Microfluidics and Nanofluidics

, Volume 3, Issue 1, pp 1–11

Evaluation of micromilled metal mold masters for the replication of microchip electrophoresis devices


  • Mateusz L. Hupert
    • Center for Bio-Modular Multi-Scale SystemsLouisiana State University
  • W. Jason Guy
    • Center for Bio-Modular Multi-Scale SystemsLouisiana State University
  • Shawn D. Llopis
    • Center for Bio-Modular Multi-Scale SystemsLouisiana State University
    • Department of ChemistryLouisiana State University
  • Hamed Shadpour
    • Center for Bio-Modular Multi-Scale SystemsLouisiana State University
    • Department of ChemistryLouisiana State University
  • Sudheer Rani
    • Center for Bio-Modular Multi-Scale SystemsLouisiana State University
    • Department of Mechanical EngineeringLouisiana State University
  • Dimitris E. Nikitopoulos
    • Center for Bio-Modular Multi-Scale SystemsLouisiana State University
    • Department of Mechanical EngineeringLouisiana State University
    • Center for Bio-Modular Multi-Scale SystemsLouisiana State University
    • Department of ChemistryLouisiana State University
    • Department of Mechanical EngineeringLouisiana State University
Research Paper

DOI: 10.1007/s10404-006-0091-x

Cite this article as:
Hupert, M.L., Guy, W.J., Llopis, S.D. et al. Microfluid Nanofluid (2007) 3: 1. doi:10.1007/s10404-006-0091-x


High-precision micromilling was assessed as a tool for the rapid fabrication of mold masters for replicating microchip devices in thermoplastics. As an example, microchip electrophoresis devices were hot embossed in poly(methylmethacrylate) (PMMA) from brass masters fabricated via micromilling. Specifically, sidewall roughness and milling topology limitations were investigated. Numerical simulations were performed to determine the effects of additional volumes present on injection plugs (i.e., shape, size, concentration profiles) due to curvature of the corners produced by micromilling. Elongation of the plug was not dramatic (< 20%) for injection crosses with radii of curvatures to channel width ratios less than 0.5. Use of stronger pinching potentials, as compared to sharp-corner injectors, were necessary in order to obtain short sample plugs. The sidewalls of the polymer microstructures were characterized by a maximum average roughness of 115 nm and mean peak height of 290 nm. Sidewall roughness had insignificant effects on the bulk EOF as it was statistically the same for PMMA microchannels with different aspect ratios compared to LiGA-prepared devices with a value of ca. 3.7 × 10−4 cm2/(V s). PMMA microchip electrophoresis devices were used for the separation of pUC19 Sau3AI double-stranded DNA. The plate numbers achieved in the micromilled-based chips exceeded 1 million/m and were comparable to the plate numbers obtained for the LiGA-prepared devices of similar geometry.


MicromillingHot-embossingMicrochip electrophoresisPolymer microfluidics

1 Introduction

Miniaturization of laboratory processes to form total analysis systems (μTAS) has gained a great deal of attention in research laboratories over recent years. In principle, μTAS offers multiple advantages over bench-top instruments, such as increased speed of analysis, high throughput, multiplexing capabilities, high levels of system integration, portability and significantly lower cost of operation due to reduced amounts of samples, reagents, and solvents used in the assay. All of these features are of great interest in the fields of genetic analyses, clinical testing, drug discovery, food control, and environmental monitoring. In addition, several steps of multi-step assays can be integrated into monolithic μTAS devices to provide high levels of automation and at the same time, minimize contamination due to the closed architecture of these devices (Anderson et al. 2000; Liu et al. 2002, 2004).

It is well recognized that in order to make microfluidics more widely available, devices need to be produced in high volumes and at low unit costs (Becker and Gartner 2000; Boone et al. 2002). Micromanufacturing techniques using polymer substrates fulfill these requirements as polymers are generally inexpensive and a variety of low-cost fabrication techniques are available for both rapid prototyping and mass production of finished devices (Becker and Gartner 2000; Soper et al. 2000; Becker and Locascio 2002). Significant cost advantages for mass-producing polymer microdevices is offered by replication techniques, which use a microfabricated master to produce inverted structures in the polymer material. Since one master can be used multiple times, the cost of a single unit can be significantly reduced. Replication techniques commonly used include casting, hot-embossing, or injection molding (Becker and Gartner 2000).

The most elaborate and often expensive step in manufacturing microstructures through replication processes is fabrication of the mold master, the quality of which determines the quality of the device. There are few materials and processes that are being used for mold master fabrication, selection of which depends on the type of replication process, physical dimensions of the microstructures, and life expectancy of the mold master. For example, photolithographically patterned photoresists are commonly used for casting poly(dimethylsiloxane) (PDMS) devices making the fabrication process very inexpensive and easy to perform in any laboratory (Duffy et al. 1998; Vilkner et al. 2004). Wet-etched silicon is another common material for low-volume imprinting of low aspect ratio microstructures (Martynova et al. 1997; Xu et al. 2000). Both photoresist and silicon are not practical, however, for high pressure, high temperature molding processes, such as hot-embossing or injection molding, due to their low thermal and mechanical strength (Esch et al. 2003). An attractive material for replication processes are metals, as they offer both high thermal and mechanical strength and high thermal conductivity required for fast heating and cooling cycles during replication.

