A factorial design approach to fracture pressure tests of
microfluidic BF33 and D263T glass chips
with side-port capillary connections
D. Jonker1, H.W. Veltkamp2, R.G.P. Sanders1, S. Schlautmann1, K. Giannasi3*, R.M. Tiggelaar1,4 and J.G.E. Gardeniers1#.
1 Mesoscale Chemical Systems Group, MESA+ Institute for Nanotechnology, University of Twente, P.O. Box 217, 7500 AE, Enschede, The Netherlands
2 Integrated Devices and Systems Group, MESA+ Institute for Nanotechnology, University of Twente, P.O. Box 217, 7500 AE, Enschede, The Netherlands
3 Micronit Microfluidics B.V., Colosseum 15, 7521 PV Enschede, The Netherlands
4 NanoLab cleanroom, MESA+ Institute for Nanotechnology, University of Twente, P.O. Box 217, 7500 AE, Enschede, The Netherlands
# Corresponding author: j.g.e.gardeniers@utwente.nl
* Currently at: Tide Microfluidics B.V., Capitool 41, 7521 PL Enschede, The Netherlands
Abstract.
The pressure stability of microfluidic glass chips was tested experimentally, with a special focus on the inserts for glued capillary connections. Destructive high-pressure experiments with demineralized water conducted at room temperature showed a difference in mean fracture pressure between the two tested glass types BF33 and D263T, with values of 192 ±25 and 159 ±25 bar, respectively. For BF33, hydrofluoric acid (HF) etching of the powder blasted (abrasive jet machined) chip insert increased the mean fracture pressure with 43 ±9 bar, whilst for D263T a decrease of -22 ±8 bar resulted. Contrary to the expected surface smoothening of the HF treatment, a rougher surface was obtained, particularly for the case of D263T, which is thought to be due to the opening of median (radial) cracks caused by the powder impact during the blasting process. The roughness obscures the effect of the tapering of the insert, preventing that factor from having a statistically significant effect on the mean fracture pressure. Nevertheless, a decrease in the mean fracture pressure and a decrease in the variance of the mean fracture pressure was observed when a taper is introduced, whereas the fracture location tends to move away from the insert-microchannel intersection towards the glue meniscus. A practical solution for cases where a high-pressure stability is required is found in applying a metal clamp around the capillary insert section. This significantly increased the fracture pressure of the chip insert section with 50 ±21 bar, by preventing bond release.
Keywords: high-pressure microfluidics, mechanical chip testing, glass fracture strength, chip
1 Introduction.
High-pressure chemistry is an example of a field, where the introduction of microfluidics has enabled scientists to conduct lab scale experiments under relatively safe circumstances with a minimal inventory of potentially hazardous reactive intermediates, following the "smaller is safer" aphorism introduced by Hendershot in 2000 [1]. High-pressure microreactors can be helpful for (fast) screening of process conditions outside of the conventional reaction parameter windows ("novel process windows" [2]). The microfluidic setting offers additional advantages, such as in-line connection to work-up units for separation of products [3,4] or to process analytics [5].
A particularly important aspect of flow chemistry based on microfluidic glass chip technology (typical channel dimension ~100µm) is that in-situ visual inspection by optical microscopy offers new opportunities to study fundamental physics and chemistry of high-pressure processes, such as phase changes and flow regimes of liquid-gas mixtures (including near- and supercritical fluids) [6,7,8,9,10]. Such studies require the microfluidic systems to be stable for pressures up to 100 bars [11], or even above, if supercritical water experiments are the goal [12].
Fracture strengths of microfluidic chips have previously been tested by several research groups [13,14], including our team [15,16]. Besides the stability of the glass chips themselves, of which we have shown that they can withstand pressures up to ca. 700 bars [17], the mechanical stability of the connection of the chip to peripheral equipment, such as valves or pumps, is highly relevant, and often is the weakest point in the setup. An additional desired property of a chip interface is that it has an as low as possible dead volume. This is particularly relevant for applications as HPLC on a chip with its requirement for a minimized residence time distribution i.e. minimal plate height (see [18,19] for reviews).
Several options for connecting a microfluidic chip have been reviewed by Temiz et al. [20]. For high pressure capillary-to-chip connections two main configurations are common, "side-port" (capillary in a microchannel insert in-plane with chip) [13,17,21] and "top-port" (capillary connected perpendicular to chip surface) [22,23]. Top-port capillary connectors attached to a chip by clamping typically achieve sealing pressures between 50 and 100 bars, with dead volumes suitable for high-end HPLC applications [24,25]. Side-port and top-port connections, varying from the use of clamps and seals to laser welding, have been compared by Lotter et al., who found minimum pressure stability values ranging from 125 bar for the laser-welded side-port connection to 500 bar for the clamp top-port connection [26]. These authors state that side-port connectors generally give much lower dead volume than top-port connectors. In our previous work, epoxy glue was used to fix a capillary into a side-port, being an insert channel micromachined in a glass chip, exhibiting fracture pressures from ca. 200 bar for side-ports with a width of 400 µm up to ca. 690 bar for 135 µm wide ports [17]. A similar approach was taken by Calleweart et al. for HPLC chips, where a minimal fracture pressure of 350 bars was observed for 118 µm deep, almost square inserts etched in a silicon substrate that was anodically bonded to a Pyrex glass plate [27].
