1 Introduction

Applications for nanomaterial are well spread within many disciplines of research and also industry. In particular, nanoparticles (NPs) have been able to cover a broad range of employment. Commercial products for daily life like cosmetics and household appliances often rely on the benefits of nanotechnology. Moreover, advantages based on NPs are firmly established within pharmaceuticals, medical devices, odor-resistant textiles and food packaging (Maynard 2007). High concentrations of NPs can be set free into the ambient especially during the production of NP-comprising goods where workmen act at a relatively high probability to be exposed to those engineered NPs directly. Beside contact with the skin, inhalation of airborne NPs can cause pulmonary inflammation instantaneously or raise the risk to suffer from other diseases (Ling et al. 2011). NPs are also well known for their ability to penetrate through the lung-barrier and enter the blood circulation (Nemmar et al. 2002). Thus, possible impacts are not limited to the lungs.

Appraising the risk potential relies on NP concentration. Sensing should be applied personally and in best case next to the respiratory passages. Direct sensing of mass concentration is worthwhile for an assessment of monitoring affected workplaces. Micro/nanoelectromechanical system (M/NEMS)-based resonant mass sensors offer a small and lightweight solution. Such technology allows a low-cost fabrication of a huge amount of sensors in parallel. Mass concentration is tracked by sensing respective resonant frequencies. Surface-deposited NPs add their masses to the resonator. Thus, a shift in resonant frequency related to the additional mass is caused. High mass sensitivity and resolution can be reached due to the small dead weight of M/NEMS-based resonators. A sensitivity of ~1.5 kHz/pg has been reached using I-shaped bulk acoustic resonators (IBAR) which were able to sense particles of ~200 nm in diameter (Hajjam et al. 2011). Recently, using a self-sensing cantilever resonator we have proven a sensitivity of 10 Hz/ng for engineered carbon aerosol NPs in bimodal size distribution at ~20/120 nm in diameter (Wasisto et al. 2012).

With decreasing weight of a resonator, its sensitivity raises. Seen from this perspective, downsizing is beneficial. NEMS-based vertical silicon nanowires (SiNWs) are desirable to support miniaturized sensors. In this work, two top-down techniques for fabrication of such wires are presented. Beside conventional photolithography, nano-imprint lithography (NIL) is applied to reach small dimensions, keeping a high quality of the pattern. Inductively coupled plasma (ICP) deep reactive ion etching (DRIE) is applied to create a crystal-independent shape of the wires. Shrinking of wire diameters by multiple sacrificial thermal oxidation and oxide stripping has been used.

This paper focuses on the parameters of significant processes in wire fabrication. Characterization of SiNWs has been performed by scanning electron microscope (SEM) for shape issues and for resonance. The latter is assisted by a shear-piezo element driven in the range of the resonant frequency of the wires.

2 Fabrication of vertical silicon nanowires

Nanowire arrays are created in top-down approaches using two different types of lithographic patterning. The general sequence is shown in Fig. 1. A 2-in. silicon wafer from semiconductor foundry Siltronix is covered by a resist (Fig. 1a). The pattern of the mask is transferred into the resist in a following step (Fig. 1b). Subsequently, the surface pattern is finalized by removal of unwanted area coating (Fig. 1c). After that treatment, the silicon is etched by cryogenic DRIE (Fig. 1d). The processing details of conventional optical lithography and NIL-based sequences are explained below.

Fig. 1
figure 1

General sequence for top-down fabrication process of vertical silicon nanowire arrays, comprising a resist coating, b nanolithography, c resist masking finishing, and d cryogenic DRIE