Various techniques for microfabricating metal mold masters have been established over recent years (Becker and Gartner 2000; Soper et al. 2000; Boone et al. 2002). The majority of these are based on the combination of standard lithographic processes (e.g., photoresist processing, dry and wet etching) to define the microstructures and metal electroplating (typically nickel) as the last step to produce a mold master. These methods involve multiple fabrication steps, are rather expensive and time consuming. Much simpler methods of metal mold master fabrication include conventional machining techniques, such as electro-discharge machining and high-precision micromilling (Ehrfeld et al. 1996; Guber et al. 2004). These methods do not involve lithographic steps and thus, do not require clean room environments making the fabrication of exquisite structures more accessible to researchers who do not have access to extensive lithography-based equipment.

High-precision micromilling offers some distinct advantages over other commonly used techniques for preparing mold masters. For example, only three fabrication steps are required; design, CNC milling, and finishing as compared to ∼ 15 steps required for X-ray LiGA or ∼ 10 steps for UV lithography-based techniques (Soper et al. 2000; Boone et al. 2002). Although micromilling can not achieve the fine resolution or minimum feature size of most lithographic techniques, it is well suited for many microfluidic applications, which usually require structures in the range of 10–500 μm with aspect ratios and inter-structure spacings that are easily obtainable using micromilling. In addition, micromilling offers the potential of fabricating multilevel structures during the same milling cycle at minimal additional cost as compared to repeating nearly all of the fabrication steps for each additional level when using lithographic techniques. Micromilling allows for a wide selection of materials to be used for the mold master fabrication as well as the ability to choose a non-standard mold insert size or shape. In comparison, lithography-based techniques are limited to metals that can be easily electroplated, thus excluding, for example, stainless steel (commonly used material for macro-molding), and to mold sizes which are usually limited by standard lithographic processing equipment (i.e., 4 or 6 in. wafer sizes).

Few reports have discussed the use of micromilling as a technique for metal microstructure fabrication. Friedrich et al. (1998) used precision micromilling with a custom made 22 μm diameter milling bit for the direct fabrication of deep X-ray lithography masks. The authors were able to produce the gold mask absorber features with a minimum width of 10 μm at an accuracy of 0.65 μm. Schaller and coworkers used home-made ground hard metal end mills (< 50 μm diameter) to cut microgrooves into brass and stainless steel (Schaller et al. 1999). When cutting stainless steel they observed tool wear after only a 100 mm length cut, whereas brass showed no significant tool wear. Madou et al. (2001) manufactured stainless steel mold inserts for molding CD-type bioanalytical platforms. Takacs et al. (2003) micromilled an array of 50 × 50 μm posts with aspect ratios of ∼ 10 designed for injection molding cell culturing devices. Zhao et al. (2003) used high-precision milling to fabricate microchannels (64 μm wide and 17 μm deep) in an aluminum plate. The aluminum plate served as a high temperature molding tool to produce a PMMA master for casting PDMS microfluidic devices. Gerlach et al. reported on precision micromilling to fabricate a brass molding master that was used for hot-embossing a polymer microfluidic device (Gerlach et al. 2002; Guber et al. 2004). The device contained 96 capillary electrophoresis columns arranged in a standard microplate format in order to simplify liquid handling. Each column was 100 μm wide and 50 μm deep with 600 pL cross injectors. Our group has also used high-precision micromilling for mold master fabrication. For example, Chen et al. (2005) used micromilling to fabricate a mold master for hot-embossing polycarbonate microdevices for electrokinetically synchronized PCR. The device employed microchannels that were 100 μm wide and 70 μm deep. Situma et al. (2005) used a mold master micromilled in brass for casting of PDMS stencils used for patterning DNA microarrays. Most recently, conventional CNC milling was used for fabrication of aluminum mold inserts for hot-embossing microelectrophoresis devices in PMMA (Mecomber et al. 2006).

A few limitations of micromilling as a method for the fabrication of mold masters have been noted in the literature. These include larger wall roughness factors of the microstructures as compared to lithography-based techniques and the inability of micromilling to produce sharp inner corners as they always have a curvature with the radius equal to the radius of the milling bit (Guber et al. 2004; Mecomber et al. 2006). The intrinsic qualities of micromilling may impose some operational limitations on the use of these masters for replicating parts in certain application areas. For example, in the case of microelectrophoresis, wall roughness has been shown to be detrimental to plate numbers produced with electrokinetically driven separations (Slater and Mayer 1995; Blom et al. 2001; Pu et al. 2003). In addition, the rounded corners of intersecting channels may be troublesome in microfluidic designs, which are dependent on well-defined volumes at channel intersections. One such case is a sample injector for electrokinetically driven separation microchips (i.e., microchip electrophoresis). The size and shape of a sample plug injected into the separation microchannel is an extremely important factor that can affect the separation efficiency (i.e., plate numbers) (Jacobson et al. 1994; Ermakov et al. 2000; Zhang and Manz 2001).