The observation that higher fracture pressures are obtained for smaller inserts [17] might be due to an increased probability of encountering a significant microcrack on larger surfaces [28]. Failure of glass microchips can be in two ways, either by the release of the bond between the parts from which the microfluidic chip was created, resulting in delamination [16], or by cracking of the chip bulk material. In the case of bond release, factors like the hydraulic diameter, deformation stresses, and effective bond energy are responsible for the maximum pressure which can be applied to the microfluidic chip. The bond release is caused by the dominance of a hydraulic pressure term over effective bonding energy and deformation stresses [15]. In the case of cracking it is found that the initiation of failure is due to microcracks at the surface of the fluid-solid interface, which may be the result of a defective surface finish caused by processing steps to fabricate the chosen insert geometry [17]. Enhanced slow crack growth at higher temperatures (studied within the range 11-125 °C [13]) was found to have a negative effect on the maximum operating pressure and durability of glass chips.
Following up on our earlier work [17], we conduct destructive pressure tests on "side-port" chip-to-capillary interfaces, in order to identify the design factors that have a significant effect on the mean value of the fracture pressure of side-port inserts in microfluidic chips. Indications are given for the operating ranges of the hydraulic pressures in which the chosen geometries function without failing. The experimental strategy follows factorial design methods, where the analysis of variance (ANOVA)
method is used to draw statistically relevant conclusions by means of the F-test. In order to construct more reliable results, power transformations are used to generate data sets that give a better approach to the normal distribution of failure data. The transformed data sets are based on the measured values for the fracture pressures. The goal is to generate an understanding of the effect that the various design factors have on the mean fracture pressure and to obtain guidelines for improving the inserts of microfluidic glass chips.
2 Materials and methods.
Chip design and fabrication - Test chips with dimensions of 10 mm length and 5 mm width,
containing different microchannel and insert layouts, were fabricated according to the process flow previously reported [17]. All the microchannels are dead-end on one side and run into an insert on the other side. The differences between the various test chips are: i) the glass type, ii) processing steps affecting the surface roughness of the insert channel, iii) the taper angle of the transition from the insert to the microchannel, iv) the size of the microchannel, and v) the geometry of the insert. Every design parameter ("factor" in the Design of Experiments, see below) has two variations, yielding a total of 32 different designs. Fig. 1 displays a schematic representation of the basic design variations, where designs 1 to 4 in the top row can be either in single or double geometry, with surfaces either rough (i.e. as resulting from the applied powder blasting process) or smoothened (by additional isotropic etching), and all variations are processed in either of the two glass types. An overview of all fabricated chips is given in Table S-1 in the Supporting Information.
Chips are fabricated either in Borofloat 33 (BF33, Schott Technical Glasses, Germany; glass type commonly used as a substrate for microfluidic chips) or in D263T (D263T, Schott Technical Glasses, Germany; glass type commonly used for microscope coverslips). The glass substrates have a thickness of 1.1 mm, therefore the final bonded stacks have a total thickness of 2.2 mm. Mechanical properties of the glass types are summarized in Table 1. The parameters of interest are whether the mean fracture pressure would differ significantly between glass types, and if the same design factors have the same effect on the mean fracture pressure for the different glass types.
Figure 1. An overview of geometries and chip designs. The first row displays a top-view of the chip layouts, the other rows
present cross-sectional views. In the second row, the dimensions for the microchannels, highlighted with orange color in the first row, are given. These channels are manufactured by HF-etching. The third and fourth row give the dimensions of the inserts, represented in row 1 by green and red color, respectively, where the third row corresponds to single-processed and
the fourth row to double-processed chips. The inserts have a bell shape as a result of the applied powder blasting process [29]. Note that for the wider microchannels in the last column, the taper in the insert is only present in the direction perpendicular to the plane of view, i.e. in the depth direction of the insert (see also main text and Fig. 2 below), because the microchannel and the insert have the same width.
Table 1. Properties of the tested glass types (source: Schott product info, www.schott.com).
Glass type / property BF33 D263T
Density (ρ) 2.2 g/cm3 2.51 g/cm3
Young’s Modulus (E) 64.0 GPa 72.9 GPa
Poisson’s ratio (μ) 0.2 0.208
A difference in roughness of the insert inner wall surface can be realized by following one of the following fabrication sequences:
1. 20 μm deep HF etching of the fluidic microchannel, followed by 380 μm deep powder blasting (also called: abrasive jet machining) of the insert. This gives inserts with relatively rough sidewalls (see below).
2. 380 μm deep powder blasting of the inserts, followed by 20 μm deep HF etching of the microfluidic channel. It is expected that with the latter process, the sidewalls of the insert channel will be smoothened by isotropic etching (as in our previous work [16,17]).
The topic of interest is whether these different sequences indeed induce a smooth or rough surface, and if the resulting difference in surface roughness affects the mean fracture pressure. Four main different process sequences are carried out, in which the order of powder blasting and HF-etching is reversed, as well as the insert is realized in one or in two substrates (single or double geometry). In case the insert is realized in two substrates which will be aligned and bonded together later (the double geometry), both substrates are powder blasted to a depth of 180 μm, followed by an HF step on both substrates to create 20 μm additional depth for the inserts on both substrates, to reach the same total insert depth as in the single geometry. As for the single geometry chips, the order of powderblasting/ etching may be reversed to vary the roughness of the surface.
The width of the mask-opening for the powder blasted insert is 500 μm. The depth of the inserts (after powder blasting and HF-etching) is 400 μm for all geometries, i.e. 400 μm per substrate in case of inserts with a single geometry cross-section, and 200 μm per substrate in case of inserts that have a double geometry cross-section. The wall angle of the powder blasted groove at the top-side of the substrate depends on the size of the blast-area (i.e. mask-opening) as well as the applied flux of Al2O3 particles and is approximately 75-82°, whereas the overall shape of the groove can be approximated by a Gaussian [29]. In a single geometry chip, a processed substrate is direct-bonded to an unprocessed substrate. In the case of the double geometries, both substrates are processed prior to aligned direct-bonding.
