nm h"#. The gold surface of one sensorwas covered by an alkylthiol and the other sensorby a PEG5 thiol. BSA adsorbs only on the hydro-phobic surface of the alkylthiol sensors. PEG isknown to block almost completely the adsorptionof proteins such as BSA or antibodies on surfaces[31]. As BSA adsorbs to the same extent on the Sisurfaces of both cantilevers, this adsorption processdoes not contribute to the di!erential signal.

The measured de#ection is of the order of 25 nmand corresponds to a bending of the alkylthiolcantilever away from its gold surface. That is, theprotein layer causes a compressive stress of about0.006 N m"#, as can be calculated with Stoney'sformula [8]. Thereby the stress of the unorderedphysisorbed protein layer on top of the HDT ismore than one order of magnitude smaller than thetypical stress of about 0.2 N m"# related to theself-assembly of thiols on gold. The adsorption of


Fig. 9. (a) De#ection traces of HDT (black) and PEG 5-func-tionalized (grey) cantilevers on injection of BSA (raw data). Thewavy behavior of the curves is due to thermal in#uences. (b)Di!erence in responses of HDT and PEG 5-covered cantilevers.Adsorption of BSA on the HDT surface is clearly observed(increase in de#ection). Injection of bu!er solution does notchange the de#ection any further. Peaks are due to the injectionprocess (indicated by arrows).

the proteins on the hydrophobic surface occurs withina few minutes. Up to now, all adsorption measure-ments of various thiols and proteins on functionalizedand non-functionalized surfaces show a compressivestress, that is, the cantilevers bend away from thesurface on which the molecules adsorb [8].

4. Conclusion

In conclusion, we have demonstrated that a mi-cromechanical array of cantilevers can be used as

a selective and quantitative chemical sensor (arti"-cial nose). It can be employed to create "ngerprintsof analytes. Its major advantages are the capabilityto use reference sensors for di!erential measure-ments that compensate for superimposed distur-bances. Likewise, calibration o!sets originatingfrom misalignment or refractive-index changes byoperation in liquids or gases can be corrected. Asexamples, quantitative and qualitative analyte de-tection and identi"cation were demonstrated bycovering the cantilevers with a system of variouspolymers and polymer blends. The cantilever de-#ection caused by a swelling of the polymer layersupon exposure to gaseous analytes was used tocreate "ngerprints for PCA and ANN recognitiontechniques. Successful identi"cation of a series ofprimary alcohols, commonly used solvents, and#avors for bakery products has been demonstrated.

The device is designed to operate in variousmedia such as ambient air, vacuum, gases, andliquids. First results of sensor array operation inliquids have been shown, such as protein adsorp-tion on cantilevers in #uids. The micromechanicaldesign of the sensors implies short response timesand high sensitivity over a wide range of operatingtemperatures, and they can be integrated seamless-ly into microelectronic devices.

Acknowledgements

We thank P.F. Seidler and P. GueH ret for helpfuldiscussions as well as their support. We acknow-ledge support from E. Delamarche and A. Bernard,IBM Zurich Research Laboratory. This work ispartially funded by the Swiss National ScienceFoundation priority program MINAST (Micro-and Nanostructure System Technology), Project7.04 NOSE.

References

[1] J.K. Gimzewski, Ch. Gerber, E. Meyer, R.R. Schlittler,Chem. Phys. Lett. 217 (1994) 589.

[2] R. Berger, Ch. Gerber, J.K. Gimzewski, E. Meyer, H.-J.GuK ntherodt, Appl. Phys. Lett. 69 (1996) 40.

[3] Y. Nakagawa, R. SchaK fer, Angew. Chem. Int. Ed. 38 (1999)1083.


[4] T. Bachels, R. SchaK fer, Chem. Phys. Lett. 300 (1999) 177.[5] J.R. Barnes, R.J. Stephenson, M.E. Welland, Ch. Gerber,

J.K. Gimzewski, Nature 372 (1994) 79.[6] G.Y. Chen, T. Thundat, E.A. Wachter, R.J. Warmack, J.

Appl. Phys. 77 (1995) 3618.[7] R. Berger, Ch. Gerber, H.P. Lang, J.K. Gimzewski, Micro-

electron. Eng. 35 (1997) 373.[8] R. Berger, E. Delamarche, H.P. Lang, Ch. Gerber, J.K.

Gimzewski, E. Meyer, H.-J. GuK ntherodt, Science 276(1997) 2021.