The most common modes of plug injection for microchip electrophoresis includes simple floating injection (Effenhauser et al. 1993; Harrison et al. 1993), floating injection with pull-back voltages (Effenhauser et al. 1994; Jacobson et al. 1994), pinched injection for small plug formation (Ermakov et al. 2000; Ocvirk et al. 2000; Wallenborg and Bailey 2000), and gated injection (Ermakov et al. 2000; Liu et al. 2000). More sophisticated injection schemes that involve the use of additional microfluidic channels and extensive voltage control schemes have also been proposed (Fu et al. 2002, 2003; Thomas et al. 2004a). Zhang and Manz (2001) have shown that through proper geometric design of the injector (loading channels that are five times narrower than the separation channel), highly reproducible and well-controlled injections can be made using simple floating conditions. This approach significantly simplifies the microelectrophoresis system by reducing the number of high-voltage power supplies required for the device.

In this paper, we describe and characterize brass molding masters fabricated via high-precision micromilling. We used numerical simulations to evaluate the effects of the non-sharp intersections of micromilled cross injectors on the size and shape of electrokinetically injected sample plugs. We also discuss the sources and possible effects of increased sidewall roughness of microfluidic channels fabricated via hot-embossing using micromilled masters on the separation efficiency of microchip electrophoresis devices. Finally, we compare a PMMA microchip electophoretic device fabricated using micromilled brass masters to LiGA-prepared masters in terms of separation performance of double-stranded DNA via gel electrophoresis.

2 Experimental

2.1 Fabrication of mold masters

Microstructures were milled on the surface of a 6.3 mm thick brass plate (alloy 353 engravers brass, McMaster-Carr, Atlanta, GA, USA) using a high-precision micromilling machine (KERN MMP 2522, KERN Micro- und Feinwerktechnik GmbH & Co. KG; Germany). According to manufacturer specifications, the micromilling machine is capable of achieving positional and repetition accuracy of ± 1 μm. The milling machine was fitted with a laser measuring system (LaserControl NT, Blum-Novotest GmbH, Germany) for automatic determination of tool length and radius, and an optical microscope (Zoom 6000, Navitar, Inc. Rochester, NY, USA) for monitoring the milling process. 500, 200, 100 and 50 μm diameter solid carbide milling bits were used in this study (McMaster-Carr or Quality Tools, Hammond, LA, USA). Micromilling was carried out at 40,000 rpm at feed rates that were optimized for maximum machining speed and quality of the microstructures. Feed rates were dependent on the size of the milling bit and were typically in the range of 200 mm/min for a 500 μm bit, 100–150 mm/min for 200 μm bit, 50–75 mm/min for 100 μm and 10–20 mm/min for a 50 μm bit. A typical milling cycle consisted of a pre-cut of the entire surface with a 500 μm milling bit to ensure parallelism between both faces of the brass plate and uniform height of the final milled microstructures over the entire pattern, a rough milling of the microstructures using a 500 μm milling bit, and a finishing cut with a smaller diameter milling bit. In the final step of mold fabrication, burrs produced at the top of the microstructures were removed by mechanical polishing on a 3 μm grain size polishing paper (Fibrmet Discs, PSA, Buehler, Lake Bluff, IL, USA). The total time required for fabrication of each mold master of the device layout described herein was less than 3 h.

2.2 Hot-embossing and assembly of PMMA microdevices

PMMA sheets (Plexiglas MC, GE Polymershapes, New Orleans, LA, USA) were cut into 135 mm diameter discs. PMMA discs were rinsed with 2-propanol and distilled water and placed in an oven at 75°C for at least 24 h to remove residual moisture from the bulk polymer. The microchannel pattern was hot embossed into the PMMA wafers using a commercial hydraulic press (PHI Precision Press model number TS-21-H-C, City of Industry, CA, USA). A home-built vacuum chamber was installed into the press to remove air (pressure < 10 kPa) so complete filling of the mold master could take place. The embossing was performed at a force of 5 kN for 5 min at 155°C. After embossing, the microfluidic chips were cut out from the patterned PMMA wafers and 2 mm diameter access holes were drilled at each end of the microchannels. Microfluidic chips were then ultrasonicated in ∼ 0.5% Alconox solution for 20 min, rinsed with DI water, ultrasonicated in DI water for 10 min, again rinsed with DI water, and dried with compressed air. The microchannels were examined under a microscope to ensure they were free from any debris. For the final assembly, a thin PMMA cover plate was thermo-bonded to the face of the microfluidic chip. For this purpose, the microfluidic chip and the cover plate were clamped between two glass plates using binder clips and placed inside a convection oven at 107°C for 0.5 h.

2.3 Numerical modeling of sample injection

Numerical simulations were conducted to examine the detailed characteristics (e.g., length, peak concentration, concentration distribution) of a material plug generated electrokinetically through a standard cross-channel configuration. Four different geometries of the injection cross were examined. All injection crosses had uniform channel widths of 100 μm and the same length of channels (i.e., buffer, waste, and sample channels were 1 mm in length and separation channel was 3 mm long) but differed by the radii of curvature in the corners (0, 25, 50, and 100 μm). The diffusion coefficient of the test species was D = 4.4 × 10−6 cm2/s, initial concentration of the plug was, c0 = 1 × 10−6 mol/L, and effective mobility of the species was μeff = 2 × 10−4 cm2/(V s) (μeff = μep + μeof, where μep is the electrophoretic mobility of the species and μeof is the electroosmotic mobility). The value of μeff was based on our initial experimental data for transport of Alexa Fluor 633 in PMMA devices (1× TBE buffer, pH 8.2). For simplicity, the simulations were two dimensional, i.e., the channel depth was infinite. While this is not the case in practice, this assumption is not critical because it does not compromise the essential physics of the effective electrokinetic and diffusive transport processes in view of the objectives of the present investigations. The same assumption has been used in several other computational investigations (Ermakov et al. 2000).

