Transfer printing of nanomaterials and microstructures using a wire bonder

Wire bonding is the most prevalent and mature back-end technology for establishing interconnections between the die and the substrate. Typically, industrial wire bonders can easily achieve high throughput of about 20 bond wires per second, high placement accuracy of less than 3 $\mu$m, and low cost of 2 USD per 1000 bond wires [30, 31]. The majority of the wire bonded connections are realized using ball stitch wire bonding. A conventional ball stitch wire bonding cycle starts with the formation of a free air ball (FAB) using an electronic flame off (EFO). Next, a ball bond is performed on the bond pad of the fabricated die, followed by stitch bond performed on the bond pad of the hosting substrate, thereby forming a wire connection between the two bond pads. After this, the wire is ripped off from the stitch bond and ready for the next cycle of ball stitch wire bonding. In our experiments, we used a commercial and non-modified high-speed automated wire bonder (ESEC 3100+, ESEC Ltd., Switzerland) together with Au wires (25/50 $\mu$m diameter) in an unconventional 'pick-and-place' fashion, to demonstrate high-throughput transfer and heterogeneous integration of SWCNTs and Si microstructures to a new host substrate. The 'ball bond' and 'stitch bond' operations were used for either pick-up or placement of the CNTs or Si microstructures, instead of forming wire bonds. The substrate temperature during all the wire bonding operations was set to be 70 $^{\circ}\mathrm{C}$, thus making the transfer process a low-temperature process.

2.2.1. Preparation of the CNT patches.

2.2.2. Preparation of the PDMS and parylene host substrates.

2.2.3. Preparation of the Si host substrate for CNT field emission cathodes.

2.2.4. Preparation of the Si microstructures on an SOI source substrate and of the MHDA dry adhesive.

The approach of transfer printing CNTs from their source substrate to a target substrate by wire bonder is illustrated in figure 1. The transfer printing process consists of two successively performed standard wire bonding cycles. The first cycle starts with the formation of the FAB (200 $\mu$m diameter). Next, a dummy 'ball bond' (3500 mN, 50 ms) is performed on a silicon surface to flatten the bottom of the FAB, which will be subsequently used as a 'stamp' for the transfer printing. After the wire capillary is centered over a CNT patch, a dummy 'stitch bond' (1200 mN, 50 ms) is utilized for picking up the CNT patch. The porous nitrocellulose substrate has relatively weak adhesion to the CNT bundles, compared to the adhesion between the CNT bundles and the Au-stamp (van der Waals force) resulting from the close and large-area contact during the dummy 'stitch bond' process. This facilitates the pick-up of the CNT patch by the Au FAB, similar to the dry transfer mechanism demonstrated in previous work [33]. The second wire bonding cycle starts with preventing the FAB formation by passing the EFO current through a grounded wire adjacent to the target substrate, instead of affecting the previously generated FAB. Since the grounded wire is placed only in proximity to the target substrate and far from the Si substrate (figure 1(a)), it will not affect the FAB formation for the next CNT transfer cycle. Then, the placement of the CNT patch on the target substrate is achieved by a dummy 'ball bond' (100 mN, 100 ms) and the disposal of the used Au FAB is completed by a dummy 'stitch bond' (3000 mN, $35\%$ ultrasonic energy, 50 ms). After the second wire bonding cycle, a new wire is generated and ready for the next transfer printing cycle. This is to ensure the repeatability of the stamping effects on the CNT samples, by forming a clean stamp for each transfer cycle.

Zoom In Zoom Out Reset image size

The wire bonder-enabled transfer printing process was investigated on PDMS, parylene, and Au/parylene target substrates since these are commonly used substrates for flexible and wearable electronics. Another merit of these polymeric substrates is their good affinity to CNTs, especially when they are locally slightly deformed by the dummy 'ball bond', which results in enhanced contact with the CNTs and thus facilitates the placement of the CNT patches. It should be noted that after the placement of the CNT patches, some CNTs remain on the Au FAB due to their good adhesion to the FAB.