[9] T. Thundat, S.L. Sharp, W.G. Fisher, R.J. Warmack, E.A.Wachter, Appl. Phys. Lett. 66 (1995) 1563.

[10] T. Thundat, R.J. Warmack, G.Y. Chen, D.P. Allison, Appl.Phys. Lett. 64 (1994) 2894.

[11] T. Thundat, E.A. Wachter, S.L. Sharp, R.J. Warmack,Appl. Phys. Lett. 66 (1995) 1695.

[12] J. Lai, Z. Shi, T. Perazzo, A. Majumdar, Sensors Actuators58 (1997) 113.

[13] J. Varesi, J. Lai, Z. Shi, T. Perazzo, A. Majumdar, Appl.Phys. Lett. 71 (1997) 306.

[14] T. Perazzo, M. Mao, O. Kwon, A. Majumdar, J.B. Varesi,P. Norton, Appl. Phys. Lett. 74 (1999) 3567.

[15] R. Berger, H.P. Lang, Ch. Gerber, J.K. Gimzewski, J.H.Fabian, L. Scandella, E. Meyer, H.-J. GuK ntherodt, Chem.Phys. Lett. 294 (1998) 363.

[16] H.-J. Butt, J. Colloid Interface Sci. 180 (1996) 251.[17] H.P. Lang, R. Berger, C. Andreoli, J. Brugger, M. Despont,

P. Vettiger, Ch. Gerber, J.K. Gimzewski, J.-P. Ramseyer,E. Meyer, H.-J. GuK ntherodt, Appl. Phys. Lett. 72 (1998)383.

[18] H.P. Lang, R. Berger, C. Andreoli, J. Brugger, M. Despont,P. Vettiger, F. Battiston, J.-P. Ramseyer, E. Meyer, T.Mezzacasa, L. Scandella, H.-J. GuK ntherodt, Ch. Gerber,J.K. Gimzewski, Appl. Phys. A 66 (1998) S61.

[19] H.P. Lang, M.K. Baller, R. Berger, Ch. Gerber, J.K. Gim-zewski, F.M. Battiston, P. Fornaro, J.P. Ramseyer, E.Meyer, H.-J. GuK ntherodt, Anal. Chim. Acta 393 (1999) 59.

[20] S.C. Minne, S.R. Manalis, C.F. Quate, Appl. Phys. Lett. 67(1995) 3918.

[21] M.I. Lutwyche, Y. Wada, Sensors Actuators A 48 (1995)127.

[22] S.A. Miller, K.L. Turner, N.C. McDonald, Proceedings ofthe Transducers '97, International Conference on Solid-State Sensors and Actuators, Chicago, June 16}19, 1997,IEEE, Piscataway, 1997, p. 455.

[23] W. GoK pel, J. Hesse, J.N. Zemel (Eds.), Sensors: A Compre-hensive Survey, Vols. 1}9, VCH, Weinheim, 1989}1996.

[24] H. Baltes, W. GoK pel, J. Hesse (Eds.), Sensors Update, Vols.1}5, VCH, Weinheim, 1996}1999.

[25] H.P. Lang, M. K. Baller, F.M. Battiston, J. Fritz, R.Berger, J.-P. Ramseyer, P. Fornaro, E. Meyer, H.-J.GuK ntherodt, J. Brugger, U. Drechsler, H. Rothuizen,M. Despont, P. Vettiger, Ch. Gerber, J.K. Gimzewski,Technical Digest 12th IEEE International Conference onMicro Electro Mechanical Systems `MEMS '99a,Orlando, FL, USA, January 17}21, IEEE, Piscataway,1999, p. 9.

[26] S. Prescesky, M. Parameswaran, A. Rawicz, R.F.B. Turner,U. Reichl, Can. J. Phys. 70 (1992) 1178.

[27] M.D. Antonik, N.P. D'Costa, J.H. Hoh, IEEE Eng. inMed. and Biol. (1997) 66.

[28] S.J. O'Shea, M.E. Welland, T.A. Brunt, A.R. Ramadan,T. Rayment, J. Vac. Sci. Technol. B 14 (1996) 1383.

[29] R. Raiteri, H.-J. Butt, J. Phys. Chem. 99 (1995) 15728.[30] T. Thundat, P.I. Oden, R.J. Warmack, Microscale Ther-

mophysical Engineering 1 (1997) 185.[31] M. Zhang, T. Desai, M. Ferrari, Biomaterials 19 (1998)

953.


Documents

A Cantilever Array-based Artificial Nose