Fig. 2 gives a schematic representation of the transition (B) from the insert (A) to the microchannel (C). The length of the powder blasted insert is 4 mm (A), followed by a fluidic microchannel (C). In the case of a direct, "flat" transition (α = 90⁰), (B) equals 0 and in this case, the length of (C) is 4 mm. For the applied tapered transition with α = 15⁰, the transition length (B) is 1 mm and the length of the microchannel (C) is 3 mm. Note that for a tapered transition, due to the relatively constant powder blasting wall inclination caused by so-called blast lag [29], a masking opening with a taper (as is used here), will also give a taper in depth along the insert.
Figure 2. Top: Schematic top view of the two different designs for the transition from the insert to the microchannel. Left: no
(C) microchannel. Bottom: Schematic side view of the transition from insert to microchannel, including the capillary and the glue layer, for a double geometry chip.
Table 2. List of design factors accompanied by their abbreviations
Factor -1 1 Abbreviation (either, or)
Glass type D263T BF33 263, 33
Surface roughness PB PB+HF P, H
Transition Flat Tapered F, T
Size of the microchannel Narrow Wide N, W
Geometry Single Double S, D
Throughout this paper, chips are identified by means of an abbreviation corresponding to the design factors implemented on the insert, see Table 2. For example, a chip made out of BF33 containing a powder blasted plus HF-etching-smoothened surface (H), flat transition (F), a wide microchannel (W) and a double geometry (D) would be abbreviated by 33HFWD.
Pressure testing setup - Polyamide coated fused silica capillaries with dimensions 250 μm and 360 μm
(Polymicro Technologies, USA), inner and outer diameter respectively, were glued into the side inserts of the chips with an epoxy glue (Araldite Rapid™, Ciba-Geigy, Switzerland) and allowed to cure at room temperature for at least 48 hours. The glue meniscus was positioned approximately 2 mm from the entrance of the insert. The test chip itself was placed in a custom-built metal clamp (see Supporting Information for details), in order to strengthen the insert part of the chip.
During each experimental run, an HPLC syringe pump (model 100DM; Teledyne Isco Inc., Lincoln, USA) containing demineralized water, pumped the liquid through a fused silica capillary towards the test chip by aid of a pump controller (series D; Teledyne Isco Inc., Lincoln, USA) and applying stainless steel fittings (F-140 & A-318, Upchurch Scientific, USA) to connect the HPLC pump to the valves. A PEEK nut (C360NFPK, VICI Valco, USA) and reducing union (C360RU1PK6, VICI Valco, USA) were used to connect the valve to the capillary. The flow rate was set at a constant rate of 200 μL/min to build up the pressure in the glass test chips at a constant rate. At the same time, the information on the HPLC readout pressure was logged by a LabVIEW-based program. By means of visual inspection, it was checked whether failure was due to chip fracture and not due to failure of tubing or connectors. In all cases, pressure was released from the system through the waste line before a new chip was connected. All experiments were carried out at room temperature (ca. 22oC).
Design of experiments and statistical analysis - The experiments are performed by means of a Design
of Experiment (DoE) approach. Factors are varied over a range of, initially, two values which are either high, indicated by "1", or low, "-1". By allowing interaction between the settings, a response is set up in which the variation of one factor, or interaction between factors, can be tested against the variation of all factors, indicating if the factor has a significant effect on the output. This is done by means of an analysis of variance (ANOVA). It is assumed that the measured variance is a result of changing inputs as well as of non-controllable factors.
The results from the ANOVA are extracted by means of an F-test, where the variance of one factor is tested to the total variance of the experiment yielding a test score represented by an F-value. The null hypothesis is that none of the factors or interactions have a significant effect on the total variance, which is tested at a significance level of α= 0.05. The P-value indicates the significance level at which the null hypothesis is failed to be rejected (i.e. the hypothesis is accepted). Important for the ANOVA is the assumption of normality, for which will be tested before statistical inference is commenced, by means of an Anderson-Darling test. Whenever the assumption of normality is rejected, a Box-Cox transformation is performed, and the resulting data set is again tested for normality. Departure from normality may result in false-positive or false negative conclusions. The assumption of normality is, in this case, defined as the underlying dataset being normally distributed.
3 Results and discussion
Insert roughness with and without isotropic etching treatment - Table 3 contains the Ra roughness values, measured with a Veeco Dektak 8 surface profiler, of BF33 and D263T substrates before and
after HF-etching. The reported values are the mean values of 5 scans over 1 mm at various locations of
the treated surfaces. The as-blasted roughness of BF33 and D263T is 2.2-2.3 μm, which is identical to previously reported data for Pyrex [30] and BF33 glass [31], obtained with a similar powder blasting system. Upon HF-etching (25% HF for BF33, 10% for D263T; giving an etch rate of ~1 μm/min for both materials) the Ra value of BF33 increases slightly, whereas that of D263T increases significantly (P-value <0.001, N=5).
Table 3. Mean roughness parameter Ra* of the different glass types after powder blasting (PB) and HF etching (PB+HF).
roughness/glass type BF33 in μm (Cv**) D263T in μm (Cv**)
Ra after PB 2.23 ± 0.17 (0.076) 2.06 ± 0.22 (0.107) Ra after PB+HF 2.31 ± 0.17 (0.074) 3.53 ± 0.36 (0.102)
* Ra is the arithmetic average of the absolute values of the profile height deviations from the mean line, recorded within the
evaluation length.
