A rolling-based printing approach for transferring arrays of patterned micro- and nano-structures directly from rigid fabrication substrates onto flexible substrates is presented. Transfer-printing experiments show that the new process can achieve high-yield and high-fidelity transfer of silicon nanomembrane components with diverse architectures to polyethylene terephthalate substrates over chip-scale areas (>1 × 1 cm2) in <0.3 s. The underlying mechanics of the process are investigated through finite element simulations of the contact and transfer process. These mechanics models provide guidance for controlling the contact area and strain in the flexible substrate during transfer, both of which are key for achieving reproducible and controlled component transfer over large areas.

The fabrication of electronic and photonic components, with critical dimensions ranging from hundreds of nanometers to micrometers to millimeters, on flexible substrates enables devices that provide sensing, communication, and information processing capabilities for a diverse range of consumer, military, industrial, and healthcare applications.1 Examples of emerging flexible devices include displays,2 solar cells,3 conformable sensors for robotics,4 structural-health monitoring,5 human-health monitoring,6–9 bio-integrated devices,6,10 and epidermal electronics.11 However, the mass production of high-performance flexible hybrid electronic devices based on high-quality inorganic semiconductors has been hindered by a lack of scalable manufacturing processes that allow for the integration of inorganic semiconductors on flexible substrates (see Ref. 12 for a recent review).

Here, we introduce a solution for high-yield fabrication of high-performance, flexible devices by combining established lithography (e.g., optical and nanoimprint lithography), and etching techniques with a new rolling-based direct-transfer (RBDT) process. In contrast to established multi-step stamp-based adhesive transfer techniques,13–15 the RBDT process enables direct transfer of components to flexible substrates with a continuous rolling motion. Direct pattern transfer to flexible substrates has been achieved for metal patterns on glass substrates using laser assistance (e.g., Ref. 16); however, the applicability for transferring high-performance semiconducting components or for roll-to-roll assembly has yet to be demonstrated. RBDT is the first example of the flexible substrate itself being used both as a rolling “stamp” and as the destination substrate for high-speed direct transfer printing. As a model system, we demonstrate the direct transfer of monocrystalline silicon nanomembranes (SiNMs) onto flexible substrates. The thinness, patternability, amenability to strain engineering, and stackability of SiNMs allow them to serve as a building block for flexible electronic or optoelectronic devices.17–19 We show the use of RBDT for high-fidelity and high-yield transfer of SiNM components to polyethylene terephthalate (PET) substrates in a process that is directly translatable to roll-to-roll assembly. The underlying mechanics of the process is investigated via finite element (FE) modeling to provide guidance on setting process parameters.

Fig. 1 presents a schematic diagram of the RBDT approach. An adhesive-coated flexible substrate (i.e., the web) is suspended above a sample that contains SiNMs to be transferred. The flexible substrate is deformed into contact with the components via a roller, and the components are then transferred by translating both the stage and the flexible substrate at the same rate with respect to the contacting roller. Success in the process requires engineering the contact such that components within the region of contact adhere uniformly, strongly, and damage-free to the adhesive coating on the flexible substrate, and remain adhered to the flexible substrate as the substrate translates away from the rolling contact.

FIG. 1.

Schematic of the rolling-based direct-transfer printing approach. A roller deforms an adhesive-coated flexible web into contact with patterned components supported on a donor wafer below. By translating the stage and web during contact, components are transferred directly to the flexible substrate with a continuous rolling motion.

FIG. 1.

Schematic of the rolling-based direct-transfer printing approach. A roller deforms an adhesive-coated flexible web into contact with patterned components supported on a donor wafer below. By translating the stage and web during contact, components are transferred directly to the flexible substrate with a continuous rolling motion.

Close modal

The concept in Fig. 1 was realized in a benchtop apparatus (Fig. 2) with programmable control of the operating conditions, including a closed-loop system for controlling and measuring contact forces, motorized stage positioning, a pulley system connected to the stage for driving the web, and a digital camera for process observation. The system allows the web angle and web tension to be adjusted over a wide range. The contacting rollers, 38 mm diameter, were fabricated by molding polydimethylsiloxane (PDMS) (10:1 Dow Corning SYLGARD® 184) into cylindrical molds.

FIG. 2.

Image of the RBDT apparatus, with the roller, web, and pulley system highlighted for clarity. The inset shows a view of a custom PDMS contacting roller.

FIG. 2.

Image of the RBDT apparatus, with the roller, web, and pulley system highlighted for clarity. The inset shows a view of a custom PDMS contacting roller.

