- Email: [email protected]
A light detection cell to be used in a micro analysis system for ammonia
Talanta 56 (2002) 331– 339 www.elsevier.com/locate/talanta
A light detection cell to be used in a micro analysis system for ammonia R.M. Tiggelaar *, T.T. Veenstra, R.G.P. Sanders, J.G.E. Gardeniers, M.C. Elwenspoek, A. van den Berg MESA + Research Institute, Uni6ersity of Twente (EL-TT), P.O. Box 217, 7500 AE, Enschede, The Netherlands Received 11 April 2001; received in revised form 20 August 2001; accepted 23 August 2001
Abstract This paper describes the design, realization and characterization of a micromachined light detection cell. This light detection cell is designed to meet the specifications needed for a micro total analysis system in which ammonia is converted to indophenol blue. The concentration of indophenol blue is measured in a light detection cell. The light detection cell was created using KOH/IPA etching of silicon. The KOH/IPA etchant was a 31 wt.% potassium hydroxide (KOH) solution with 250 ml isopropyl alcohol (IPA) per 1000 ml H2O added to it. The temperature of the solution was 50 °C. Etching with KOH/IPA results in 45° sidewalls ({110} planes) which can be used for the in- and outcoupling of the light. The internal volume of the realized light detection cell is smaller than 1 ml, enabling measurements on samples in the order of only 1 ml. Measurements were performed on indophenol blue samples in the range of 0.02 to 50 mM. In this range the measurements showed good reproducibility. © 2002 Elsevier Science B.V. All rights reserved. Keywords: KOH/IPA etching; Light detection cell; Berthelot; Ammonia detection
1. Introduction The light detection cell presented in this paper is part of a micro analysis system for ammonia. This ammonia detection system or MAFIAS (Micro Ammonia Fluid Injection Analysis System) is a micro total analysis system (m-TAS) which can be used for the determination of the concentration of ammonia in water [1]. In MAFIAS concentra* Corresponding author. Tel.: +31-53-489-4420; fax: +3153-489-3343. E-mail address: [email protected] (R.M. Tiggelaar).
tion measurements are carried out using an optical detection method, since this method is used most commonly in micro total analysis systems. Furthermore, the internal volumes of micro total analysis systems in general are small (order of 10 ml) and therefore, the size of the sample used in the light detection cell has to be in the microliterrange. The MAFIAS is made of silicon and Pyrex and the light detection cell will be made of the same materials to ensure that monolithic integration of the cell with the analysis system will be possible. The latter aspect is important for reduction of sample volume and sample dispersion.
0039-9140/02/$ - see front matter © 2002 Elsevier Science B.V. All rights reserved. PII: S 0 0 3 9 - 9 1 4 0 ( 0 1 ) 0 0 5 9 9 - 9
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
332
The analytical chemical method applied in MAFIAS is the Berthelot reaction mechanism [1 –5]. Using this reaction mechanism the ammonia content of a sample reacts with hypochlorite and phenol to form indophenol blue. The complete conversion takes several minutes and the reaction temperature should be lower than 38 °C to avoid bubbles in the solution. After conversion, the concentration of indophenol blue is determined using the light detection cell. The concentration of this substance is directly related to the initial ammonia concentration. For the development of a practical analysis system for continuous measurement of ammonia samples it was specified that the ammonia concentration should be detectable in the range 6– 600 mM [1]. After conversion by the Berthelot reaction this results in concentrations of indophenol blue in the range 0.15–15 mM. The concentrations of the used starting reagents in the Berthelot reaction are 5.8 mM hypochlorite and 5.8 mM phenol. Though hypochlorite is a strong oxidant and therefore, might be expected to etch silicon, the hypochlorite concentrations in the system will be as low as 75 mg l − 1, which is considered to be of minor significance with respect to the etching of silicon. Furthermore, the system is flushed thoroughly after each sample.
2. Design considerations
2.1. Effects of reflections, light in- and outcoupling The intensity of the light exiting from a light cell is influenced by two effects. The first effect is
the absorption of light by the solution (Lambert– Beer law). The second effect which influences the intensity of the outcoming light is reflection of light at surfaces between different media. If a beam of light hits a surface, part of the light is reflected from that surface and another part is transmitted into the medium behind the surface. This is quantified by the reflectance R, which is the ratio of the reflected power to the incident power and strongly depends on the incident angle qi as well as on the refractive indices of the incident and reflecting media. Using Snell’s law of reflection and Fresnel’s equations, the reflectance R as a function of the incident angle qi can be calculated for interfaces between different media. At constant angle of incidence (qi), the intensity of a beam of light travelling through a channel with a non-absorbing solution is only influenced by the number of reflections. In this special case the decrease in light intensity can be expressed as a function of the reflectance R Eq. (1): I= R xI0
(1)
in which I is the measured light intensity, I0 the initial light intensity, R the reflectance and x the number of reflections. In Fig. 1 a possible light-ray tracing diagram in a micromachined silicon light detection cell is shown. As can be seen in Fig. 1, the initial light intensity I0 is decreased due to in- and outcoupling by means of mirror planes, reflections in the detection channel itself (in the following these reflections are called ‘internal reflections’) and absorption by the solution. Using this information with Eq. (1) and the law of Lambert–Beer,
Fig. 1. Light-ray tracing diagram in a micromachined light detection cell.