** Between brackets is the Cv value, coefficient of variance, given as the standard deviation divided by Ra
The surface profile scans after powder-blasting have a spiky appearance, whereas after HF etching the high asperities are leveled off. Powder blasting erodes the surface by the generation of both lateral and radial cracks due to the impact of the powder particles. The roughness measured after powder blasting mainly originates from the lateral cracks, which lead to chipping off microscopically small parts from the glass surface [30]. The radial (or median, i.e. normal to the particle impact) microcracks cannot be detected by the surface profiler with a stylus-tip radius of 2.5 μm, even more so since it has been reported that upon unloading (i.e. when particle impact has ceased), the radial crack closes [32], so that a crack opening may not even be apparent at the surface where these cracks end. On the other hand, the radial crack length can easily be up to 100 μm or more [33], meaning that even if the isotropic nature of HF etching smoothens the surface asperities, etching into these cracks and opening them up may lead to enhanced surface roughness [30]. Most likely, this is what has happened in our smoothening process. It is therefore questionable whether the removal of 20 μm from the powder blasted glass surface, as was done here, will alleviate the bulk damage caused by the powder blasting process.
The powder blasted structures for both types of glass have a milky appearance. They become transparent after HF treatment, in particular in the case of BF33. Fig. 3 shows top-view SEM-images of powder blasted D263T and BF33 surfaces before and after etching. Whereas the morphological appearance of powder blasted BF33 and D263T is nearly the same, see Figs. 3(a) and (c), significantly more HF etched pits are visible in D263T than in BF33, see Figs. 3(b) and (d). Also, the shape of the pits is different: BF33 shows hemispherical features, compared to elliptical ditches on D263T.
Figure 3. Typical top-view SEM-images of powder blasted BF33 (a, b) and D263T (c, d) surfaces before (a, c) and after to
HF-etching (b, d). All images have the same scale (scale bars are 100 µm).
Influence of a metal clamp - An experiment was conducted in which chips of two BF33 designs (i.e.
33HFND and 33HTWD) were tested with and without a metal clamp (see Supporting Information for clamp details). Two factors, viz. chip design and application of metal clamp, were tested in two settings (two designs, clamp or no-clamp), with four repeats yielding a total sample size of 16 samples. The dataset was transformed and tested for normality, which was failed to be rejected at p=0.571. From the statistical results in Table S-4 (a detailed discussion of the statistical inference is given in the Supporting Information), it follows that the clamp has a significant effect on the mean fracture pressure of BF33 glass chips of design 33HFND and 33HTWD, with little influence of the chip design. The mean fracture pressures with and without clamp (based on results for 8 chips for both cases) are 234±58 bar and 184±20 bar, respectively. Note that the standard deviation for the clamped chips is rather large, which is caused by one outlier in the data set with an exceptionally high fracture pressure. It was chosen to keep the outlier in the dataset because its origin was unclear.
With an optical microscope it was observed that all chips tested without the clamp failed due to bond release, whereas chips tested with the clamp failed due to cracking of the glass. Therefore, all later experiments are carried out with a metal clamp around the insert of the glass test chips. In this way, the fracture pressure due to cracking of the glass is measured, and not due to bond release at the insert or fracture or delamination of the glue bond (the latter topics have been investigated in our previous work [17]). Fig. 4 shows an example of a chip for which the failure mechanism was by bond release.
Figure 4. A 33HTWD chip after pressure test and consecutive bond release failure. The periodic discoloring across the
delaminated chip is due to Newton rings which are believed to be caused by the combined effect of the deformation of either of the glass slides, the angle of the incident light, and a thin water film between the two glass slides. This type of observation is a typical characteristic of bond release.
25-factorial design with four repeats for two glass types - Measurements on 175 chips have been used
to analyze the influence of the different chip design parameters on the fracture pressure, based on a full factorial design with 5 factors on 2 levels and 4 repeats. 128 out of 175 values for fracture pressure were selected for further analysis, all being from chips that failed by glass cracking. The selection was done by visual inspection. More examples of the various types of chip failure are shown in the supporting information. The remaining 47 chips failed either due to bond release, incorrect application of the glue which prevented conducting an appropriate pressure test, or defects caused by underetching into the area that was intentionally masked (this mostly occurred for D263T chips). The factors used during the factorial design are displayed in Table 4, and varied over two settings (-1, 1). The null hypothesis is that none of the factors or interactions between the factors influence the mean fracture pressure of the glass chips.
Table 4. Design factors and settings for a 25 factorial design experiment.
Factor -1 1
roughness PB PB+HF
transition Flat Tapered
size μ-channel Narrow Wide
geometry Single Double
glass type D263T BF33
Fig. 5 shows the 95% confidence intervals for the fracture pressures of chips manufactured from BF33 or D263T glass. Every interval is based on 4 measurements conducted for every setting. The data are analyzed prior to applying an ANOVA to see if the assumptions of normality and homoscedasticity hold. The P-values for the Anderson-Darling and multiple comparisons test are 0.013 and 0.036 respectively (see Supporting Information for more details). This means that the assumption of normality does not hold and within-group data are likely to be drawn from different underlying distributions. This might increase the probability on type I errors when considering the F-statistic P-values. When P-values are close to the significance level of α=0.05, care must be taken with drawing conclusions. The factors that were found to have a significant effect on the mean fracture pressure are displayed in Table 5.