Close modal

RBDT printing tests were conducted to quantify transfer yield as a function of contact force, web tension, and transfer speed. In all tests, the flexible substrates (i.e., the web) were 50 μm thick PET sheets (Grafix Plastics) that were 5 cm in width. Norland Optical Adhesive 61 (Norland Products, Inc.) was used as the adhesive coating on the PET substrates. After transfer, UV exposure was used to cure the adhesive layer fully and increase its stiffness to close to that of the underlying substrate.

The samples that were transferred consisted of 224 rectangular-shaped arrays of SiNMs that were 300 nm thick and had lateral dimensions ranging from single micrometers up to millimeters (see supplementary material). They were fabricated on 1.2 × 1.2 cm2 chips from a silicon-on-insulator (SOI) wafer with a 300 nm crystalline-Si device layer and a 400 nm thick buried oxide using conventional optical lithography and reactive ion etching. The NM arrays were released in place via etching the buried oxide layer with hydrofluoric acid (49%). After removal of the oxide, the NMs became adhered to the Si handle wafer below due to weak secondary bonds (i.e., van der Waals forces and hydrogen bonding).

The web tension needs to be high enough to prevent the web from remaining adhered to the chip containing the SiNMs and low enough to prevent NM wrinkling/delamination upon removal of the tension in the web. Optical images of the transferred components indicated that keeping the maximum strain in the web ≲ 1% was sufficient to avoid NM wrinkling (see Sec. IV for determination of web strain), which, for the aforementioned apparatus configuration, corresponded to a web tension per unit width of the web, F′w, of <0.17 N/mm. To determine an appropriate contact force for comprehensive studies of the effect of transfer speed on printing yield, a series of transfer tests was performed over the contact force range 2N < Fcontact < 8 N (corresponding to a contact force per unit length of the contact, F′contact, of 0.167–0.667 N/mm), with F′w = 0.11 N/mm and a transfer speed of 19 mm/s. At a contact force of 0.33 N/mm, the width of the contact zone between the web and the SiNM chip is approximately twice the lateral size of the largest NM components (see Sec. IV for determination of contact size), which proved to be sufficient for establishing full contact between the web and the underlying components within the contact zone. Because the quality (e.g., lack of distortion) and yield of the components using the 0.33 N/mm contact force were high, that force was chosen for the speed studies.

The transfer yield and transfer quality were characterized as a function of transfer speed. To gain insight into the dependence of SiNM frame architecture on transfer yield, the SiNM sample was divided into seven sections, with each of the seven sections (denoted I to VII) consisting of 32 structures featuring rectangular NM strips linked to one another through a common frame (henceforth referred to simply as a NM frame—see images (a) and (b) in Fig. 4 for optical images of frames). The frames in each section exhibited the same values for the width of the strips and the spacing between them (see supplementary material), but ranged in strip number from two to twenty and in strip length from 25 μm to 1.5 mm. The ratio of the strip width to the spacing between the strips, henceforth referred to as the spacing ratio, for each section is listed in Table I. NM transfer yield was quantified for each section by counting the number of frames transferred successfully to the PET web. Two different levels of web tension (F′w = 0.11 N/mm and F′w = 0.04 N/mm) were used, and similar yield results were obtained for both. The average transfer yield (averaged over all sections) is plotted in Fig. 3. For both values of web tension, transfer yield was found to be the highest for a transfer speed of 40 mm/s (0.3 s transfer time for the 1.2 × 1.2 cm2 chip), with an average yield of ∼90% averaged over all sections.

TABLE I.

Results for transfer yield for each section of the NM sample, averaged over speeds >2 mm/s.

SectionIIIIIIIVVVIVII
NM width (μm) 10 10 10 
Spacing ratio (NM width: NM spacing) 1:1 1:2 1:4 1:1 1:2 1:4 1:10 
Yield (%), averaged over speeds >2 mm/s 91.5 90.6 87.1 79.9 81.3 77.7 75.0 
SectionIIIIIIIVVVIVII
NM width (μm) 10 10 10 
Spacing ratio (NM width: NM spacing) 1:1 1:2 1:4 1:1 1:2 1:4 1:10 
Yield (%), averaged over speeds >2 mm/s 91.5 90.6 87.1 79.9 81.3 77.7 75.0 
FIG. 3.

Transfer yield as a function of transfer speed at two different web tensions. The contact force was 4 N (F′contact = 0.33 N/mm) for all tests.

FIG. 3.

Transfer yield as a function of transfer speed at two different web tensions. The contact force was 4 N (F′contact = 0.33 N/mm) for all tests.