Full size image

2.1 Surface masking by lithography

Prior to a coating, a dip in hydrofluoric acid (HF-dip) removes the natural oxide layer on the wafer. Optical lithography is performed with AZ 5214 E (MicroChemicals GmbH, Ulm, Germany) as photoresist. A thin film is achievable by adding AZ EBR Solvent (MicroChemicals GmbH, Ulm, Germany) to the resist (1:1 vol). A spinning velocity of 3,000 rpm for 35 s results in a homogeneous layer thickness of ~350 nm. The whole surface area is coated uniformly by this method. After a softbake at 110 °C for 50 s, UV lithography changes the chemistry of the resist. An optical power density of 3 mW/cm2 provided by a mask aligner MJB4 (SÜSS MicroTec AG, Garching, Germany) at a timeframe of 10 s exposes the resist. For structures within the submicrometer range, the lithography has to be very well arranged. A slight pattern deviation is able to reduce the performance of the desired sensors. Thus, parameters are optimized for high pattern quality. Concerning the given values, an exposure resolution below 300 nm is achievable by applying a vacuum contact between mask and wafer. It guarantees the highest structural conformity in comparison with other possible setups like soft or hard contact.

The mask itself has been made by e-beam lithography. An electron beam pattern generator (EBPG 5000 plus, Vistec Electron Beam GmbH, Jena, Germany) firstly transfers the lateral information in a sequence into a 930 nm thick PMMA-layer (AR-P 641.09, ALLRESIST GmbH, Strausberg, Germany) at the photomask. Several different patterns as shown in Fig. 2 are arranged at a single mask to test a range of different diameters and distances of wires. Different metals coatings have been tested to optimize the shaping template. In this paper, an aluminum layer of 60 nm serves as a shadowing mask. The aluminum is structured by a standard lift-off process. Smallest intended structure sizes are at 200 nm in diameter.

Fig. 2
figure 2

Layout (a) and photograph of the mask (b) prepared by e-beam and aluminum lift-off ready for patterning the silicon wafer substrate containing various diameters and distances of the wire arrays between 200 nm and 5 μm

Full size image

After the mask-depending exposure of the resist, the wafer is immersed into AZ 726 MIF developer (MicroChemicals GmbH, Ulm, Germany) for 50 s. This step is very critical because of dark erosion. It impairs the shape and thickness of the resist mask. While removing the sample from the beaker, the liquid still covers the surface and the development is ongoing. An immediately following deionized (DI)-water immersion stops the process and fixes the structures.

Whereas optical lithography is well established concerning structuring, it suffers from dependency on exposure wavelength. A possibility to avoid this disadvantage and still keep most of the advantages given by resist technology, NIL is applied in a second approach to fabricate SiNWs. The silicon wafer is treated by a HF-dip as before and the surface is activated for 2 min by oxygen plasma. As resist, UV-Cur06 (micro resist technology GmbH, Berlin, Germany) is used to spin coat a 163 ± 2 nm thick layer during 60 s of coating at a spinning velocity of 4,000 rpm. Before, a stamp made of PDMS (SYLGARD 184, Dow Corning, Midland, USA) with a homogeneous hexagonal pattern of 150 nm high pillars at diameters of 400 nm and distances of 2,500 nm has been prepared from a master template. The coated wafer and the stamp are aligned manually after a 2 min prebake of the resist at a temperature of 100 °C. The stamp is pressed into the resist layer by gauge pressure of 5 bars. A UV-LED (Winger Electronics GmbH, Dessau-Roßlau, Germany) with an emission line at 405 nm is used to expose the resist layer. To harden the complete layer a time of 5 min is required. After detaching the stamp the residual resist layer is etched for 20 s with oxygen plasma. The plasma power is 200 W and high-frequency (HF) power is 10 W. As reaction gas 10 sccm O2 are flowing in the reaction chamber and the chamber pressure is 0.5 Pa. An acceleration voltage of −22 V is built up and the wafer is cooled down to −60 °C. After this maskless etching process the holes in the resist reached the underlying silicon wafer and the standard NIL-process is finished.