Figure 5. 95% Confidence intervals for fracture pressures of BF33 and D263T chips, plotted versus the different design
Table 5. Factors that have a significant effect on the mean fracture pressure with their description, F-value, P-value, and
coefficient respectively. The coefficient is displayed mainly to show its positive or negative effect in a linear model.
FactorI D
Factor or interaction F-value P-value Coefficient (bar)
E glass type 50.269 2.3E-10 11.7
AE roughness• glass type 31.584 1.8E-7 9.2
AD roughness• geometry 17.697 5.8E-05 -6.9
ABC roughness• transition• microchannel 5.918 1.7E-02 4
The statistical analysis (see Supporting Information) shows that the residuals fail to pass the Anderson-Darling test for normality (P-value <0.005). Furthermore, the histogram on the residuals displays a non-normal distribution, which implies that a linear regression model does not estimate the fracture behavior well enough, based on the input parameters used. Departure from normality is mostly observed in the BF33 data set, furthermore, whereas the BF33 data contain mostly high valued outliers, the data set for D263T contains mostly low valued outliers (Fig. 6). From these results it is concluded that it is meaningful to set up a factorial design with 4 factors on 2 levels and 4 repeats, eliminating the glass type from the analysis, and check for consistencies among the results for BF33 and D263T separately.
Figure 6. Boxplot of the fracture pressures for BF33 and D263T. Outliers are represented by means of an asterisk (*).
24-factorial design experiment on BF33 glass chips- The same experimental data that were used for
the 25-design are implemented in a 24-design and analyzed for the same factors, except for the glass type. This means that the results for a single glass type are analyzed considering only the design changes. In total 64 chips were analyzed, which all fractured because of glass cracking. The factors and interactions that were found to have a significant effect on the mean fracture pressure are displayed in Table 6 with their coefficient, F-values and P-values (for the complete statistical analysis, see Supporting Information). In the case of BF33, relative to the 25-factorial design the number of main effects and interactions that exhibit a significant effect on the fracture pressure have increased. Furthermore, the 3 factors found significant in the 25 analysis (except for the glass type) are also found significant here: Roughness is the most significant factor, where its positive coefficient indicates that an HF smoothened surface has a higher mean fracture pressure. Note that this is different from the 25 analysis, where the HF treatment had an overall negative coefficient, i.e. a reducing effect on the mean fracture pressure. Note also the outcome in Table 6 that a high level for the roughness or for the geometry, each on their own, gives a positive coefficient, i.e. increases the mean fracture pressure, whereas their interaction (the combined effect of roughness and geometry) leads to a decrease in fracture pressure. The implication of this finding will be discussed in more detail below.
Table 6. 64 destructive pressure test on BF33 chips have revealed that the factors and interactions considered in this table
have been found to have a significant influence on the mean fracture pressures of BF33 glass chips. The factors have been found significant at a level of α=0.05 with their description, F-value, P-value, and coefficient, respectively. The coefficient column contains a transformed value with unit 1/bar and is displayed here mainly to show whether it is positive or negative.
Factor or interaction F-value P-value Coefficient
(transformed)* roughness 34.26 0.000 0.00526 roughness/geometry 22.82 0.000 -0.00429 transition/microchannel 6.78 0.012 0.000054 roughness/transition/microchannel 6.08 0.017 0.000221 microchannel 5.97 0.018 -0.000062 geometry 5.48 0.023 0.0000210
* see Supporting Information for details
Table 7. Fracture pressures (in bar) for BF33 chips of single or double geometry, with rough or smoothened inserts. No
discrimination is made for microchannel size or tapering so that each mean value and standard deviation originates from 16 measurement values.
Roughness / Geometry Single Double
PB 159±19 201±28
PB+HF 233±42 214±51
Table 7, which summarizes the data for the HF treatment on powder blasted inserts in BF33, demonstrates that smoothened surfaces outperform their rough counterparts. The result that the double geometry has a higher fracture pressure than the single geometry may partially be explained by the fact that the powder blasted channel for the chosen dimensions has a bell-shaped (Gaussian) cross-section [29], which leaves an indent at the joining edge of the two bonded substrates, whereas the double geometry approaches more closely a circular, stress-relieved shape. However, the results in Table 7 show that the effect of the HF smoothening treatment is much more effective on the single than on the double geometry. Furthermore, the data in Table 3 show that HF etching increases surface roughness, which above was attributed to the opening up (and incomplete removal) of radial cracks. These two findings indicate that an additional factor might be responsible for the difference in mechanical stability of single and double geometry: the double geometry has more surface defects, caused by etching into radial cracks, because it has a powder blasted surface on both sides of the insert channel, and thus the probability of crack propagation is higher for this geometry.
From Fig. 5 it can be derived that a tapered transition reduces the variance in the fracture pressures, but also leads to a decrease in the mean fracture pressures when compared to the flat counterpart. To evaluate this in more detail, cracking initiation was located by means of optical microscopy and scored binomially for the occurrence (1) or absence (0) at certain locations. Several images of chips containing different characteristic crack patterns are given in the Supporting Information, a representative example is given in Fig. 7.
Figure 7. Image of a 33PTNS chip after a pressure test and failure. The crack path seems to originate from the glue
meniscus and evolves through one of the glass slides to the outer interface. For this chip it is observed that the crack misses both the tapering and the microchannel so that it scores a (1) for the meniscus and a (0) for transition.