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
333
Fig. 2. {111} planes as mirrors (a) and {110} planes as mirrors (b).
an expression for the intensity of the light coming out of a light detection cell is found: I = I0(R 2mirr R xint)e − mCd
(2)
in which I is the measured light intensity, I0 the initial light intensity, m the extinction coefficient of the absorbing compound, C the concentration of this compound, d the optical pathlength, Rmirr the reflectance of the mirrors, Rint the reflectance at the place of the internal reflections and x the number of internal reflections. It should be kept in mind that the numerical value of the reflectance Rint depends on the used materials and the solution flowing through the channel. For example, if the light detection cell is made completely in silicon, internal reflections are on the solution/silicon interfaces. However, if the light detection cell consists of silicon and Pyrex, there are internal reflections on the solution/silicon interfaces as well as on the solution/Pyrex interfaces. The reflectances of the solution/silicon interface and the solution/Pyrex interface are not the same. In the latter case, in Eq. (2) the term R xint needs to be replaced by: x2 1 R xint, Pyrex R int, silicon
(3)
in which Rint, Pyrex is the reflectance at the place of a solution/Pyrex interface, x1 the number of internal reflections on solution/Pyrex interfaces, Rint, silicon the reflectance at the place of a solution/ silicon interface and x2 the number of internal reflections on solution/silicon interfaces. The use of a detection cell for the determination of the concentration of a certain chemical compound is only useful if an acceptable efficiency in terms of intensity of the outcoming light is reached. Furthermore, for a high sensitivity the optical pathlength (d) should be in the mm-range. The optical path length of a light detection cell
can be increased easily by processing in the plane of wafers (as shown in Fig. 1) instead of perpendicular to wafer surfaces. For absorption measurements performed parallel to channels with extremely small cross-sectional dimensions (in the order of 50 mm), an optical detection system based on evanescent field sensing has been presented [6]. However, realizing a detection cell based on the absorption of the evanescent field requires multiple extra process steps. If the detection cell is fabricated using silicon and Pyrex, the application of a tracing diagram as shown in Fig. 1 for wavelengths in the range of 1330–1500 nm (IR-range) is reasonably efficient. This is because at these wavelengths the reflectance R of a silicon/water interface is 55– 64% [7]. However, in the wavelength range used in the ammonia detection system, 590–600 nm, the reflectance R of a silicon/water interface is only 20–30%. So when internal reflections occur if light with a wavelength of about 600 nm is used, the intensity of the outcoming light decreases very rapidly. Verpoorte et al. experienced this rapid decay in light intensity in their detection cell [8]. In order to reach a high optical pathlength as a means of improving detection limits, they designed and realized a light cell with multiple internal reflections. The light cell was made in silicon using standard KOH etching techniques. Therefore, the mirroring planes were {111} planes. These planes have an angle of 54.7° with respect to the surface of (100) silicon. Using a light source with a wavelength of 600 nm, after five reflections only 0.15% of the initial light power was left [8]. So for a high efficiency of a light detection cell using wavelengths in the visible range, the number of internal reflections in the detection channel has to be reduced. A further increase of the efficiency
334
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
of the light detection cell can be realized by the deposition of a coating with a high reflectance (R) in the detection channel. This is treated in Section 2.3. The minimization of the number of internal reflections can be done in two ways. First, if {111} planes are used as mirrors, the incident angle qi can be changed such that no internal reflections in a channel occur. The second way is the use of {110} planes as mirror planes. The {110} planes have an angle of 45° with respect to the surface of (100) oriented silicon. Both methods are shown in Fig. 2. The use of {111} planes may cause problems with the incoupling and outcoupling of light. If the refractive index of the sample varies, the light beam will not couple straight into the channel. This may cause unwanted internal reflections, resulting in a lower light efficiency of the light detection cell. The advantage of the use of {110} planes as mirrors is that light can be coupled in perpendicular to a Pyrex wafer, therewith avoiding unexpected internal reflections. Therefore, the use of 45° mirrors ({110} planes) is favored over using 54.7° mirrors ({111} planes).
2.2. Etching {110} planes in silicon In (100) oriented silicon {111} planes can be revealed using standard etchants like KOH. The fabrication of 45° mirrors is, however, less trivial. The literature describes two techniques to achieve 45° mirrors using anisotropic etching of silicon [9]: 1. making {111} mirrors on wafers that are cut 9.7° off the [100] axis. For example, the normal of the wafer is tilted 9.7° off the [100] direction towards the flat; 2. revealing {110} planes on (100) oriented wafers using etchants other than pure aqueous KOH. The major disadvantage of the first technique is that it is not possible to reveal two exactly opposite 45° mirrors (with respect to the surface of the wafer) in these wafers. Because of this disadvantage, the second technique will be used here.
In order to make {110} planes in (100) Si, masks should be aligned parallel to the 100 direction, that is 45° from the 110 flat. Furthermore, for revealing {110} planes an etchant is necessary in which the following relation between the etch rates R of different major crystal orientations ({100}, {110} and {111}) is available [10]Eq. (4)): R100 \ R110 \ R111
(4)
Pure aqueous KOH, a standard etchant for silicon, does not fulfill the requirement in Eq. (4). However, for several other etchants the mentioned condition is valid, like for EDP (solution consisting of ethylenediamine, water, pyrocatechol and pyrazine) type ‘B’ (1 l:320 ml:160 g:0 g) and type ‘S’ (1 l:133 ml:160 g:6 g) as well as for KOH with isopropyl alcohol (IPA) added to it or TMAH (tetramethyl ammonium hydroxide) [9– 12]. If the roughness of the {110} planes (measured directly after etching) is taken as a reference, EDP ‘B’ and TMAH give the best results. If KOH/IPA is used as etchant for revealing {110} planes, the use of a fresh etching solution in the range 30–35 wt.% and a temperature of about 50 °C is reported to be optimal in terms of roughness of the revealed planes [13].
2.3. Ways to impro6e the reflecting quality of {110} planes After etching the {110} planes the reflecting quality of these planes can be improved by several ways. First, the surface roughness of the {110} planes can be reduced. Second, the reflecting quality of mirror planes can be increased by the deposition of a coating with a high reflectivity on these planes. The surface roughness of etched {110} planes can be reduced significantly using repeated wet thermal oxidation steps and HF baths (for removing the grown oxide); [13] reports a reduction of the surface roughness of {111} planes by a factor 2–3. Furthermore, during processing of silicon the number of high temperature steps (e.g. annealing steps and oxidation steps) should be avoided as much as possible. It is reported that a clear difference was seen between the roughness of
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
sidewalls of wafers subjected to three high-temperature steps and wafers submitted to only one high-temperature step: the average initial roughness of the sidewalls was significantly lower for silicon submitted to one high-temperature step [13]. The deposition of a coating with a high reflectivity is another method for improving the reflecting quality of mirror planes. The application of a proper high reflectivity coating may improve the reflectance of the {110} planes by a factor 5 [14]. In theory, silver (Ag) is the best coating for wavelengths around 600 nm, but silver oxidizes when it is in contact with solutions containing water. To prevent these oxidation effects, platinum (Pt), aluminum (Al) and gold (Au) are good alternatives for silver (Ag).