The equations solved during the simulations are the Laplace equation,
$$ \ifmmode\expandafter\vec\else\expandafter\vec\fi{\nabla }^{2} \phi = 0, $$
governing the electrical potential distribution, φ(x,y), and the species mass transport equation,
$$ \frac{{\partial c}} {{\partial t}} + \ifmmode\expandafter\vec\else\expandafter\vec\fi{u}_{{{\text{eff}}}} \cdot \ifmmode\expandafter\vec\else\expandafter\vec\fi{\nabla }c = D\ifmmode\expandafter\vec\else\expandafter\vec\fi{\nabla }^{2} c, $$
governing the species concentration distribution c(x,y), where \( \ifmmode\expandafter\vec\else\expandafter\vec\fi{u}_{{{\text{eff}}}} = \mu _{{{\text{eff}}}} \ifmmode\expandafter\vec\else\expandafter\vec\fi{E} \) is the effective electrokinetic velocity; D is the mass diffusion coefficient, and \( \ifmmode\expandafter\vec\else\expandafter\vec\fi{E} \) is the electrical field calculated from the definition, \( \ifmmode\expandafter\vec\else\expandafter\vec\fi{E} = - \ifmmode\expandafter\vec\else\expandafter\vec\fi{\nabla }\phi . \) A zero species flux condition, \( \ifmmode\expandafter\vec\else\expandafter\vec\fi{n} \cdot \ifmmode\expandafter\vec\else\expandafter\vec\fi{\nabla }c = 0, \) was imposed on the channel walls to which the electrical field vector is set to be tangent \( ({\text{i}}{\text{.e}}{\text{.}},\;\ifmmode\expandafter\vec\else\expandafter\vec\fi{n} \cdot \ifmmode\expandafter\vec\else\expandafter\vec\fi{\nabla }\phi = 0). \)

The simulations were performed using CoventorWare ANALYZERTM software (SwitchSim module ver. 2004, Coventor, Inc., Cary, NC, USA) with a spatial grid appropriately refined to render the results grid independent and a time-step of 50 ms, which is more than two orders of magnitude smaller than the electrophoretic time to transport the species from the center of the cross to the exit of the separation channel. A fluorescence detector signal was simulated from the results assuming that the intensity is proportional to the integral of the species concentration across the channel. This signal was recorded as a function of time for “detector” location at 350 μm from the center of the cross junction.

2.4 Measurement of the electroosmotic flow (EOF)

The electroosmotic flow (EOF) in assembled devices was measured using the method described elsewhere (Huang et al. 1988). The procedure involved filling the entire chip with a 20 mM buffer. After filling the chip, one reservoir was emptied and filled with the same type of buffer but of lower ionic strength (18 mM). An electric field (400 V/cm) was then applied to the reservoirs containing the low and high ionic strength buffers and the current was monitored continuously using a multimeter (Fluke 189 Logging Multimeter, Fluke Co., Everett, WA, USA) interfaced to a personal computer. The time required for the current to reach a plateau was measured from the plot and the linear velocity calculated. Dividing the linear velocity by the electric field strength produced EOF values [cm2/(V s)]. The electric field was supplied by a Spellman high-voltage power supply (CZ1000R, Plainview, NY, USA). The EOF measurements were carried out in TBE (Tris–borate–EDTA) buffer, pH 8.2.

2.5 DNA separations

A linear polyacrylamide (LPA) sieving matrix was prepared from high-viscosity-average molecular mass powder (∼ 6 MDa) (Polysciences Inc., Warrington, PA, USA) and dissolved in 1× TTE (50 mM Tris/50 mM TAPS/2 mM EDTA) buffer for electrophoretic separations of double-stranded DNA fragments. The 4% (w/v) LPA was replaced between each run from the anodic end of the separation channel. The electrophoresis buffer, consisting of 1× TTE buffer, was also changed after each run. pUC19 Sau3AI was obtained from Abgene Inc. (Rochester, NY, USA) and labeled with 1 μM TO-PRO-3 dye (Molecular Probes, Inc., Eugene, OR, USA). The DNA ladders were loaded by applying 400 V/cm across the injection channel for 25 s with the anodic and cathodic buffer reservoirs floating. Electrophoretic separations were run at 110 V/cm; sample leakage into the injection cross was prevented by applying pull-back voltages. Microchips used for DNA separations were hot embossed in PMMA and had a dual T-type injector with 500 μm offset between the sample and waste channels. All microchannels were 25 μm wide and 120 μm (LiGA) or 75 μm (micromilled) deep.

Fluorescence detection was performed using an in-house constructed near-IR laser-induced fluorescence (LIF) system. The excitation source consisted of a diode laser (630 nm) (NT 54-151, Edmund Industrial Optics, Barrington, NJ, USA). The laser was focused to a 10 μm diameter spot into the microchannel using a 40× (NA = 0.65) microscope objective. The resulting emission was collected through the same objective, routed through the dichroic beam splitter and filtered using a stack of optical filters. The filtered fluorescence emission was ultimately detected by a photomultiplier tube (RT 1508, Hamamatsu, San Jose, CA, USA).