Close modal

It is apparent that transfer yield is relatively poor at lower transfer speeds (<40 mm/s). We attribute the poor performance at low speeds to the rate-dependent properties of the polymer-based optical adhesive, with adhesive strength decreasing at lower peeling rates (e.g., analogous to that of Ref. 13, and exploited in the stamp-based transfer-printing work14,20–23). At speeds of 40 mm/s and higher, the average yield approaches 90%, with yields for several of the sections reaching 100%. Yields did not increase at speeds of 50 mm/s and higher, a result that we attribute to a combination of (i) no significant improvement in the rate-dependent adhesion beyond a transfer speed of ∼40 mm/s; and (ii) lag in the force feedback loop, leading to fluctuations in the quality of the roller–substrate contact at higher rates.

The effect of frame architecture on transfer yield can be seen in the results of Table I. The results for transfer yield, excluding those for the 2 mm/s transfer speed because of the overall low yield, were averaged for each section to generate the data in Table I. Overall, the frames with the wider, more densely spaced NM strips (i.e., up to 1:1 spacing ratio) had the highest yields. With the slight exception of the 5–μm-wide strips with a 1:2 spacing ratio, there is a monotonic trend of decreasing yield as the NM width and spacing ratio get smaller. We attribute the high yields for the frames with the wider NMs and larger spacing ratios to the larger area of contact between the frames and the adhesive-coated PET during contact. This result has a direct implication for component-based device design prior to transfer—namely, with electronic/optoelectronic components that exhibit small feature sizes and small spacing ratios, incorporating larger-area frames or pads into the pattern design can enhance transfer yield significantly.

Fig. 4 shows a number of NM structures that were transferred successfully using F′contact = 0.33 N/mm, F′w = 0.11 N/mm, and a transfer speed of 50 mm/s (0.24 s transfer time for the 1.2 × 1.2 cm2 chip). The higher-magnification images reveal that the NM arrays are flat and were transferred damage-free to PET. Multiple cycles of bending the PET substrate with SiNMs on the surface around the contacting roller (dia. = 3.81 cm) confirmed that no delamination or wrinkling of the SiNMs occurred due to post-transfer mechanical deformation of the printed structures.

FIG. 4.

Optical micrographs of SiNM microstructures transferred to PET over a ∼1 × 1 cm2 area, with a contact force of 0.33 N/mm, a web tension of 0.11 N/mm, and a transfer speed of 50 mm/s. (a) NM frames with 1:10 spacing ratio. (b) NM frames with 1:1 spacing ratio. (c) Circular array of NM pads. (d) Chip-scale transfer of NM structures ranging in length from 25 μm to 1.5 mm, including those of (a)–(c).

FIG. 4.

Optical micrographs of SiNM microstructures transferred to PET over a ∼1 × 1 cm2 area, with a contact force of 0.33 N/mm, a web tension of 0.11 N/mm, and a transfer speed of 50 mm/s. (a) NM frames with 1:10 spacing ratio. (b) NM frames with 1:1 spacing ratio. (c) Circular array of NM pads. (d) Chip-scale transfer of NM structures ranging in length from 25 μm to 1.5 mm, including those of (a)–(c).

Close modal

The fidelity of the RBDT process for transferring the NMs onto PET without disturbing their arrangement was quantified by measuring the change in spacing between pairs of structures due to transfer. Optical images of the NMs before and after transfer, analyzed with a custom Matlab script for tracking the their positions with respect to one another, revealed sub-micron printing fidelity, with |Δx| = 0.5 ± 0.4 μm, |Δy| = 0.7 ± 0.7 μm, and |Δθ| = 0.1 ± 0.1° (N = 30).

To understand the mechanics that govern the RBDT process, parametric finite-element studies were performed to investigate the effects of the radius of the roller, the thickness of the web, the web tension, and the contact force on: (1) the width of the contact between the web and the substrate and (2) the strain in the web. The system was modeled using finite element (FE) analysis (COMSOL Multiphysics®, using the Structural Mechanics Module). The 2D symmetric model consisted of a compliant cylindrical roller, a thin elastic web with an elastic modulus that was high relative to that of the roller, and a rigid substrate. The Young's modulus and Poisson's ratio used for the roller (PDMS) were 2 MPa and 0.499, respectively, and 2 GPa and 0.4, respectively, for the web (PET). Tension applied to the web renders the influence of adhesive forces at the web–substrate interface negligible compared to the effect of external loading; therefore, adhesion was not included in the model to reduce computation time. Additional details regarding model validation and convergence can be found in the supplementary material. Fig. 5 shows an example case with a roller diameter and web thickness of 3.81 cm and 100 μm, respectively. The length of the contact (i.e., the length in the z-direction) is 50.8 mm. The roller has a contact force of 0.104 N/mm applied to its center axle, and a web tension of 0.054 N/mm is applied to the web in the vertical direction to create a web angle of 90°. The insets in Fig. 5 show an enlarged view of the end of the web (red box) and a zoom-in of the contact region (green box). The half-width, a, of the contact, defined as half of the width of the contact area between the PET and the underlying rigid substrate, is illustrated in the lower inset of Fig. 5 (green box).