3. Prototype realization The revealing of {110} planes in (100) oriented silicon is done with a KOH/IPA etchant. The 100 direction of the silicon wafers are found by means of a Vangbo alignment-mask [15]. Before the KOH/IPA etching step the number of hightemperature steps applied to the silicon is minimized to one high-temperature step: the growth of a silicon nitride layer at 850 °C (250 nm). This silicon nitride layer is used as mask during KOH/ IPA etching. After patterning the wafer parallel to the 100 direction, the silicon is etched using a 31 wt.%
335
KOH solution with 250 ml IPA (per 1000 ml H2O) added to it. The temperature of the solution was 50 °C. Measurements showed that the etch rate of the mentioned KOH/IPA solution was about 10 mm h − 1. In Fig. 3(a) a cross-section of an etched sample is shown. The depth of the channel is 200 mm. As can be seen, the sidewalls have an angle of 45° with respect to the surface. In Fig. 3(b) a SEM picture of a part of a KOH/ IPA etched channel is shown. From Fig. 3(b) follows that the {110} planes are relatively rough after etching. The roughness of the etched sidewalls is improved using wet thermal oxidation steps and HF baths. After two wet thermal oxidation steps (1 h; 1150 °C) and HF baths, the average roughness of {110} planes was reduced from 1700 to 93 nm (Sloan Dektak 3030; scanned length 500 mm). Two different types of detection cells are realized. The first series is KOH/IPA etched and improved with two thermal oxidation and HF baths (type A). In the second series of cells (type B) also a platinum coating (40 nm) is deposited on the whole interior of the detection channel, leaving uncovered spots allowing the entry and exit of the light beam. Not only does this coating increase the reflectivity of the mirror planes, it also protects the silicon (the {110} planes) against etching and/or oxidation effects. To increase adhesion a chromium (Cr) layer of 10 nm is deposited on the silicon and Pyrex before the deposition of the platinum layer.
Fig. 3. Cross section of {110} planes (a) and a SEM-picture (b).
336
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
Fig. 4. Photograph of a light detection cell (without Pt coating) and a schematic drawing.
After processing of the silicon bottom wafer, the etched structures are covered with Pyrex by means of anodic bonding. In Fig. 4 a photograph of a light detection cell is shown as well as a schematic drawing. Fluidic inlets and outlets are made using powder blasting [16]. The inlet and outlet are in the lower and upper part of the ‘S’-shaped channel structure in order to minimize dead volume in the detection channel. Concentration measurements are done in the central part of the ‘S’-shaped channel structure. The length of this light path, the actual detection channel, is in the range of 5.0– 7.0 mm, the depth of the structure is 200 mm and the width of the structure (at the topside) is 0.8 mm. So the volume of the channelpart in which concentration measurements are done is in the range of 600– 840 nl. The size of each detection cell is 1.2× 1.2 cm.
4. Experimental The measurement set up used for the characterization of the detection cell was the same for all experiments. Light was coupled in by a LED (Agilent Technologies, HLMP-EL08-VY000) mounted on a xy-stage. The exiting light intensity was measured by a photodiode (Silonex, SLSD-71N1) which was also mounted on a xystage. No lenses were used to enhance the focussing of the light. A lock-in amplifier (Model SR 830, Stanford Research) was used to read
the signal from the photodiode. This amplifier was also used to control the LED. Two measurement series were performed. In the first series, a comparison was made between the efficiency of the cells of type A and type B. In the second series, the best cell type (B) was used to measure the response of the light cell for the desired concentration range of indophenol. Indigo carmine was used as the light absorbing dye for the first experiment series. Indigo carmine has its absorption peak at 595 nm and therefore, will give a good indication of the performance of the light cells when indophenol will be the absorbing component. Various concentrations of indigo carmine were prepared by diluting the initial solution with demineralised water. The initial solution was made by dissolving an amount of indigo carmine powder (Sigma-Aldrich, I2387) in demineralised water. The concentrations of indigo carmine were in the range of 0 v/v% (pure demineralised water) to 50 v/v%. The measurement results are shown in Fig. 5. The intensity of the incoming light (I0) was kept constant for both measurement series. The concentration of indophenol blue solutions was ranging from 0.02 to 50 mM. The solutions were made by dissolving indophenol powder (Sigma-Aldrich, 57412, standard Fluka purity) in demineralised water. The mentioned range exceeds the concentration range of the MAFIAS specifications (0.15–15 mM).
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
5. Results and discussion As can be seen in Fig. 5, light detection cells of type B give a higher output signal then type A. This output signal is defined as the difference of the absorption plateau and the baseline (Fig. 6(b)). The baseline corresponds to the photodiode-output (which is directly related to the exiting light intensity) for demineralised water. This baseline will be different for different cells: since platinum has a higher reflectance than silicon, cells of type B will have a higher baseline signal then cells of type A. Since cells of type B have a higher initial light intensity output (baseline value), the effect of light absorbance will be stronger for these cells compared to cells of type A. Absorbance takes away a certain percentage of light (a certain percentage of the baseline). For equal concentrations in both cell types an identical percentage of the light is absorbed by the solution. Since the baseline for type B cells is higher than for type A cells, the peak resulting from a sample will be deeper for type B cells.