The high-voltage power supply was assembled in-house using four independent high-voltage modules (EMCO, Sutter Creek, CA, USA) configured into separate electrophoretic modes. The high-voltage power supplies and relays were controlled by a computer using an analog output (D/A) card (PCI-DDA04/12, National Instruments, Austin, TX, USA). Software written in LabView (National Instruments) was used for both collection of LIF signal and control of the high-voltage power supply. Caution: the electrophoresis uses high voltages and special care should be taken when handling the electrodes.

3 Results and discussion

3.1 Injection of sample plug in microelectrophoresis devices fabricated using micromilled mold masters

High-precision micromilling is capable of producing in just one fabrication cycle multi-level structures with highly vertical sidewalls and aspect ratios exceeding 20:1 (see Fig. 1a). However, the micromilling process is unable to make sharp inside corners due to the intrinsic feature of the process itself—finite size of the milling bit (see Fig. 1b). Using smaller milling bits can minimize the curvature of corners (Fig. 1c) but at the same time the achievable height of the structure is limited by the useful flute length of the milling bit. For example, aspect ratios of commercially available micromilling bits are typically less than 3, which means that milling bits with diameters of 25 μm have useful flute lengths of 75 μm and thus, the maximum height of the microstructures is ∼ 75 μm.
Fig. 1

SEM photomicrographs of the microstructures milled in brass. a High-aspect ratio structure for hot-embossing of micromixer. The narrower walls are 20 μm wide and 400 μm tall (aspect ratio 20:1). b Cross structure finished with a 100 μm radius milling bit and a schematic representation of the milling process and the designed shape of the cross (dotted lined) with sharp corners. c Cross structure finished with a 25 μm radius milling bit

When using micromilling tools for the molding of microchip electrophoresis devices, ideal control of the geometry of the injector cross is not possible due to curvature of the corners introduced by the milling process. The presence of curved corners creates an additional injection volume (Vadd). Since the radius of the curvature is equal to the radius of the milling bit (R), the magnitude of the additional volume is proportional to the square of the milling bit radius for a given channel height (h) and can be calculated as:
$$ V_{{{\text{add}}}} = (4 - \pi )hR^{2} $$

For example, the additional volume introduced by two intersecting, 100 μm high channels is 53.6, 215, and 858 pL for cross structures micromilled with 25, 50, and 100 μm milling bits, respectively. One should also note that since the radius of curvature is a finite value determined by the size of the milling bit used to fabricate the mold master, the relative effect of the curvature of the corners will be higher for narrower than for wider channels of the same radius of curvature. For example, the injector volume defined by cross channels with sharp corners that are 100 μm high is 1 nL for 100 μm wide channels and only 63 pL for 25 μm wide channels. Therefore, the radius of curvature to channel width ratio (R/W) is a good metric for characterizing the performance of cross injectors of microchips molded from micromilled masters.

To investigate the differences between size and shape of the sample plug injected into a microchip electrophoresis device using simple cross injectors with different curvatures in the corners, numerical simulations were performed. Figure 2 presents the results of numerical simulations for cross injectors with radii of curvature ranging from 0 to 100 μm. As expected, the sample plug injected into the separation channel becomes longer with increasing radii of curvature in the corners. The FWHM of the electrophoretic peaks generated by a moving plug were (assuming a linear velocity of u = 0.334 mm/s): 1.04, 1.12, 1.27 and 1.64 s for 0, 25, 50, and 100 μm radii of curvature, respectively. The length of the plugs measured at 20% peak height were: 510, 530, 590, and 750 μm for 0, 25, 50, and 100 μm radii of curvature, respectively. Additional peak areas for different radii values after subtracting the area of the peak for R = 0 were: 0.068, 0.198, and 0.554 μmol s/L for R = 25, 50, and 100 μm, respectively. These results indicate that the effect of round corners on the sample plug size is relatively small (10–20%) for injectors with R/W < 0.5 and increases to ∼ 50% for R/W = 1.
Fig. 2

Numerical simulations of loading and dispensing of sample into cross injectors with different geometries. a The pictures present only the central part of the simulated micro-channels. 1 Waste, 2 buffer, 3 sample, and 4 separation microchannel. Loading: 1 at 33.4 V, 3 at GND, 2 and 4 float. Dispensing: 1 and 3 float, 2 at GND, 4 at 66.8 V. Eload = Edisp = 167 V/cm; tload = 7.5 s, tdisp = 11 s. t0 corresponds to the start of dispensing; t1 and t2 were taken at 1 s increments. b, c Each data point is the average concentration across the separation channel obtained 350 μm downstream from the center of the injector. b Loading and dispensing parameters were the same as in a. c Loading parameters were the same as in a, dispensing: 1 and 3 at 28 V, 2 at GND, 4 at 66.8 V

Data presented in Fig. 2b indicates the presence of peak tailing for all geometries investigated. The empirical asymmetry factor (B/A)(Foley and Dorsey 1983) measured at 10% of peak height was 2.54, 2.42, 2.40, and 2.40 for 0, 25, 50, and 100 μm radii of curvature, respectively. Similar values of B/A indicate that the magnitude of peak tailing is not dependent on the geometry of the injection cross but rather originates from the injection scheme used. Simple injections without pull-backs (as used in this simulation) can lead to sample leakage from the injection channel into the separation channel producing peak tailing. Such leakage is caused by both diffusion of the sample and migration of the sample into the electric field and can be eliminated by using properly designed pull-back voltages applied to the sample and waste reservoirs during the separation step.