Among the above-mentioned dummy 'ball bond' and 'stitch bond' parameters, the critical ones are related to the pick-up and placement operations of the CNT patches, for realizing reliable transfer of the CNTs with minimal defects induced. Specifically, for picking up the CNT patches from the nitrocellulose substrate, initial tests revealed that bonding forces of 1000–1400 mN and bonding time of 50–100 ms yielded good and comparable results. For the placement of the CNT patches on the target polymeric substrates, a low bonding force of 100–200 mN and a bonding duration of 100 ms was a suitable combination. These parameters were selected to both ensure sufficient contact between the CNTs and the target substrates and, at the same time, to avoid large local deformation of the target polymeric substrates by the FABs, which might result in mechanically induced defects in the transferred CNTs.

The approach of transferring Au FABs (160 $\mu$m diameter) with stamped CNTs to an Au-coated Si substrate for realizing field emission sources is illustrated in figure 2. Field emitters are key components of various electronic devices, such as displays, ionization sources, and x-ray sources. The transfer process is realized by a single standard wire bonding cycle. The cycle starts with the formation of a FAB stamp, followed by a dummy 'ball bond' (1200 mN, 50 ms) on the source substrate to directly pick up the CNT patch. Then, a dummy 'stitch bond' (4000 mN, $25\%$ ultrasonic energy, 50 ms) places the Au FAB together with the CNT patch into a through-substrate hole of the target substrate. After this, a new wire is generated for the next transfer cycle. The field emission measurement setup is illustrated in figure 3. A high voltage was applied (AU-50P6-L, Matsusada Precision Inc., Japan) between the transferred CNT-Au cathode and a tungsten (W) anode. The emission current and the applied voltage were recorded using LabVIEW and digital multimeters (HP34401A).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The process of transfer printing Si microstructures from a source substrate to a target substrate is illustrated in figure 4. Here, the square Si microstructures serve as dummy dies for validating the transfer printing of microchips, but could potentially be replaced with III–V components and other more expensive materials. The transfer process constitutes two standard wire bonding cycles. The first cycle starts with the formation of a FAB (100 $\mu$m diameter). Next, a dummy 'ball bond' (500 mN, 50 ms) realizes the flattening of the FAB and the inking of the dry adhesive (MHDA) simultaneously. Then, a dummy 'stitch bond' (500 mN, 50 ms) achieves the pick-up of the Si die by breaking the narrow BOX supporting pillar on the SOI donor substrate (figure 4). The second wire bonding cycle begins with a dummy 'FAB formation' step by setting the EFO current to almost zero to avoid damage on the attached Si die. Then, a dummy 'ball bond' (100 mN, 100 ms) realizes the placement of the Si die on an adhesive layer (Nitto Denko Corp., Japan) on the target substrate, followed by disposal of the used FAB by a dummy 'stitch bond' (500 mN, $30\%$ ultrasonic energy, 50 ms). After this, a new wire is generated for the next cycle of die transfer. The adhesive layer can be peeled off from the target substrate, thus serving as a flexible polymeric substrate for hosting the Si dies.

Zoom In Zoom Out Reset image size

To demonstrate the viability and repeatability of the proposed transfer printing approach, we performed twenty cycles of the transfer process on PDMS, parylene, and Au/parylene substrates, respectively. The CNT patches before and after the transfer process are shown in figure 5. The CNT patches picked up by the deformed Au FABs are clearly visible in the scanning electron microscopic (SEM) images in figure 5(b), and were all successfully transferred onto PDMS, parylene, and Au/parylene substrates (figures 5(c)–(e)). For all the tested substrates, the achieved placement accuracy was within $\pm$4 $\mu$m, which is close to the specified alignment accuracy of the wire bonder ($\pm$3 $\mu$m, at $\pm$3$\sigma$).