337
So, for equal concentrations the output signal (as defined in Fig. 6(b)) will be larger for type B cells than for type A cells. This effect is clearly visible in Fig. 5, in which the output signals of cell type A are about 73% of the values of cell type B. As was mentioned in Section 4, for all measurements the initial light intensity (I0) was the same. Each point in Fig. 5 consists of three independent datapoints. The standard deviation of these series varies between 0.009 and 0.072 nA. With respect to a total signal range of 10 nA, it can be concluded that the measurements show good reproducibility. Subsequently, the cells of type B were used for the determination of various concentrations of indophenol blue. Single absorption measurements of the highest (50 mM) and lowest (0.02 mM) concentrations are given in Fig. 6. As can be seen the injected sample plugs are detected clearly. The response of the system as shown in Fig. 6(a) after the plug has passed the light cell may be a result of the actuation of the valve that is used to change between the sample fluid and a reference fluid (demineralised water). In Fig. 6(b) this actuation is not visible, since the output signal of the
Fig. 5. Photodiode output signal as function of the indigo carmine concentration; optical path length 6 mm. The upper line represents a measurement series performed with a light detection cell in which a 40 nm platinum (Pt) coating was deposited on the interior. The lower line is a measurement series performed with a cell without a coating.
338
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
Fig. 6. Time dependent detector signal: lowest concentration measurement of indophenol blue (a; 0.02 mM) and highest concentration (b; 50 mM).
photodiode is much larger than in the case of Fig. 6(a). The used photodiode operates linear over 6 decades, whereas the lock-in amplifier (Model SR 830, Stanford Research) takes unwanted offsets away. The detection limit of this lock-in amplifier is 10 − 15 A. All measured signal outputs were well above this value (] 5 ×10 − 13 A). For our measurements the value of mCd (Eq. (2)) was very small compared to unity (for indophenol concentrations of 0.02– 50 mM, the value of mCd is in the range 2×10 − 6 – 40 ×10 − 3). For these values, the Lambert– Beer law can be simplified to a linear approximation. In Fig. 7 all concentration measurement series are presented with the expected linear fit. Though the output signal of the detection cell is not linear (slope of trendline is 0.83 instead of unity), it is clearly possible to measure a concentration range of 3 decades in the range needed for the MAFIAS system (0.15–15 mM/0.03 – 3 mg l − 1). The S/N ratio for the low-level concentrations is estimated to be in the order of 10. Therefore, the theoretical detection limit of the presented detection cell lies around 0.002 mM indophenol.
revealed using a KOH/IPA etchant. The surface roughness of {110} planes is reduced from 1700 to 93 nm by thermal wet oxidation steps and HF baths. The reflecting quality of these {110} planes is further improved by depositing a platinum coating (40 nm) on the whole interior of the detection cell. The light detection cell is designed for the determination of indophenol blue concentrations in the range 0.15–15 mM. Concentration measurements with indophenol were done in the range of 0.02– 50 mM. The measurements showed high reproducibility. Therefore, it is clear that the indophenol concentration range which can be expected in the ammonia micro analysis system can be measured easily.
6. Conclusions A light detection cell has been designed, realized and tested. Light is coupled in and out a channel using 45° mirrors. These mirrors are {110} planes in (100) oriented silicon which are
Fig. 7. Measured output signal as function of the concentration of indophenol blue.
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
Acknowledgements This research was funded by the Dutch Science Foundation (STW, TEL 3467). J.W. Berenschot, M.J. de Boer, A.J. Nijdam and H. Wensink are thanked for their technical assistance and/or fruitful discussions.
References [1] T.T. Veenstra, MAFIAS —An Integrated Lab-on-a-Chip for Ammonium Measurement, PhD Thesis, University of Twente, 2001. [2] P.L. Searle, The Analyst 109 (1984) 549. [3] C.J. Patton, S.R. Crouch, Anal. Chem. 49 (3) (1977) 464. [4] T.T. Ngo, A.P.H. Phan, C.F. Yam, H.M. Lenhoff, Anal. Chem. 54 (1982) 46. [5] R.G. Harfmann, S.R. Crouch, Talanta 36 (1989) 261. [6] G. Pandraud, T.M. Koster, C. Gui, M. Dijkstra, A. van den Berg, P.V. Lambeck, Sens. Act. A85 (2000) 158.
339
[7] D.J. Sadler, M.J. Garter, C.H. Ahn, S. Koh, A.L. Cook, J. Micromech. Microeng. 7 (1997) 263. [8] E. Verpoorte, A. Manz, H. Lu¨ di, A.E. Bruno, F. Maystre, B. Krattiger, H.M. Widmer, B.H. van der Schoot, N.F. de Rooij, Sens. Act. B6 (1992) 66. [9] C. Strandman, L. Rosengren, H.G.A. Elderstig, Y. Ba¨ cklund, J. Microelectro-mech. Sys. 4 (4) (1995) 213. [10] H. Seidel, L. Csepregi, A. Heuberger, H. Baumga¨ rtel, J. Electrochem. Soc. 137 (11) (1990) 3612. [11] Y. Ba¨ cklund, L. Rosengren, J. Micromech. Microeng. 2 (1992) 75. [12] M. Sekimura, H. Naruse, Tech. Digest. 10th Int. Conf. On Solid-State Sensor and Act. (Transducers ’99), Sendai, Japan, June 7 – 10, 1999, pp. 550 – 551. [13] T.A. Kwa, P.J. French, R.F. Wolffenbuttel, P.M. Sarro, L. Hellemans, J. Snauwaert, J. Electrochem. Soc. 142 (4) (1995) 1226. [14] J.E. van der Linden, P.M. de Dobbelaere, P.P. van Daele, M.B. Diemeer, Jpn. J. Appl. Phys. 37 (1998) 3730. [15] M. Vangbo, Y. Ba¨ cklund, J. Micromech. Microeng. 6 (1996) 279. [16] H. Wensink, J.W. Berenschot, H.V. Jansen, M.C. Elwenspoek, Proc. 13th IEEE Int. Conf. On Micro Electro Mechanical Systems (MEMS-2000), Miyazaki, Japan, January 23 – 27, 2000, pp. 769 – 774.