Figure 2c shows results of a numerical simulation of plug injection with pull-back voltages applied to the sample and waste reservoirs during the transport phase for R = 0 μm and R = 100 μm. The plug appears to move slower as indicated by the longer migration time for the peaks as compared to injection without pull-back voltages. This is due to the lower effective electric field imposed on the sample (167 V/cm without pull-back and ∼ 110 V/cm with pull-back). The lengths of the plugs measured at 20% peak height were: 320 and 460 μm for 0 and 100 μm radii of curvature, respectively (for a linear velocity of u = 0.220 mm/s). More importantly, peak tailing could be almost eliminated by applying pull-backs (B/A = 0.98 for R = 0 μm and B/A = 1.12 for R = 100 μm) indicating that the only effect of curved corners is in terms of the length of the injection plug when proper injection schemes are used.

Simulation results also indicated that using cross injectors with large R/W leads to the injection of relatively long plugs. Longer injection plugs may compromise separation efficiency due to overloading introducing extra-column effects on the total plate height. The efficiency in the separation can be expressed in terms of the plate height (H = L/N, where L is the length of the column and N is the number of theoretical plates). If the on-column sample spreading is dominated by longitudinal diffusion, the contributions to the plate height can be determined using:
$$ H_{{{\text{tot}}}} = H_{{{\text{inj}}}} + H_{{{\text{det}}}} + H_{{{\text{diff}}}} = \frac{{l^{{\text{2}}}_{{{\text{inj}}}} }} {{12L_{{{\text{sep}}}} }} + \frac{{l^{{\text{2}}}_{{{\text{det}}}} }} {{12L_{{{\text{sep}}}} }} + \frac{{2D_{{\text{M}}} }} {u} $$
where Hdiff, Hinj, and Hdet are the contributions of axial diffusion, injection plug length, and detection length, respectively. DM is the diffusion coefficient of the analyte in the buffer, u is the linear velocity of the analyte, linj is the injection plug length, ldet is the detector observation length, and Lsep is the separation length. Equation 4 indicates that for higher separation speeds where the contribution from sample diffusion decreases, the relative effect of longer injection plugs for devices with high R/W injectors will be more pronounced. The effect of high R/W will also be more pronounced for the separation of species with smaller diffusion coefficients.

In order to generate well-defined short axial extent sample plugs suitable for high-performance separations, pinched injection can be used (Jacobson et al. 1998; Ermakov et al. 2000). This method uses spatial containment of the sample in the cross intersection by means of electrokinetic focusing before dispensing it into the separation channel. During the dispensing step, pull-back voltages are applied to minimize sample leakage from the sample and waste channels into the separation channel. Studies by Ermakov et al. (2000) have shown that increasing the pinching voltage leads to shorter sample plugs but also lowers the maximum concentration of the analyte (Cmax) in the plug. In a similar way, increasing the amount of pull-back applied during dispensing leads to shorter plugs and lower sample concentrations in the plug. Ermakov et al. (2000) have used Cmax/σ, where σ is the standard deviation of the sample peak, as one of the metrics to determine optimal injection conditions. The electric field distribution was described by using relative electric field strengths in the channels, єi = Ei/Emax, where Ei is the electric field strength in channel i in that portion of the channel that is far from the cross intersection and where it is assumed to be uniform and constant and Emax is the maximum field value. These authors found that Cmax/σ reaches a maximum for |є3L| = 0.6 and |є4D| = 0.3, which can be described as a weak pinch combined with a medium sample pull-back (channel assignment as in Fig. 2, L and D denote loading and dispensing step, respectively).

Figure 3 presents data of numerical simulations of pinched injections at |є3L| = 0.6 and |є4D| = 0.3 for both a cross injector with sharp corners and rounded corners with R = 100 μm. Clearly, pinched injection allows for generating much shorter plugs in both geometries as compared to floating injection and floating injection with pull-backs. For example, the plug lengths measured at 20% peak height were: 165 and 255 μm for 0 and 100 μm radii of curvature, respectively. Extended plug lengths were observed using injectors with rounded corners accompanied by much higher Cmax as compared to sharp-corner injectors. This indicates that in the case of rounded corner injectors, stronger pinching may be used without sacrifice of Cmax significantly. Indeed, as shown in Fig. 3 the electric field conditions that allow for stronger pinching, |є3L| = 0.3, lead to much shorter plugs (200 μm) and a Cmax that is roughly the same as optimal conditions for sharp-corner injectors.
Fig. 3

Numerical simulations of pinched injection in cross injectors with 0 and 100 μm radii of curvature (R0 and R100, respectively). Each data point is the average concentration across the separation channel obtained 350 μm downstream from the center of the injector. 1 corresponds to |є3L| = 0.6 and |є4D| = 0.3; 2 corresponds to |є3L| = 0.3 and |є4D| = 0.3