The regions of interest for this scoring test are: the transition from insert to the microchannel, the glue meniscus and the microchannel. The chips were also scored binomially for containing a tapered transition or not. Afterwards, all binomial data vectors were compared by means of Pearson correlation.
Table 8. Pearson correlation coefficient values when correlating the tapered transition to the location of crack initiation.
Factor/Region R²
Tapered/meniscus 0.331
Tapered/transition -0.509
In Table 8, it is seen that a tapered side insert shows a positive correlation with fracturing at the glue meniscus and a negative correlation with fracturing at the transition towards the microchannel. Although the coefficients are not convincing, their direction is opposite, indicating that applying a tapered transition not only affects the mean fracture pressure and its variance, but also the location of the cracking initiation, showing an affinity to fracturing at the meniscus. The difficulty during the optical analysis is that for many of the chips the crack runs from the glue meniscus through the end of the tapered transition where the taper joins the microchannel (or vice versa) so that both meniscus and transition are scored with a (1), therewith blurring the test statistics. The only way to obtain the correct direction of the crack propagation would be to do a microscopic in situ study during the failure of the chip, which most likely requires high-speed imaging. This could be a topic for future study.
24-factorial design experiment on D263T glass chips - Exactly the same procedure was followed for
the analysis of 64 chips made from D263T glass. Table 9 shows that, compared to the 25-factorial design and the analysis of the BF33 glass chips, fewer factors are found significant for the fracture pressure of D263T chips. F- and P-values are lower, indicating more variance during the measurements. Remarkable is also the negative coefficient for factor A, which implies that a powder blasted surface has a higher fracture pressure than an HF smoothened surface. This clearly differs from the BF33 case, with a positive effect of roughness on mean fracture pressure. Similar to BF33, the interaction factor geometry-roughness has a negative effect on mean fracture pressure. Finally, a comparison of Figs. 3 and 5 demonstrates that the data for D263T show higher variance.
Table 9. Results on significant factors in 24 factorial design on D263T with a significance level of α= 0.05.
Factor or interaction
F-value P-value Coefficient (transformed)*
roughness 5.08 0.028 -3536
roughness• geometry 4.84 0.032 -3452
* see Supporting Information for details
Comparison of the BF33 and D263T results - Fig. 6 indicates that the choice of glass type is
significant for the mean fracture pressure of glass chips, with BF33 chips fracturing at higher pressures than D263T chips. The different design factors that have been studied also seem to have a different impact on the mean fracture pressure, where it is found that the results for D263T carry weaker conclusions (lower p-values) than those for BF33, pointing out that different, less obvious, intrinsic properties of the glasses are playing a part. A lead to an explanation might be found in the experiments in which the surface roughness of the side insert was changed by HF etching. Above it was already discussed that the effect of the HF treatment is more effective for the single insert geometry, and in fact does not give the expected smoothening but rather an increase in surface roughness (Table 3), which is attributed to the widening of the radial cracks by HF etching [30]. As the summarized data in Fig. 8 show, the effect of the HF treatment is opposite for the two glass types: the fracture pressure becomes higher after HF treatment for BF33, and lower for D263T, whereas Table 3 again shows that the roughness increase by the treatment is significantly higher for the D263T glass chips.
It is thus concluded that the mean chip fracture pressure has a correlation with a glass property that underlies the effect that HF etching has on defects generated by powder blasting. Note that the Cv values in Table 3 for D263T are higher, both before and after HF treatment. This can be interpreted as D263T glass having a surface with higher aspect ratio asperities than BF33. The valleys between such asperities might be considered points for crack initiation during the pressure fracturing process, in the case of powder blasted surface. This might partially explain the trends indicated in Fig. 8.
Figure 8. Boxplots on fracture pressure of different materials under similar surface treatment conditions, not considering the
differences in chip geometry. A Boxplot is used, as the Anderson-Darling normality test had shown that the untransformed datasets distributions where non-normal. Symbols indicate outliers (compare Fig. 5).
A remaining question is how HF etching affects the surface of the different glass types if there are no asperities, or hidden microcracks or other damage caused by powder blasting, present. To verify this, test channels were etched in clean, unprocessed substrates of BF33 and D263T glass having the same 20 μm channel depths and applying an etching rate of 1 μm/min in both cases (which requires different etching media, see Materials and Methods). The substrates were rinsed and dried. The roughness was measured before and after etching by means of Atomic Force Microscopy, giving the data in Table 10. From these measurements, it is derived that, while etching shows an increase in roughness for both materials the scale on which the effect takes place is much smaller than the roughness witnessed after powder blasting, or the roughness that was achieved after etching a powder blasted channel. Furthermore, no significant difference exists between the two materials. From this, it is concluded that the HF etching of the glass in itself does not cause surface damage which might affect fracturing. In this respect, it is worth mentioning that in our previous work [17] it was found that for insert channels fabricated by only HF etching a mean fracture pressure of at least 690 bar (the limit of the applied pump) was achieved (for an insert width of 120 µm; greater HF etching depths are impractical due to mask layer limitations).
Table 10. Roughness data on BF33 and D263T before and after etching in HF at similar rates to similar depths.