A light detection cell to be used in a micro analysis system for ammonia R.M. Tiggelaar *, T.T. Veenstra, R.G.P. Sanders, J.G.E. Gardeniers, M.C. Elwenspoek, A. van den Berg MESA + Research Institute, Uni6ersity of Twente (EL-TT), P.O. Box 217, 7500 AE, Enschede, The Netherlands Received 11 April 2001; received in revised form 20 August 2001; accepted 23 August 2001
Abstract This paper describes the design, realization and characterization of a micromachined light detection cell. This light detection cell is designed to meet the specifications needed for a micro total analysis system in which ammonia is converted to indophenol blue. The concentration of indophenol blue is measured in a light detection cell. The light detection cell was created using KOH/IPA etching of silicon. The KOH/IPA etchant was a 31 wt.% potassium hydroxide (KOH) solution with 250 ml isopropyl alcohol (IPA) per 1000 ml H2O added to it. The temperature of the solution was 50 °C. Etching with KOH/IPA results in 45° sidewalls ({110} planes) which can be used for the in- and outcoupling of the light. The internal volume of the realized light detection cell is smaller than 1 ml, enabling measurements on samples in the order of only 1 ml. Measurements were performed on indophenol blue samples in the range of 0.02 to 50 mM. In this range the measurements showed good reproducibility. © 2002 Elsevier Science B.V. All rights reserved. Keywords: KOH/IPA etching; Light detection cell; Berthelot; Ammonia detection
1. Introduction The light detection cell presented in this paper is part of a micro analysis system for ammonia. This ammonia detection system or MAFIAS (Micro Ammonia Fluid Injection Analysis System) is a micro total analysis system (m-TAS) which can be used for the determination of the concentration of ammonia in water [1]. In MAFIAS concentra* Corresponding author. Tel.: +31-53-489-4420; fax: +3153-489-3343. E-mail address: [email protected] (R.M. Tiggelaar).
tion measurements are carried out using an optical detection method, since this method is used most commonly in micro total analysis systems. Furthermore, the internal volumes of micro total analysis systems in general are small (order of 10 ml) and therefore, the size of the sample used in the light detection cell has to be in the microliterrange. The MAFIAS is made of silicon and Pyrex and the light detection cell will be made of the same materials to ensure that monolithic integration of the cell with the analysis system will be possible. The latter aspect is important for reduction of sample volume and sample dispersion.
0039-9140/02/$ - see front matter © 2002 Elsevier Science B.V. All rights reserved. PII: S 0 0 3 9 - 9 1 4 0 ( 0 1 ) 0 0 5 9 9 - 9
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
332
The analytical chemical method applied in MAFIAS is the Berthelot reaction mechanism [1 –5]. Using this reaction mechanism the ammonia content of a sample reacts with hypochlorite and phenol to form indophenol blue. The complete conversion takes several minutes and the reaction temperature should be lower than 38 °C to avoid bubbles in the solution. After conversion, the concentration of indophenol blue is determined using the light detection cell. The concentration of this substance is directly related to the initial ammonia concentration. For the development of a practical analysis system for continuous measurement of ammonia samples it was specified that the ammonia concentration should be detectable in the range 6– 600 mM [1]. After conversion by the Berthelot reaction this results in concentrations of indophenol blue in the range 0.15–15 mM. The concentrations of the used starting reagents in the Berthelot reaction are 5.8 mM hypochlorite and 5.8 mM phenol. Though hypochlorite is a strong oxidant and therefore, might be expected to etch silicon, the hypochlorite concentrations in the system will be as low as 75 mg l − 1, which is considered to be of minor significance with respect to the etching of silicon. Furthermore, the system is flushed thoroughly after each sample.
2. Design considerations
2.1. Effects of reflections, light in- and outcoupling The intensity of the light exiting from a light cell is influenced by two effects. The first effect is
the absorption of light by the solution (Lambert– Beer law). The second effect which influences the intensity of the outcoming light is reflection of light at surfaces between different media. If a beam of light hits a surface, part of the light is reflected from that surface and another part is transmitted into the medium behind the surface. This is quantified by the reflectance R, which is the ratio of the reflected power to the incident power and strongly depends on the incident angle qi as well as on the refractive indices of the incident and reflecting media. Using Snell’s law of reflection and Fresnel’s equations, the reflectance R as a function of the incident angle qi can be calculated for interfaces between different media. At constant angle of incidence (qi), the intensity of a beam of light travelling through a channel with a non-absorbing solution is only influenced by the number of reflections. In this special case the decrease in light intensity can be expressed as a function of the reflectance R Eq. (1): I= R xI0
(1)
in which I is the measured light intensity, I0 the initial light intensity, R the reflectance and x the number of reflections. In Fig. 1 a possible light-ray tracing diagram in a micromachined silicon light detection cell is shown. As can be seen in Fig. 1, the initial light intensity I0 is decreased due to in- and outcoupling by means of mirror planes, reflections in the detection channel itself (in the following these reflections are called ‘internal reflections’) and absorption by the solution. Using this information with Eq. (1) and the law of Lambert–Beer,
Fig. 1. Light-ray tracing diagram in a micromachined light detection cell.