For etching purposes, a metal mask is preferable due to the high degradation of resist while ICP–DRIE processes. Sulfur hexafluoride (SF6) is employed to etch the silicon for 1 min causing a small hole in the silicon, which is used after the next step. The recipe contains 150 W plasma power, 4.8 W high-frequency power, 129 sccm SF6 flow rate and 0 °C wafer temperature. The complete substrate is metalized with 30 nm thick layer of aluminum. After the metallization a lift-off procedure is performed. The resist and the metal layer are removed afterwards. Only the metal dots are remaining at the silicon. The metal mask of 30 nm thickness is then used for cryogenic DRIE.

2.2 Cryogenic ICP–DRIE

Next to lithography, DRIE is the most critical processing step during fabrication procedure of SiNWs with high aspect-ratio. From literature, there are two common DRIE methods available named Bosch process and cryogenic dry etching, respectively. The Bosch process has been stated generally capable to produce nanowires (Abdolvand and Ayazi 2008; Nakamura et al. 2013). Even so, the principle suffers from several disadvantageous characteristics. Beside the passivation layer, which is arranged within a repeated process step and has to be removed after finally finishing the etching, the smoothness of etched sidewalls depends on the process cycle time. This phenomenon is well known as scalloping effect. The smaller the aimed structures, the more complex is the process. Thus, lateral dimensions in the nanometer regime are challenging. In our process, cryogenic DRIE is applied for deep etching. Thus, a continuous material removal without alternating gases provides smooth sidewalls (Stranz et al. 2011; Sökmen et al. 2010). Figure 3 illustrates the principle etching mechanism of the cryogenic DRIE. SF6 and O2 are provided as continuous gas flow inside the reactor. During the process those gases react with silicon and form a solid passivation layer of SiOxFy at surface temperatures below −75 °C (Pereira et al. 2009). Due to directed kinetic energy transfer by the ions, the bottom is by far more easily cracked, than the sidewalls. Parameters should be arranged in a way to save the sidewall passivation. Key parameters are oxygen content and temperature. By adjustment of those conditions, the shape of the sidewalls is modifiable in a broad range around the preferred etching shape. The passivation layer desorbs during the heating from process temperature to normal ambient conditions until the layer is vanished completely. Dry etching does not depend strongly on crystallographic directions like wet etching methods by KOH or TMAH chemistry. Thus, its anisotropic nature allows forming of SiNW arrays independent of crystal orientation of the wafer. Furthermore, no hard mask like thermally grown oxide is needed to maintain a high selectivity between photolithographic resist and silicon which is in the range of 100:1 in the case of cryogenic DRIE (Stranz et al. 2011).

Fig. 3
figure 3

Schematic of the cryogenic reactive ion etching chemistry illustrated at an etched face. The gases used in the etching are SF6 and O2 which create a passivation layer on silicon at cryogenic temperatures

Full size image

The conventionally patterned wafers are etched within 20 min at −95 °C at a reactor pressure of 1.5 Pa inside a SI 500 C cryogenic dry etcher (Sentech Instruments GmbH, Berlin, Germany). A continuous flow of SF6 and O2 is kept stable at 129 and 11 sccm, respectively. By adjusting the parameters, the etch rate can be raised up to ~5 μm/min depending on the surface pattern and the overall resist-free area.

After cryogenic DRIE, the residual photoresist layer from the top of the wires is washed off using piranha etching, i.e. a mixture of hydrogen peroxide (H2O2) and sulfuric acid (H2SO4, 98.08 g/mol) tempered at ~180 °C inside a quartz beaker containing the wafer. The inspection by SEM shows a maximum aspect-ratio of ~40 at a diameter of ~765 nm (Fig. 4d). As a conclusion, the resist patterning errors, including mask misalignment and dark erosion caused structure shrinkage of ~120 nm from both sides in this case. However, large parts of that SiNW-array section have not been fabricated. Other structures have been formed in an array of an intended wire diameter of D = 2 μm at a distance G = 5 μm. The aspect-ratio is 17.5 here. This array has been formed homogeneously without failure. As visible in Fig. 4, the etching is more aggressive at the rim, than at the middle part of the array section, which can be attributed to faster material removal of etching products at the rim. This causes the outer rows of wires to be undercut, while the slender wires at a distance of three rows and more from the rim still keep their cylindrical shape. As a result, the outer rows lose ground within the process and tumble down onto the surface. The sidewall of such a wire is still covered by its passivation thus forming a mask to the underlying silicon as shown in Fig. 4a. Ongoing etching then leads to the formation of the walls visible in Fig. 4 at parts b, c and f.