Description of glass surface Ra-values (nm)
BF33 untreated 0.365
BF33 after 20 min in 25% HF (1μm/min) 0.878
D263T untreated 0.417
D263T after 20 min in 10% HF (1μm/min) 0.912
During powder blasting, two types of cracks are observed, lateral cracks and radial/median cracks. It is believed that the lateral cracks, parallel to the surface of the substrate, induce material removal, while radial/median cracks cause a decreased strength of the material. Above it has already been discussed that etching 20 μm away from the glass surface does not remove all bulk damage caused by the powder blasting process, particularly not the radial cracks which can run as deep as 100 μm [33]. Erosion rates and microcrack formation depend on the mechanical characteristics of the substrate onto which the powder blasting is applied, more specifically, the fracture mechanics relations for equilibrium crack growth involve the ratio of hardness-to-modulus ratio, H/E [34]. It was reported that as the ratio H/E of a material is lower, crack growth with a lateral component becomes dominant over subsurface median crack growth. For the two investigated glass types the hardness to modulus ratio does not differ much, 0.073 versus 0.080 for BF33 and D263T, respectively (calculation based on Knoop hardness, Table 1), although the slightly higher ratio would indicate a higher probability for more (deeper) radial/median crack growth in D263T.
4 Conclusions.
A factorial design experiment on the mean fracture pressure for microfluidic chips of two glass types, BF33 and D263T, was carried out, testing a large number of chips with different geometries. A first finding is that a metal clamp over the insert part of the chip reduces the probability of chip failure by bond release or delamination of the glue layer. The surface roughness of the side inserts for connecting glass capillaries to the chip turns out to be a major factor in the stability of the chips, with some clear differences between the two glass types. The mean fracture pressure between the two tested glass types BF33 and D263T, taking in account all the designs are 192 ±25 and 159 ±25 bar, respectively. For BF33, hydrofluoric acid (HF) etching of the powder blasted (abrasive jet machined) chip insert increased the mean fracture pressure with 43 ±9 bar, whilst for D263T a decrease of -22 ±8 bar resulted. Roughness is related to the damage caused by the powder blasting process applied to
manufacture the side inserts, and subsequent HF treatment cannot remove all this damage. It is concluded that chip fracturing has a correlation with a glass property that underlies the effect that HF etching has on the defects generated by powder blasting, intrinsic mechanical properties cannot explain the higher fracture pressures of BF33 chips compared to D263T. The effects of the roughness factor in most experiments is so high, as well as the variance in the measurements that it causes, that it is statistically less evident in this study that a tapered transition from the insert to the microchannel improves the pressure stability of the chips. In previous work, in which HF etched inserts (although less deep) were tested, the transition had a very significant effect and aided in raising the fracture pressure of the chips to ~ 690 bar [17]. The current experiments show that fracture in case of a tapering does not occur at the point where the insert and the microchannel join but at the location of the meniscus of the glue-front in the insert.
Acknowledgements
This work was partially funded by the Dutch network for Nanotechnology, NanoNext NL, in the subprogram 10C, "Microdevices for chemical processing".
References
[1] D.C. Hendershot, Process minimization: Making plants safer, Chem. Eng. Progr. 96 (2000) 35-40. [2] V. Hessel, Novel process windows - Gate to maximizing process intensification via flow chemistry, Chem. Eng. Technol. 32 (2009) 1655-1681
[3] A. Adamo, R.L. Beingessner, M. Behnam, J. Chen, T.F. Jamison, K.F. Jensen, J.-C.M. Monbaliu, A.S. Myerson, E.M. Revalor, D.R. Snead, T. Stelzer, N. Weeranoppanant, S. Yee Wong, P. Zhang, On-demand continuous-flow production of pharmaceuticals in a compact, reconfigurable system, Science 352 (2016) 61-67.
[4] H.R. Sahoo, J.G. Kralj, K.F. Jensen, Multistep continuous-flow microchemical synthesis involving multiple reactions and separations, Angew. Chem Int. Ed. 46(2007) 5704-5708.
[5] D.L. Browne, S. Wright, B.J. Deadman, S. Dunnage, I.R. Baxendale, R.M. Turner, S.V. Ley, Continuous flow reaction monitoring using an on-line miniature mass spectrometer, Rapid Commun. Mass Spectr. 26 (2012) 1999-2010.
[6] R. Blanch-Ojea, R.M. Tiggelaar, J. Pallares, F.X. Grau, J.G.E. Gardeniers, Flow of CO2–ethanol and of CO2–methanol in a non-adiabatic microfluidic T-junction at high pressures, Microfluid. Nanofluid. 12 (2012) 927-940.
[7] A. Martin, S. Camy, J. Aubin, Hydrodynamics of CO2-ethanol flow in a microchannel under elevated pressure, Chem. Eng. Sci. 178 (2018) 297-311.
[8] F. Benito-Lopez, R.M. Tiggelaar, K. Salbut, J. Huskens, R.J.M. Egberink, D.N. Reinhoudt, J.G.E. Gardeniers, W. Verboom, Substantial rate enhancements of the esterification reaction of phthalic anhydride with methanol at high pressure and using supercritical CO2 as a co-solvent in a glass microreactor, Lab Chip 7 (2007) 1345-1351.
[9] S. Marre, C. Aymonier, P. Subra, E. Mignard, Dripping to jetting transitions observed from supercritical fluid in liquid microcoflows, Appl. Phys. Lett. 95 (2009) 134105.
[10] F. Trachsel, B. Tidona, S. Desportes, P.R. Von Rohr, Solid catalyzed hydrogenation in a Si/glass microreactor using supercritical CO2 as the reaction solvent, J. Supercrit. Fluids 48 (2009) 146-153. [11] S. Marre, Y. Roig, C. Aymonier, Supercritical microfluidics: opportunities inflow-through chemistry and materials science, J. Supercrit. Fl. 66 (2012) 251-264.