FIG. 5.

Schematic diagram of the FE model used to examine the contact mechanics of the RBDT process.

FIG. 5.

Schematic diagram of the FE model used to examine the contact mechanics of the RBDT process.

Close modal

For the parametric studies, the radius of the roller was varied from ∼1 cm to ∼2 cm, the thickness of the web was varied from 50 μm to 150 μm, the web tension was varied from 0.054 N/mm to 0.162 N/mm, and the contact force was varied from 0 N/mm to 0.492 N/mm. Matlab scripts were written to process the exported FE files in order to determine the precise values of the width of the contact and the maximum strain at the outer surface of the flexible web. To be consistent with other contact analyses performed in the literature, the size of the contact is quantified as the contact half-width, defined as half of the width of the contact zone between the PET and the underlying rigid substrate.

FE studies were used to establish a general analytical expression for the contact half-width as a function of the loading, geometry, and elastic properties of the system. Based on the analysis performed in Ref. 24, the following non-dimensional parameters were chosen:

a*=araFEaraw,
(1)
h*=haFE,
(2)

with

ar=4RFcontactπEr*,
(3)
aw=4RFcontactπEw*,
(4)

and

1R=1Rr+h+1Rs,
(5)
1Er*=1vr2Er+1vs2Es,
(6)
1Ew*=1vw2Ew+1vs2Es,
(7)
Fcontact=FcontactL,
(8)

where a* is the normalized contact half-width, h* is the normalized web thickness, and aFE is the contact half-width determined from the FE simulations. h is the thickness of the web, R is the radius, E is the elastic modulus, ν is the Poisson's ratio, Fcontact is the contact force, L is the contact length, and the subscripts r, w, and s denote the roller, the web, and the substrate, respectively. It should be noted that over the range of relatively low web tensions investigated here, web tension had a negligible effect on the contact radius; hence, Fw was not included in the normalized quantities above.

Fig. 6(a) is a non-dimensional plot of a* versus h* from the FE results. The FE results collapse onto a single curve, indicating that the normalization (Eqs. (1) and (2)) is appropriate. The curve approaching 1 as h* increases reflects the fact that when aFE becomes small (i.e., h* becomes large), the prediction for the contact half-width of the compliant roller, ar, becomes large compared to aFE. A nonlinear least-squares fit was performed on the FE data using the following equation:

a*=1αeβh*.
(9)

The best-fit values for the fitting coefficients were α = 1.4169 and β = 2.5881. The plot of Fig. 6(b) presents the error between the fit and the FE results. The error is <5% for the range of h* tested, demonstrating that Eq. (9) can be used to describe the finite element results.

FIG. 6.

(a) Plot of normalized contact half-width (a*) versus normalized web thickness (h*). (b) Error plot from the fit shown in (a).

FIG. 6.

(a) Plot of normalized contact half-width (a*) versus normalized web thickness (h*). (b) Error plot from the fit shown in (a).

Close modal

Given the geometry and elastic properties of the system, Eq. (9) can be used to determine the contact force required to achieve a targeted contact half-width during transfer. The procedure for determining the force to apply during operation of an RBDT apparatus is as follows. First, calculate the constants ρ and ω as:

ρ=4RπEr*,
(10)
ω=4RπEw*,
(11)

with R and E defined as above. Then, calculate h* and C as

h*=hatarget,
(12)
C=1αeβh*,
(13)

where atarget is the targeted contact half-width. The contact force required to achieve a targeted contact half-width is then calculated as

Fcontact=L(atargetρ(ρω)C)2.
(14)

The average contact pressure, pavg, at the web-substrate interface can be calculated as

pavg=Fcontact2Latarget.
(15)

Equations (14) and (15) can therefore be used to determine the force required to achieve either a targeted contact half-width or a desired average pressure during tool operation, depending on the sizes of the components being transferred and any restrictions on pressure within the contact (e.g., to avoid embedding the components into an adhesive layer), respectively. It should be noted that if an adhesive coating with elastic properties that are comparable to those of the web is present on the outer surface of the web, then the value for h should be replaced with h + hadhesive. If the adhesive coating has an appreciable thickness and has elastic properties that differ significantly from those of the web, then again h should be replaced with h + hadhesive; however, atarget would then be a lower bound if the adhesive is more compliant, and an upper bound if the adhesive is stiffer, than the web material. The degree to which a mismatch in the elastic properties of the web and adhesive affects the contact half-width was not investigated rigorously here.