Fig. 4
figure 4

SEM images (30° tilted) of results for cryogenic dry etching of conventionally patterned arrays by UV-photolithography depicting different wire aspect ratios (i.e. 17.5 at b–d; 39.7 at f, g), merging wires at an array with small gap between wires (e) and a sketch (a), illustrating the effect of tumbling wires at the rim

Full size image

Unfortunately, there is no possibility to achieve wires in every array section of the pattern. Due to geometrical changes, the wires will end up in cones or the diameter shrinks till the wire falls to the ground (cf. Fig. 4e). While the gap between the wires causes conically shaped wires till they merge, the etching at the rim of the array itself causes an undercut.

Additional treatment by sacrificial thermal oxidation and HF-dip for oxide stripping can further shrink wire diameters. The high reproducibility of oxide formation by thermal growth offers a well-controlled shrinkage of 3D silicon structures (Stranz et al. 2010). The principle itself is illustrated by Fig. 5. In the given example, a wire at a diameter of 3.47 ± 0.03 μm after cryogenic DRIE is treated for three cycles. The process of DRIE differed from the previous recipe by temperature (−80 °C) and O2 supply (10 sccm). Thermal oxidation is performed in two short steps (30 min, each) and a finalizing long-term step (1.5 h) at 1,100 °C. During the short steps, thermal oxide has grown 315 and 275 nm into the silicon wires, respectively. After third oxidation, the oxide has diminished the silicon by another 815 nm around the wire. These values have been proven by SEM inspection after HF-dip for 10 min each step. A more detailed explanation is written by Stranz et al. (2011). With a final diameter of 660 nm, the aspect-ratio for these structures has reached values beyond 60.

Fig. 5
figure 5

Basic sequence of sacrificial thermal oxidation for wire shrinking with subsequently HF-treatment for stripping the oxide

Full size image

However, the density of nano-sized wires that can be achieved in an area this way is limited. Therefore, in this study NIL was applied as an alternative. A master with pillars ordered regularly within a square pattern at a pitch of 480 nm and pillar diameters of 300 nm serves as a template to produce PDMS stamps. Those stamps include the inverse topography (i.e. pattern of holes), to imprint resist pillars again. The UV-based method does not suffer from optical constrictions.

Thus, a much higher density of wires is possible by lithography. For dry etching, the parameters have to be well arranged. The upper part of Fig. 6 shows SEM images of a series of samples etched at different substrate temperatures. The value for the temperature given for the examples is measured just below the sample containing carrier in the reactor. Thus, the real surface temperature of the sample is slightly higher depending on the thermal energy input by the etching. As discussed before, the temperature is a critical parameter for the etching profile. In case of a setpoint of T = −75° C, the passivation is not stable at all for this array section. Using artifacts an etch rate of 114 nm/min was measured. With decreasing temperature, two main characteristics are obvious. Firstly, the anisotropy

$$ A = 1 {-}(R_{\text{hor}} / R_{\text{ver}} ) $$
(1)
Fig. 6
figure 6

Parametric sweeps of temperature, oxygen supply and HF-power at a series of equally prepared samples for evaluation of etch rate and anisotropy

Full size image

increases. R hor is the horizontally orientated etching at the bottom, i.e. for an undercut R hor is positive causing anisotropy <1. The etched depth perpendicular to the bottom is represented by R ver. At T = −95° C, the sidewalls are almost perpendicular to the bottom surface. 10 K below that, the sidewall appears to be perfectly formed. By characterizing SEM images, values in the range of A = 1 are calculated as compared in Table 1. Secondly, the overall etching rate decreases. Values for the rate are also given by Table 1. Parameters for these processes are collected in Table 2.