[12] A.K. Goodwin, G.L. Rorrer, Conversion of Glucose to Hydrogen-Rich Gas by Supercritical Water in a Microchannel Reactor, Ind. Eng. Chem. Res. 47 (2008) 4106-4114.
[13] M. Andersson, K. Hjort, L. Klintberg, Fracture strength of glass chips for high-pressure microfluidics. J Micromech. Microeng. 26 (2016) 095009
[14] Y. Huang, A.S.S. Vasan, R. Doraiswami, M. Osterman, M. Pecht, MEMS reliability review. Device and materials reliability, IEEE Trans. Device Mater. Rel. 12 (2012) 482-493.
[15] M.T. Blom, N.R. Tas, G. Pandraud, E. Chmela, J.G.E. Gardeniers, R. Tijssen, M. Elwenspoek, A. van den Berg, Failure mechanisms of pressurized micro channels, model and experiments, J. Micro Electromech. Syst. 10 (2001) 158-164.
[16] R.E. Oosterbroek, D.C. Hermes, M. Kakuta, F.Benito-Lopez, W. Verboom, J.G.E. Gardeniers, A. van den Berg, Fabrication and mechanical testing of glass chips for high pressure synthetic or analytical chemistry, Microsyst. Technol. 12 (2006) 450-454.
[17] R.M. Tiggelaar, F. Benito-López, D.C. Hermes, H. Rathgen, R.J.M. Egberink, F.G. Mugele, D.N. Reinhoudt, A. van den Berg, W. Verboom, H.J.G.E. Gardeniers, Fabrication, mechanical testing and application of high-pressure glass microreactor chips, Chem. Eng. J. 131 (2007) 163-170.
[18] G. Desmet, S. Eeltink, Fundamentals for LC miniaturization, Anal. Chem. 85 (2013) 543-556. [19] X. Yuan, R.D. Oleschuk, Advances in microchip liquid chromatography, Anal. Chem. 90 (2018) 283-301.
[20] Y. Temiz, R.D. Lovchik, G.V. Kaigala, E. Delamarche, Lab-on-a-chip devices: How to close and plug the lab? Microelectr. Eng. 132 (2015) 156-175.
[21] W. Wang, C. Gu, K.B. Lynch, J.J. Lu, Z. Zhang, Q. Pu, S. Liu, High-pressure open-channel on-chip electro osmotic pump for nanoflow high performance liquid chromatography, Anal. Chem. 86 (2014) 1958-1964.
[22] M.T. Blom, E. Chmela, J.G.E. Gardeniers, J.W. Berenschot, M. Elwenspoek, R. Tijssen, A. van den Berg, Local anodic bonding of Kovar to Pyrex aimed at high-pressure, solvent-resistant microfluidic connections, J. Micromech. Microeng. 11 (2001) 382-385.
[23] A. Mehra, X. Zhang, A.A. Ayón, I.A. Waitz, M.A. Schmidt, C.M. Spadaccini, A six-wafer combustion system for a silicon micro gas turbine engine, J. Microelectromech. Syst. 9 (2000) 517-527.
[24] Y. Shintani, K. Hirako, M. Motokawa, Y. Takano, M. Furuno, H.Minakuchi, M. Ueda, Polydimethylsiloxane connection for quartz microchips in a high-pressure system, Anal. Sci. 20 (2004) 1721-1723.
[25] A.G. Chambers, J.S. Mellors, W.H. Henley, J.M. Ramsey, Monolithic integration of two-dimensional liquid chromatography-capillary electrophoresis and electrospray ionization on a microfluidic device, Anal. Chem. 83 (2012) 842–849.
[26] C. Lotter, J.J. Heiland, V. Stein, M. Klimkait, M. Queisser, D. Belder, Evaluation of pressure stable chip-to-tube fittings enabling high-speed chip-HPLC with mass spectrometric detection, Anal Chem. 2016, 88, 7481-7486.
[27] M. Callewaert, J. Op De Beeck, K. Maeno, S. Sukas, H. Thienpont, H. Ottevaere, H. Gardeniers, G. Desmet, W. De Malsche, Integration of uniform porous shell layers in very long pillar array columns using electrochemical anodization for liquid chromatography, Analyst 139 (2014) 618-625. [28] A.A. Griffith, The phenomena of rupture and flow in solids, Phil. Trans. Roy. Soc. London. Series A, 221 (1921) 163-198.
[29] H. Wensink, M.C. Elwenspoek, Reduction of sidewall inclination and blast lag of powder blasted channels, Sens. Act. A 102 (2002) 157-164.
[30] H. Wensink, S. Schlautmann, M.H. Goedbloed, M.C. Elwenspoek, Fine tuning the roughness of powder blasted surfaces, J. Micromech. Microeng. 12 (2002) 616-620.
[31] Q.-S. Pu, R. Luttge, J.G.E. Gardeniers, A. van den Berg, Comparison of capillary zone electrophoresis performance of powder blasted and hydrogen fluoride etched microchannels in glass, Electrophoresis 24 (2003) 162-171.
[32] M. Achtsnick, P.F. Geelhoed, A.M. Hoogstrate, B. Karpuschewski, Modelling and evaluation of the micro abrasive blasting process, Wear 259 (2005) 84-94.
[33] A.G. Evans, T.R. Wilshaw, Quasi-static solid particle damage in brittle solids—I. Observations analysis and implications, Acta Mater. 24 (1976) 939-956.
[34] B.R. Lawn, A.G. Evans, D.B. Marshall, Elastic/plastic indentation damage in Ceramics: The median/radial crack system, J. Am. Ceram. Soc. 63 (1980) 574-581.