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
333
Fig. 2. {111} planes as mirrors (a) and {110} planes as mirrors (b).
an expression for the intensity of the light coming out of a light detection cell is found: I = I0(R 2mirr R xint)e − mCd
(2)
in which I is the measured light intensity, I0 the initial light intensity, m the extinction coefficient of the absorbing compound, C the concentration of this compound, d the optical pathlength, Rmirr the reflectance of the mirrors, Rint the reflectance at the place of the internal reflections and x the number of internal reflections. It should be kept in mind that the numerical value of the reflectance Rint depends on the used materials and the solution flowing through the channel. For example, if the light detection cell is made completely in silicon, internal reflections are on the solution/silicon interfaces. However, if the light detection cell consists of silicon and Pyrex, there are internal reflections on the solution/silicon interfaces as well as on the solution/Pyrex interfaces. The reflectances of the solution/silicon interface and the solution/Pyrex interface are not the same. In the latter case, in Eq. (2) the term R xint needs to be replaced by: x2 1 R xint, Pyrex R int, silicon
(3)
in which Rint, Pyrex is the reflectance at the place of a solution/Pyrex interface, x1 the number of internal reflections on solution/Pyrex interfaces, Rint, silicon the reflectance at the place of a solution/ silicon interface and x2 the number of internal reflections on solution/silicon interfaces. The use of a detection cell for the determination of the concentration of a certain chemical compound is only useful if an acceptable efficiency in terms of intensity of the outcoming light is reached. Furthermore, for a high sensitivity the optical pathlength (d) should be in the mm-range. The optical path length of a light detection cell
can be increased easily by processing in the plane of wafers (as shown in Fig. 1) instead of perpendicular to wafer surfaces. For absorption measurements performed parallel to channels with extremely small cross-sectional dimensions (in the order of 50 mm), an optical detection system based on evanescent field sensing has been presented [6]. However, realizing a detection cell based on the absorption of the evanescent field requires multiple extra process steps. If the detection cell is fabricated using silicon and Pyrex, the application of a tracing diagram as shown in Fig. 1 for wavelengths in the range of 1330–1500 nm (IR-range) is reasonably efficient. This is because at these wavelengths the reflectance R of a silicon/water interface is 55– 64% [7]. However, in the wavelength range used in the ammonia detection system, 590–600 nm, the reflectance R of a silicon/water interface is only 20–30%. So when internal reflections occur if light with a wavelength of about 600 nm is used, the intensity of the outcoming light decreases very rapidly. Verpoorte et al. experienced this rapid decay in light intensity in their detection cell [8]. In order to reach a high optical pathlength as a means of improving detection limits, they designed and realized a light cell with multiple internal reflections. The light cell was made in silicon using standard KOH etching techniques. Therefore, the mirroring planes were {111} planes. These planes have an angle of 54.7° with respect to the surface of (100) silicon. Using a light source with a wavelength of 600 nm, after five reflections only 0.15% of the initial light power was left [8]. So for a high efficiency of a light detection cell using wavelengths in the visible range, the number of internal reflections in the detection channel has to be reduced. A further increase of the efficiency
334
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
of the light detection cell can be realized by the deposition of a coating with a high reflectance (R) in the detection channel. This is treated in Section 2.3. The minimization of the number of internal reflections can be done in two ways. First, if {111} planes are used as mirrors, the incident angle qi can be changed such that no internal reflections in a channel occur. The second way is the use of {110} planes as mirror planes. The {110} planes have an angle of 45° with respect to the surface of (100) oriented silicon. Both methods are shown in Fig. 2. The use of {111} planes may cause problems with the incoupling and outcoupling of light. If the refractive index of the sample varies, the light beam will not couple straight into the channel. This may cause unwanted internal reflections, resulting in a lower light efficiency of the light detection cell. The advantage of the use of {110} planes as mirrors is that light can be coupled in perpendicular to a Pyrex wafer, therewith avoiding unexpected internal reflections. Therefore, the use of 45° mirrors ({110} planes) is favored over using 54.7° mirrors ({111} planes).
2.2. Etching {110} planes in silicon In (100) oriented silicon {111} planes can be revealed using standard etchants like KOH. The fabrication of 45° mirrors is, however, less trivial. The literature describes two techniques to achieve 45° mirrors using anisotropic etching of silicon [9]: 1. making {111} mirrors on wafers that are cut 9.7° off the [100] axis. For example, the normal of the wafer is tilted 9.7° off the [100] direction towards the flat; 2. revealing {110} planes on (100) oriented wafers using etchants other than pure aqueous KOH. The major disadvantage of the first technique is that it is not possible to reveal two exactly opposite 45° mirrors (with respect to the surface of the wafer) in these wafers. Because of this disadvantage, the second technique will be used here.
In order to make {110} planes in (100) Si, masks should be aligned parallel to the 100 direction, that is 45° from the 110 flat. Furthermore, for revealing {110} planes an etchant is necessary in which the following relation between the etch rates R of different major crystal orientations ({100}, {110} and {111}) is available [10]Eq. (4)): R100 \ R110 \ R111
(4)
Pure aqueous KOH, a standard etchant for silicon, does not fulfill the requirement in Eq. (4). However, for several other etchants the mentioned condition is valid, like for EDP (solution consisting of ethylenediamine, water, pyrocatechol and pyrazine) type ‘B’ (1 l:320 ml:160 g:0 g) and type ‘S’ (1 l:133 ml:160 g:6 g) as well as for KOH with isopropyl alcohol (IPA) added to it or TMAH (tetramethyl ammonium hydroxide) [9– 12]. If the roughness of the {110} planes (measured directly after etching) is taken as a reference, EDP ‘B’ and TMAH give the best results. If KOH/IPA is used as etchant for revealing {110} planes, the use of a fresh etching solution in the range 30–35 wt.% and a temperature of about 50 °C is reported to be optimal in terms of roughness of the revealed planes [13].
2.3. Ways to impro6e the reflecting quality of {110} planes After etching the {110} planes the reflecting quality of these planes can be improved by several ways. First, the surface roughness of the {110} planes can be reduced. Second, the reflecting quality of mirror planes can be increased by the deposition of a coating with a high reflectivity on these planes. The surface roughness of etched {110} planes can be reduced significantly using repeated wet thermal oxidation steps and HF baths (for removing the grown oxide); [13] reports a reduction of the surface roughness of {111} planes by a factor 2–3. Furthermore, during processing of silicon the number of high temperature steps (e.g. annealing steps and oxidation steps) should be avoided as much as possible. It is reported that a clear difference was seen between the roughness of
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
sidewalls of wafers subjected to three high-temperature steps and wafers submitted to only one high-temperature step: the average initial roughness of the sidewalls was significantly lower for silicon submitted to one high-temperature step [13]. The deposition of a coating with a high reflectivity is another method for improving the reflecting quality of mirror planes. The application of a proper high reflectivity coating may improve the reflectance of the {110} planes by a factor 5 [14]. In theory, silver (Ag) is the best coating for wavelengths around 600 nm, but silver oxidizes when it is in contact with solutions containing water. To prevent these oxidation effects, platinum (Pt), aluminum (Al) and gold (Au) are good alternatives for silver (Ag).