Table 1 Comparison of anisotropic behavior and etch rate by samples etched at different cryogenic temperatures, oxygen gas amounts and various HF-power settings according to the recipes of Table 2
Full size table
Table 2 Parameters for dry etching in comparison for the different series of parameter sweeping
Full size table

The second key parameter for building up a passivation layer is the ratio of oxygen inside the gas mixture. Figure 6 also contains results for different amount of oxygen. An oxygen flow rate of 9 sccm ends up in a very low value of anisotropy, even at the low power parameters (Table 2). There are no wires for characterizing left at this sample. The increasing etch rate by increasing the available amount of oxygen during the process points out, that there is a lack of oxygen for small doses of that gas. The anisotropy has been found straight next to ~1 (Table 1). However, the best result for perpendicular sidewalls is achieved by the lowest tested amount of 11 sccm, which was sufficient to generate a stable layer for passivation. A third parameter, which affects both anisotropy and etch rate is the power of the high frequency amplifier (HF-power). Some results are also depicted in Fig. 6. At high values, the ions reach the sample with more kinetic energy. Therefore the physical contribution to etching is increasing in significance. As a result, additional kinetic energy accelerates the etching speed. The anisotropy has been found to slightly vary between 0.95 and 0.98, which allows high aspect ratios in either case. However, the NIL resist does not allow excessive etching durations. Aspect ratios above 10 have not been reached this way. Therefore, a hard mask is preferable for preparing high aspect ratios.

Such a mask has been used for preparation of the wires for resonant frequency detection. As already stated in the course of the description of NIL, an aluminum layer with a thickness of 30 nm is suitable to be used as a hard mask. Etching of those samples prepared by NIL and lift-off has been performed as follows. Plasma power is 140 W and high-frequency power is 4.8 W. With a flow rate of 15 sccm O2 and 129 sccm SF6 the reactive atmosphere is established. The pressure inside the reactor is set to 0.5 Pa. For the process the wafer is cooled down to −95 °C. A wire height of 4.8 μm is achieved by an etching duration of 40 min. The ions are accelerated by a voltage of −31.5 V.

3 Characterization of nanowire resonators

3.1 Preparation for resonant frequency measurements

The different types of prepared SiNWs have to be characterized as resonators in a resonant frequency measurement. Experiments are performed inside a SEM chamber (~10−6–10−7 kPa). The samples comprising SiNWs are placed on a shear piezoactuator by a conductive carbon tape. Elastic cables build the electrical connection to the piezo. They are plugged into a bridge connector at the chamber wall. Mechanical vibration of the piezo is excited by a waveform generator (HP 33120A) by applying an AC sinus voltage Vpp of 10 V at frequencies up to 15 MHz. The generator is adjusted manually or by a LabView program, respectively. Inside the SEM the resonance response can be simultaneously detected, while the frequency is swept by the generator.

4 Results and discussion

For resonance characterization of wires made by conventional UV photolithography, a wire at a diameter of ~650 nm and height of ~40 μm has been checked for resonant response. Part of an array of such resonators is represented by Fig. 7a. A magnification around the tested wire is given by Fig. 7b. Figure 7c and d shows the wire resonating next to its resonant frequency. By the top view (Fig. 7d), the amplitude is detectable directly, i.e. calculations taking the viewing angle into account are not needed. Maximal displacement in the range of ~10 μm shows up at the resonant frequency. Figure 7e shows the graph of displacement while sweeping frequency. The bare SiNW resonates at 452.22 kHz. The quality factor calculated from the Lorentz-Fit is Q = 1,222. Shape and dimensions of the wire affects this factor. As a result of sacrificial thermal oxidation, the wire’s geometry differs from an ideal cylindrical shape. It appears to be slightly conical. By its low mass of ~30 pg it can be a potential alternative to previous tested MEMS-based piezoresistive cantilever resonators at a mass of 2.2 μg (Wasisto et al. 2012).