3.2 Wall roughness and its effect on electrophoretic separation efficiency

SEM investigations of the walls of the microstructures milled in brass revealed the presence of characteristic milling marks along the sidewall (Fig. 4a, b). Such milling marks are due to imperfections of the cutting edge of the milling bit. They propagate in the same direction as the milling bit movement (parallel to the mold master floor) and are a direct indicator of the quality of the milling bit. During hot-embossing, the milling marks are transferred to the sidewall of the microchannel (Fig. 4c, d). To quantify the quality of the sidewall roughness of micromilled structures, various sections of the molding die were characterized using a stylus profilometer. It was not possible to directly measure the sidewall roughness of the microstructures due to the large size of the mold insert and the central placement of the microstructures on the mold, which made them inaccessible to the stylus. Instead, an edge of a small brass cube (5 × 10 × 20 mm3) was machined using the same milling bit and conditions as those used for the mold insert fabrication. Similarly, it was not possible to directly access the sidewalls of the hot-embossed microchannels to measure roughness using the profilometer. Therefore, to evaluate the roughness of the hot-embossed sidewalls, the hot-embossed microdevice was carefully and precisely fly cut to expose the sidewall to the profilometer stylus. Figure 5 presents profilometer scans of various areas of the micromilled brass master. The roughness parameters; Ra (average roughness) and Rpm (mean peak height) for both micromilled mold master and hot-embossed PMMA chip are summarized in Table 1.
Fig. 4

SEM photomicrographs of micromilled molding master finished with a 100 μm radius milling bit (a, b) and its replicate hot-embossed into PMMA (c, d). The width and depth of the channel, as determined by the dimensions of the mold master, was 100 and 90 μm, respectively

Fig. 5

Profilometric scans of different areas of the brass mold insert and of the surface of a PMMA cover plate. Data were obtained for mold insert finished with a single axial pass of 200 μm diameter milling bit. The scanning speed of the profilometer stylus was 20 μm/s with a force of 2 mg

Table 1

Summary of roughness data

Tested area

Micromilled brass

Hot-embossed PMMA

Ra (nm)

Rpm (nm)

Ra (nm)

Rpm (nm)

Top (bottom)a





Sidewall along





Sidewall top-to-bottom





Floor (face)a





All scans were performed over a 1 mm length except for sidewall top-to-bottom which was measured over 300 μm length

aIndicates part of embossed PMMA chip

As expected, the lowest roughness was observed for the mechanically polished top of the metal microstructures as the quality of the surface finish can be easily controlled by proper selection of the polishing medium. Much higher roughness was observed for the sidewalls, which resulted from the fact that the sidewalls could not be subjected to post-milling polishing. The roughness measured along the milling path (i.e., parallel to the floor of the mold insert) was ∼ 2× lower than the roughness measured from the top to the bottom of the microstructures. This difference stems from the fact that the sidewall finish in the direction parallel to the floor can be controlled to some extent by the milling conditions (i.e., feed rate and rotational speed of the milling bit), whereas the top-to-bottom roughness is in part an inverse image of the cutting edge of the milling bit. Relatively high roughness was also observed for the molding die floor. Although this roughness does not directly influence the quality of the molded microchannels, it can affect the thermo-sealing of cover plates to the plastic replicates as shown by Chen et al. (2004). If this is the case, vacuum-assisted thermo-sealing can be used to alleviate this problem.

It is important to note that the characteristics (i.e., shape, cutting edge roughness, and length) of the milling bit may change during the machining process due to bit wear or breakage (Schaller et al. 1999; Takacs et al. 2003). This may lead to inconsistency in microstructure dimensions and surface finish at different parts of the mold master. To minimize such problems, two strategies were employed. First, brass was used as the material for the mold master. Brass has been shown to have excellent machining properties (low tool wear) and the required strength for hot-embossing (Schaller et al. 1999; Guber et al. 2004). Second, a two step milling process was used. In the first roughing step, bulk material was removed using a large (500 μm) diameter bit. In the second finishing step, a smaller diameter milling bit was used to fine finish the structures.

Data presented in Table 1 indicate that the roughness factors observed in the molding die were precisely transferred to the embossed polymer. This observation is in agreement with literature reports, which indicated hot-embossing as a technique capable of replicating structures down to the nanometer range (Chou et al. 1996). The important observation is that the magnitude of the sidewall roughness of the PMMA microchannels hot embossed using micromilled masters is not significantly larger than a double layer thickness (10–100 nm) typically observed in microchip electrophoretic separations. If the wall roughness is significantly larger than the double layer thickness it may become a factor influencing electrophoretic separations. For example, it has been shown that rough microchannel walls produce smaller EOF values than smooth microchannel walls prepared in the same material (Pu et al. 2003). The authors found that glass microchannels fabricated by powder-blasting (roughness of 1–5 μm) produced an EOF that was 4–7% smaller than in glass microchannels fabricated via HF etching (roughness of 3–15 nm). To evaluate if a similar effect exists in microchannels produced from micromilled mold masters, the EOF was measured in microchannels with varying aspect ratios but similar cross-sectional areas to eliminate possible artifacts due to Joule heating. Therefore, if the roughness of the sidewalls influenced the bulk EOF, one would expect that channels having different aspect ratios will develop different EOFs. The EOF values measured in TBE buffer (pH 8.2) were 3.70 ± 0.13 × 10−4 cm2/(V s) (RSD = 3.5%), 3.73 ± 0.13 × 10−4 cm2/(V s) (RSD = 3.5%), 3.60 ± 0.15 × 10−4 cm2/(V s) (RSD = 4.2%), 3.72 ± 0.21 × 10−4 cm2/(V s) (RSD = 5.6%), and 3.67 ± 0.17 × 10−4 cm2/(V s) (RSD = 4.6%) for 0.25, 0.5, 1, 2, and 4 aspect ratio microchannels, respectively. These results indicate that there are no statistically significant differences between the measured EOFs for channels of different aspect ratios with the scale of roughness factors produced via micromilling. For comparison, a LiGA–prepared microchannel (Ra = 20 nm) produced a statistically similar EOF value (3.76 ± 0.15 × 10−4 cm2/(V s) (RSD = 4.0%)).