3. Prototype realization The revealing of {110} planes in (100) oriented silicon is done with a KOH/IPA etchant. The 100 direction of the silicon wafers are found by means of a Vangbo alignment-mask [15]. Before the KOH/IPA etching step the number of hightemperature steps applied to the silicon is minimized to one high-temperature step: the growth of a silicon nitride layer at 850 °C (250 nm). This silicon nitride layer is used as mask during KOH/ IPA etching. After patterning the wafer parallel to the 100 direction, the silicon is etched using a 31 wt.%
335
KOH solution with 250 ml IPA (per 1000 ml H2O) added to it. The temperature of the solution was 50 °C. Measurements showed that the etch rate of the mentioned KOH/IPA solution was about 10 mm h − 1. In Fig. 3(a) a cross-section of an etched sample is shown. The depth of the channel is 200 mm. As can be seen, the sidewalls have an angle of 45° with respect to the surface. In Fig. 3(b) a SEM picture of a part of a KOH/ IPA etched channel is shown. From Fig. 3(b) follows that the {110} planes are relatively rough after etching. The roughness of the etched sidewalls is improved using wet thermal oxidation steps and HF baths. After two wet thermal oxidation steps (1 h; 1150 °C) and HF baths, the average roughness of {110} planes was reduced from 1700 to 93 nm (Sloan Dektak 3030; scanned length 500 mm). Two different types of detection cells are realized. The first series is KOH/IPA etched and improved with two thermal oxidation and HF baths (type A). In the second series of cells (type B) also a platinum coating (40 nm) is deposited on the whole interior of the detection channel, leaving uncovered spots allowing the entry and exit of the light beam. Not only does this coating increase the reflectivity of the mirror planes, it also protects the silicon (the {110} planes) against etching and/or oxidation effects. To increase adhesion a chromium (Cr) layer of 10 nm is deposited on the silicon and Pyrex before the deposition of the platinum layer.
Fig. 3. Cross section of {110} planes (a) and a SEM-picture (b).
336
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
Fig. 4. Photograph of a light detection cell (without Pt coating) and a schematic drawing.
After processing of the silicon bottom wafer, the etched structures are covered with Pyrex by means of anodic bonding. In Fig. 4 a photograph of a light detection cell is shown as well as a schematic drawing. Fluidic inlets and outlets are made using powder blasting [16]. The inlet and outlet are in the lower and upper part of the ‘S’-shaped channel structure in order to minimize dead volume in the detection channel. Concentration measurements are done in the central part of the ‘S’-shaped channel structure. The length of this light path, the actual detection channel, is in the range of 5.0– 7.0 mm, the depth of the structure is 200 mm and the width of the structure (at the topside) is 0.8 mm. So the volume of the channelpart in which concentration measurements are done is in the range of 600– 840 nl. The size of each detection cell is 1.2× 1.2 cm.
4. Experimental The measurement set up used for the characterization of the detection cell was the same for all experiments. Light was coupled in by a LED (Agilent Technologies, HLMP-EL08-VY000) mounted on a xy-stage. The exiting light intensity was measured by a photodiode (Silonex, SLSD-71N1) which was also mounted on a xystage. No lenses were used to enhance the focussing of the light. A lock-in amplifier (Model SR 830, Stanford Research) was used to read
the signal from the photodiode. This amplifier was also used to control the LED. Two measurement series were performed. In the first series, a comparison was made between the efficiency of the cells of type A and type B. In the second series, the best cell type (B) was used to measure the response of the light cell for the desired concentration range of indophenol. Indigo carmine was used as the light absorbing dye for the first experiment series. Indigo carmine has its absorption peak at 595 nm and therefore, will give a good indication of the performance of the light cells when indophenol will be the absorbing component. Various concentrations of indigo carmine were prepared by diluting the initial solution with demineralised water. The initial solution was made by dissolving an amount of indigo carmine powder (Sigma-Aldrich, I2387) in demineralised water. The concentrations of indigo carmine were in the range of 0 v/v% (pure demineralised water) to 50 v/v%. The measurement results are shown in Fig. 5. The intensity of the incoming light (I0) was kept constant for both measurement series. The concentration of indophenol blue solutions was ranging from 0.02 to 50 mM. The solutions were made by dissolving indophenol powder (Sigma-Aldrich, 57412, standard Fluka purity) in demineralised water. The mentioned range exceeds the concentration range of the MAFIAS specifications (0.15–15 mM).
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
5. Results and discussion As can be seen in Fig. 5, light detection cells of type B give a higher output signal then type A. This output signal is defined as the difference of the absorption plateau and the baseline (Fig. 6(b)). The baseline corresponds to the photodiode-output (which is directly related to the exiting light intensity) for demineralised water. This baseline will be different for different cells: since platinum has a higher reflectance than silicon, cells of type B will have a higher baseline signal then cells of type A. Since cells of type B have a higher initial light intensity output (baseline value), the effect of light absorbance will be stronger for these cells compared to cells of type A. Absorbance takes away a certain percentage of light (a certain percentage of the baseline). For equal concentrations in both cell types an identical percentage of the light is absorbed by the solution. Since the baseline for type B cells is higher than for type A cells, the peak resulting from a sample will be deeper for type B cells.