Fig. 7
figure 7

Frequency response of a wire prepared by sacrificial oxidation stripping beginning with an array of wires shown at side view (30° tilted) (a), showing a close-up (b) and SEM-detectable resonant vibration from side view (b) and top view (c) ending up in a graph relating displacement to accelerated frequency (e)

Full size image

The theoretical resonant frequency f res of a wire is defined by

$$ f_{res} = \frac{{d \cdot \beta_{0}^{2} }}{{2\pi \cdot h^{2} }} \cdot \sqrt {\frac{{E_{mat} }}{{16 \cdot \rho_{mat} }}} $$
(2)

wherein d is the wire diameter, β 0 = 1.875 the coefficient for resonating in the fundamental mode, h is defined as the wire height and E mat and ρ mat are the Young’s modulus and the density of the material, respectively (Fujita et al. 2001). By Eq. (2), the resonant frequency of the previous described wire is at 423.927 kHz, assuming a Young’s modulus of 130 GPa and a density of 2336 kg/m3. Thus, the theoretical value is by 28.3 kHz below the measured result. The difference is explained by the geometry deviation between design and real shape.

The NIL-patterned wires at a diameter of 300 nm would have a theoretical resonant frequency below ~348 kHz for aspect-ratios above 100. Thus, etching with resist masking is not feasible. Furthermore, due to passivation stability relying on temperatures below −75 °C high aspect-ratios is tremendously challenging. However, wires resonating in the low MHz regime have been prepared by etching at anisotropy of ~0.97. By an aluminum masking of 30 nm as described before, wires of 4.8 μm in length have been prepared. Their inverted conical shape ends up in a diameter of ~130 nm next to the anchor at the bottom. The resonant frequency is additionally reduced by vapor deposition of aluminum on the top of the wires of 300 nm in thickness. Due to a slight angle disalignment while preparation, aluminum grains are also deposited on the wire sidewalls (Fig. 8a). The wire shadowed area for metal deposition is visible in Fig. 8c at the left hand side of the wire. The resonant frequency is swept using the same equipment as written before. Using Eq. (2) for the given diameter at the narrowest position of 130 nm and the height h = 4.8 μm, the theoretical resonant frequency of an ideally cylindrical wire is at f res ≈ 5.9 MHz. However, due the additional weight caused by the inverse conical shape of the wire as well as the aluminum on top a lower resonant frequency can be expected. The waveform generator has been controlled by a LabView routine, sweeping the frequency of excitation in steps of 100 Hz upwards each second. The response of the wire is recorded in a video for evaluation. Figure 8b and d represent the responses next to the resonant frequencies for motion in direction alongside the metal free cavity at the bottom and perpendicular to it, respectively. A non-symmetrical resonance peak is observed which is attributed to the hardening effect well-known for nonlinear spring-mass systems (Mahmoodi et al. 2009). In Fig. 8e and f, the amplitude of the vibration is shown for the respective frequency range.

Fig. 8
figure 8

Frequency response of a NIL and ICP–DRIE prepared SiNW (a, 30° tilted, c) showing two different fundamental modes at orthogonal directions due to aluminum deposition (b, d) by relating amplitude to acceleration frequency (e, f)

Full size image

5 Conclusion

Top-down approaches had been applied to fabricate SiNWs in silicon. Conventional UV-based photolithography and Soft-UV-NIL had been utilized as prepatterning methods. In both cases ICP–DRIE at cryogenic temperatures had been subsequently etched SiNW at high aspect-ratios, i.e. 39.7 and 10, respectively. Further aiding treatment by sacrificial thermal oxidation allowed aspect-ratios >60. Influences of critical parameters while etching have been discussed with an eye to anisotropy and etch rate. Best results for anisotropy have been found with A = 0.99, i.e. a faint undercut of wires. Resonant responses have been found for the different SiNWs in the kHz and MHz regime and have been tracked by video recording of the SEM-images during automated frequency sweeping. As conclusion, due to their resonant behavior and small masses those resonators are promising for utilization as NP-mass sensitive sensors.