The roughness of the walls of microchannels has also been demonstrated as a factor that can lead to decreased efficiency of electrophoretic separations through Taylor–Aris dispersion (Slater and Mayer 1995; Bianchi et al. 2001; Blom et al. 2001). In such case, Eq. 4 must be modified to include additional plate height contributions to Htot from Taylor dispersion (Htaylor). Studies of Slater and Meyer (1995) have indicated that channel roughness has to be on the order of at least 10% of the channel dimension to generate a noticeable loss in separation efficiency due to Htaylor. Considering that the microchannel sizes are in the order of 10–100 μm, it is not likely that the roughness of the microchannels fabricated through hot-embossing using micromilled mold masters would lead to any significant loss in separation efficiency due to mechanical dispersion. On the other hand, for electroosmotic separations where very small Debye lengths are observed, Blom et al. (2001) have found experimentally that 3 μm square periodic perturbations on a 100 μm wide channel will increase the effective diffusion coefficient, Deff, of the sample by a factor of 0.65 for a 1 mm/s EOF velocity and by 16.25 for 5 mm/s flows. These findings indicate that the sidewall roughness observed in polymer microchannels fabricated through hot-embossing of micromilled mold masters may actually lead to increased Taylor–Aris dispersion when channel widths approach 10 μm and high electric fields, high sample velocities, are used.

3.3 Electrophoretic separation of double-stranded DNA

To demonstrate the applicability of using micromilled molding tools for manufacturing polymer microchip electrophoresis devices, a separation of double-stranded DNA was performed in similar topographical devices manufactured using either a LiGA or micromilled molding master. pUC19 Sau3AI DNA stained with 1 μM TO-PRO-3 was separated in a 4% LPA gel using both devices (Fig. 6). Pull-back voltages were applied during the separation stage immediately after injection in order to minimize peak tailing. High quality separations were observed for both devices. The differences in the peak intensities observed between these devices may result from differences in the depth of the microchannels (120 μm for LiGA and 75 for micromilled). For the device embossed using a micromilled molding tool, the calculated plate numbers were 1.3 × 106 m−1, 1.4 × 106 m−1 and 8.4 × 105 m−1 for the 105, 341 and 955 bp fragments, respectively. By contrast, the plate numbers generated in the LiGA device were 1.8 × 106 m−1 for the 105 bp, 1.7 × 106 m−1 for 341 bp and 3.9 × 105 m−1 for the 955 bp fragments. These values compare well to ones previously reported for gel separations of double-stranded DNA fragments in PMMA microchannels (Lee et al. 2001; Qi et al. 2002; Thomas et al. 2004b). The results shown here suggest that structural features in the molding tools produced through differences in micromanufacturing demonstrate negligible effect on electrophoretic performance of each device.
Fig. 6

Separation of pUC19 Sau3AI double-stranded DNA stained with 1 μM TO-PRO-3 dye using a PMMA microchip with an offset T injector. Einj = 400 V/cm; tinj = 25 s; Esep = 110 V/cm; separation length 3.5 cm; 4% LPA was used as the sieving matrix. PMMA microchips were hot embossed using either a high-precision micromilled master or b LiGA-made master

4 Conclusions

We have successfully used high-precision micromilling to fabricate molding masters that can be used for hot-embossing microfluidic chips in polymers. The entire fabrication process from design to the finished device was less than 6 h (< 3 h for brass master fabrication, < 3 h for hot-embossing, post-machining, cleaning, and final assembly) making this technique a very attractive method for rapid prototyping of polymer-based microfluidic devices. As an example of using micromilled masters for fabricating polymer microdevices, a microchip electrophoresis device was fabricated and tested via experimentation and numerical simulations. The results of the numerical simulations indicated that additional injection volume present in the cross injector due to the curvature of micromilled corners led to increased sample plug size. This effect could be minimized for designs with radius of curvature to channel width ratios (R/W) of less than 0.5. In order to generate short sample plugs pinched injections with strong pinching potentials have to be used. Increased wall roughness for microdevices fabricated via micromilling did not seem to affect the performance of the device in any significant way compared to LiGA-prepared microchips. Direct comparison of a microchip electrophoresis device produced from a micromilled master to one made using LiGA showed similar analytical figures of merit for both devices with the number of theoretical plates for the separation of double-stranded DNA exceeding one million plates/m.


The authors gratefully acknowledge the financial support of the National Institutes of Health (R24-EB0002115) and National Science Foundation (EPS-0346411). The authors would also like to thank Dr. Varshni Signh of the Center for Advanced Microstructures and Devices (CAMD, LSU) for help with obtaining SEM images.

Copyright information

© Springer-Verlag 2006