337
So, for equal concentrations the output signal (as defined in Fig. 6(b)) will be larger for type B cells than for type A cells. This effect is clearly visible in Fig. 5, in which the output signals of cell type A are about 73% of the values of cell type B. As was mentioned in Section 4, for all measurements the initial light intensity (I0) was the same. Each point in Fig. 5 consists of three independent datapoints. The standard deviation of these series varies between 0.009 and 0.072 nA. With respect to a total signal range of 10 nA, it can be concluded that the measurements show good reproducibility. Subsequently, the cells of type B were used for the determination of various concentrations of indophenol blue. Single absorption measurements of the highest (50 mM) and lowest (0.02 mM) concentrations are given in Fig. 6. As can be seen the injected sample plugs are detected clearly. The response of the system as shown in Fig. 6(a) after the plug has passed the light cell may be a result of the actuation of the valve that is used to change between the sample fluid and a reference fluid (demineralised water). In Fig. 6(b) this actuation is not visible, since the output signal of the
Fig. 5. Photodiode output signal as function of the indigo carmine concentration; optical path length 6 mm. The upper line represents a measurement series performed with a light detection cell in which a 40 nm platinum (Pt) coating was deposited on the interior. The lower line is a measurement series performed with a cell without a coating.
338
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
Fig. 6. Time dependent detector signal: lowest concentration measurement of indophenol blue (a; 0.02 mM) and highest concentration (b; 50 mM).
photodiode is much larger than in the case of Fig. 6(a). The used photodiode operates linear over 6 decades, whereas the lock-in amplifier (Model SR 830, Stanford Research) takes unwanted offsets away. The detection limit of this lock-in amplifier is 10 − 15 A. All measured signal outputs were well above this value (] 5 ×10 − 13 A). For our measurements the value of mCd (Eq. (2)) was very small compared to unity (for indophenol concentrations of 0.02– 50 mM, the value of mCd is in the range 2×10 − 6 – 40 ×10 − 3). For these values, the Lambert– Beer law can be simplified to a linear approximation. In Fig. 7 all concentration measurement series are presented with the expected linear fit. Though the output signal of the detection cell is not linear (slope of trendline is 0.83 instead of unity), it is clearly possible to measure a concentration range of 3 decades in the range needed for the MAFIAS system (0.15–15 mM/0.03 – 3 mg l − 1). The S/N ratio for the low-level concentrations is estimated to be in the order of 10. Therefore, the theoretical detection limit of the presented detection cell lies around 0.002 mM indophenol.
revealed using a KOH/IPA etchant. The surface roughness of {110} planes is reduced from 1700 to 93 nm by thermal wet oxidation steps and HF baths. The reflecting quality of these {110} planes is further improved by depositing a platinum coating (40 nm) on the whole interior of the detection cell. The light detection cell is designed for the determination of indophenol blue concentrations in the range 0.15–15 mM. Concentration measurements with indophenol were done in the range of 0.02– 50 mM. The measurements showed high reproducibility. Therefore, it is clear that the indophenol concentration range which can be expected in the ammonia micro analysis system can be measured easily.
6. Conclusions A light detection cell has been designed, realized and tested. Light is coupled in and out a channel using 45° mirrors. These mirrors are {110} planes in (100) oriented silicon which are
Fig. 7. Measured output signal as function of the concentration of indophenol blue.
R.M. Tiggelaar et al. / Talanta 56 (2002) 331–339
Acknowledgements This research was funded by the Dutch Science Foundation (STW, TEL 3467). J.W. Berenschot, M.J. de Boer, A.J. Nijdam and H. Wensink are thanked for their technical assistance and/or fruitful discussions.
References [1] T.T. Veenstra, MAFIAS —An Integrated Lab-on-a-Chip for Ammonium Measurement, PhD Thesis, University of Twente, 2001. [2] P.L. Searle, The Analyst 109 (1984) 549. [3] C.J. Patton, S.R. Crouch, Anal. Chem. 49 (3) (1977) 464. [4] T.T. Ngo, A.P.H. Phan, C.F. Yam, H.M. Lenhoff, Anal. Chem. 54 (1982) 46. [5] R.G. Harfmann, S.R. Crouch, Talanta 36 (1989) 261. [6] G. Pandraud, T.M. Koster, C. Gui, M. Dijkstra, A. van den Berg, P.V. Lambeck, Sens. Act. A85 (2000) 158.
339
[7] D.J. Sadler, M.J. Garter, C.H. Ahn, S. Koh, A.L. Cook, J. Micromech. Microeng. 7 (1997) 263. [8] E. Verpoorte, A. Manz, H. Lu¨ di, A.E. Bruno, F. Maystre, B. Krattiger, H.M. Widmer, B.H. van der Schoot, N.F. de Rooij, Sens. Act. B6 (1992) 66. [9] C. Strandman, L. Rosengren, H.G.A. Elderstig, Y. Ba¨ cklund, J. Microelectro-mech. Sys. 4 (4) (1995) 213. [10] H. Seidel, L. Csepregi, A. Heuberger, H. Baumga¨ rtel, J. Electrochem. Soc. 137 (11) (1990) 3612. [11] Y. Ba¨ cklund, L. Rosengren, J. Micromech. Microeng. 2 (1992) 75. [12] M. Sekimura, H. Naruse, Tech. Digest. 10th Int. Conf. On Solid-State Sensor and Act. (Transducers ’99), Sendai, Japan, June 7 – 10, 1999, pp. 550 – 551. [13] T.A. Kwa, P.J. French, R.F. Wolffenbuttel, P.M. Sarro, L. Hellemans, J. Snauwaert, J. Electrochem. Soc. 142 (4) (1995) 1226. [14] J.E. van der Linden, P.M. de Dobbelaere, P.P. van Daele, M.B. Diemeer, Jpn. J. Appl. Phys. 37 (1998) 3730. [15] M. Vangbo, Y. Ba¨ cklund, J. Micromech. Microeng. 6 (1996) 279. [16] H. Wensink, J.W. Berenschot, H.V. Jansen, M.C. Elwenspoek, Proc. 13th IEEE Int. Conf. On Micro Electro Mechanical Systems (MEMS-2000), Miyazaki, Japan, January 23 – 27, 2000, pp. 769 – 774.