Zoom In Zoom Out Reset image size

The quality of the transferred CNT layers was investigated by Raman spectroscopy (WITec Alpha 300R, 532 nm laser). A representative Raman spectrum of the pristine CNTs on the source substrate is shown in figure 6(a). The characteristic radial breathing mode (RBM) peaks, D peak, and G peaks of SWCNTs were all present. The clear RBM peaks below 200 cm⁻¹ revealed sub-2 nm diameters of the SWCNTs [34]. Raman spectra from five randomly selected locations exhibited very low ratios between the D peak and G peak intensities ranging from 0.017 to 0.023, indicating the high quality of the pristine CNTs on the source substrate [34]. Potential transfer-induced defects of CNTs on the PDMS and parylene host substrates were evaluated by such ratios. Figures 6(b) and (c) show the Raman spectra of the transferred CNTs from five randomly selected locations on the PDMS and parylene substrates, respectively. The resulting ratios between the D and G peak intensities were 0.027–0.040 and 0.063–0.121 for the CNTs on the PDMS and parylene substrates, respectively. Therefore, a slight increase of defects was induced in the transferred CNTs on the PDMS substrate, whereas more defects were present in the CNTs on the parylene substrates. However, the overall ratios between the D and G peak intensities in the transferred CNTs were still relatively low, indicating that our transfer method preserves the decent quality of the CNTs.

Zoom In Zoom Out Reset image size

A four-wire resistance measurement (2450 SourceMeter SMU, Keithley Instruments, USA) was conducted on the CNT patches transferred to the Au/parylene substrate to verify the contact between the CNTs and the Au electrodes. All the seven tested CNT patches on the Au/parylene substrate were found to be electrically conductive with an average measured resistance of 195$\pm$50 $\Omega$. The CNT transfer speed of approximately 0.35 s per CNT patch was realized without elaborate process optimization. The demonstrated CNT patches on the Au/parylene substrate could find applications in flexible electronics, for example as skin contact pads.

To verify the potential of using the wire bonder-enabled transfer printing for fabricating CNT-based field emission sources, we performed twenty transfer printing experiments. The transferred CNT patches together with the Au balls embedded in a Si substrate are shown in figure 7(b). A high yield of $90\%$ and a transfer speed of approximately 0.25 s per CNT/Au electrode were achieved without elaborate process optimization. The CNT patches were located 10 $\mu$m below the surface of the backside opening. By adjusting the backside etch depth, different CNT embedding depths with respect to the backside opening surface can be achieved, as illustrated in figure 7(a). Thus, the turn-on voltage of the field emission can be tuned accordingly due to different shielding effects. For example, for a cathode array with 2 $\times$ 2 CNT patches, a backside etch depth of 100 $\mu$m resulted in a CNT embedding depth of 110 $\mu$m and a turn-on voltage of  ∼2.6 kV (two tests). Instead, a backside etch depth of 200 $\mu$m resulted in a CNT embedding depth of 10 $\mu$m and a turn-on voltage of  ∼1 kV (six tests), as shown in figures 7(a) and (c). According to the Fowler–Nordheim (F–N) theory [35], the dependence of the field emission current on the local electrical field can be described using the following equation:

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle I={\alpha_{tip}}A{E_{local}^{2}}\exp{\left(-\frac{B}{E_{local}}\right)} \nonumber \end{align} \tag{ 1 }$$

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle A=\frac{{q_e}^{3}}{8{\pi}h{\phi}} \nonumber \end{align} \tag{ 2 }$$

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle B=\frac{8{\pi}{\sqrt{2m_e{\phi}^3}}}{3q_e{h}} \nonumber \end{align} \tag{ 3 }$$

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle E_{local}={\beta}U \nonumber \end{align} \tag{ 4 }$$

where $\alpha_{tip}$ is the emission area of the CNT emitters, E_local is the local electric field at the CNT emission tips, q_e is the electron charge, h is Planks constant, $\phi$ is the work function of carbon, m_e is the effective mass of an electron, and U is the applied voltage. $\beta$ is the local electric field enhancement factor, which reflects the local amplification effect on the electric field at the CNT emission tips due to the nanoscale sharp features of the CNTs. Transforming the above equation, the linear relationship between ln(I/U²) and 1/U can be expressed using:

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle \ln\left(1/{U^2}\right)=\left(-B/{\beta}\right)(1/U)+\ln\left({\alpha_{tip}}A{\beta}^2\right). \nonumber \end{align} \tag{ 5 }$$

Zoom In Zoom Out Reset image size

The typical measured emission data can be plotted (F–N plot) in this format, as shown in the inset of figure 7(c). The plot exhibits good linearity, indicating the emission data agrees with the F–N field emission model. From the plot, a high $\beta$ value of $8.55\times10^{6}$ m⁻¹ was extracted (using a carbon work function of 5 eV), demonstrating the significantly high field enhancement effect of the CNTs.

The achieved field enhancement factor and the emission current density of 38.2 mA cm⁻² of our transferred CNT emitters are comparable to previous work [36–38], especially for transferred CNTs [37], although the turn-on field and field enhancement factor are lower than that of vertically aligned CNTs that were directly grown by CVD [38]. The advantages of our method lie in the flexibility of the choice of the host substrate, which is very limited for directly grown CNTs, and in the preservation of the pristine CNTs as compared to drop casting methods. In addition, another potential benefit of our method is the possibility to bond the host substrate together with the embedded CNT emitters to target substrates for realizing vacuum sealed devices, using a hermetic sealing method demonstrated previously with similar wire bonded Au bumps [39].

The field emission stability was investigated (two tests) by applying a constant voltage and recording the emission current variations over time (figure 7(d)). Despite slight degradation, the CNT cathodes demonstrated a stable emission current of approximately 10 $\mu$A at 1560 V for over 12 min. The chamber pressure stayed constant at $1\times10^{-5}$ mbar during the entire measurement period. The demonstrated field emitters can be used as miniaturized electron sources for applications such as micro-ionization sources and x-ray sources.

To demonstrate the feasibility and reliability of applying the transfer printing approach on Si microstructures, we performed thirty transfer cycles on the prepared Si dies. For each transfer cycle, the Au FAB was deformed with a flattened top surface and MHDA attached due to a thiol-Au semi-covalent bond [40] (figure 8(a)). An MHDA self-assembled monolayer (SAM) can readily form on an Au surface, while in our case the attached MDHA could be a monolayer or multi-layers, as a result of excessive supply of the dried MHDA adhesive. The carboxyl tail groups of MHDA made the flattened Au top surfaces hydrophilic, thus improving the affinity between the flattened Au balls and the top native oxide of the Si dies. The Si dies prepared on the SOI substrate and residues of the BOX supporting pillars after the transfer can be seen in figures 8(b) and (c). Figure 8(d) shows an attached Si die on the flattened Au ball surface, demonstrating the feasibility of our transfer principle. The Si dies were finally transferred to an adhesive layer on a glass substrate with a high yield of 93.3%, due to stronger die-adhesive adhesion compared to die-Au ball affinity. An advantage of the presented transfer printing process is that even if the first transfer attempt failed, an additional transfer cycle could make up for the missing die on the target substrate. Right after transfer, the residues of the MHDA coating remained on the top surfaces of the Si dies. However, these residues were easily removed by rinsing in isopropanol (IPA), as depicted in figures 8(e) and (f).

Zoom In Zoom Out Reset image size

While the achieved placement accuracy of the Si dies (thirty samples) was very good (within $\pm$4 $\mu$m), there was a significant rotational misalignment ($\pm$30°) of the Si dies, as shown in figures 8(e) and (f). We believe this can be ascribed to the straining and relaxation of the Au wires during the wire bonding operations. The demonstrated transfer printing approach features a transfer speed of approximately 0.35 s per die, and can potentially be used to transfer light-emitting diodes (LEDs) from their fabrication substrate to a different host substrate.