FE studies were also performed to determine a general expression for the maximum strain at the outer surface of the flexible web as a function of the loading and the geometric and elastic properties of the system. In the theoretical case in which the roller does not undergo deformation, the equation for the maximum strain in the web is

εmax,th=εbending,th+εstretching,th,
(16)

with

εbending,th=h2(Rr+h2),
(17)
εstretching,th=Fw(1vw2)Ewhb,
(18)

where b is the width of the web. With elastic deformation of the roller, the Rr term in Eq. (16) needs to be modified to reflect the actual local radius of curvature, not merely the radius of the roller, as the roller gets compressed under an applied load

εbending,act=h2Rmin.
(19)

Therefore, an analytical expression is needed for Rmin, where Rmin is the minimum radius of curvature that the web experiences when the roller deforms. Eq. (18) can remain unchanged.

The FE results were analyzed to determine Rmin numerically. The following non-dimensional parameter was chosen to normalize Rmin:

R*=RminRr+h.
(20)

Fig. 7(a) presents the results for R* as a function of h*. The FE results collapse onto a single curve, indicating that the normalization (Eqs. (2) and (20)) was appropriate.

FIG. 7.

(a) Non-dimensional plot of normalized bending radius (R*) of the web versus normalized web thickness (h*). The inset in (a) shows a strain map, revealing the location of the maximum web strain to be outside of the contact zone (h = 75 μm, F′contact≈ 0.433 N/mm, F′w = 0.162 N/mm). (b) Error plot from the fit shown in (a).

FIG. 7.

(a) Non-dimensional plot of normalized bending radius (R*) of the web versus normalized web thickness (h*). The inset in (a) shows a strain map, revealing the location of the maximum web strain to be outside of the contact zone (h = 75 μm, F′contact≈ 0.433 N/mm, F′w = 0.162 N/mm). (b) Error plot from the fit shown in (a).

Close modal

A non-linear least-squares fit was performed on the FE data using the following equation:

R*=1γh*δ.
(21)

The best-fit values for the fitting coefficients were γ = 0.114 and δ = −0.893. The plot of Fig. 7(b) presents the error between the fit and the FE data. The error is <3%, demonstrating good agreement between Eq. (21) and the finite-element results.

Given the loading and the geometric and elastic properties of the system, Eq. (21) can be used in combination with Eqs. (18) and (19) to determine the maximum strain in the web. The procedure for calculating εmax is as follows. First, calculate Rmin as:

Rmin=(Rr+h)(1γh*δ),
(22)

with h* defined in Eq. (12). Then, calculate εmax as

εmax,act=h2Rmin+Fw(1vw2)Ewhb.
(23)

With atarget prescribed, Eqs. (22) and (23) can therefore be used to determine the maximum strain in the web as a function of the web tension, Fw. It is worth noting that because the web lies flat (i.e., no bending) within the contact zone, the second term in Eq. (23), the term representing the theoretical strain solely due to stretching can be used to approximate the strain in the web within the contact zone during transfer.

To visualize the strain near the contact zone, the inset of Fig. 7(a) presents a map of the εxx strain distribution under load (h = 75 μm, F′contact ≈ 0.433 N/mm, F′w = 0.162 N/mm). The strain map reveals that the location of the maximum strain in the web is close to but outside of the contact zone. The implication for the transfer performance of the tool is that all transferred components will undergo a strain cycle caused by the variation of the strain in the underlying web from εstretching, thεmax, actlεmax, thεstretching, th → 0 as they travel from within the contact zone → just beyond the contact zone → along the contacting roller → down the assembly line → removal from the assembly line. The degree to which the strain in the web imparts strain to the components will depend on the geometric and mechanical properties of the components, which have been analyzed and discussed elsewhere (e.g., Ref. 25).

We have demonstrated a new rolling-based direct-transfer process and characterized its performance. The process allows for high-yield, high-fidelity transfer of NM-based components with diverse architectures to flexible substrates over >1 × 1 cm2 areas. A model-based prescription for controlling the contact size and strain in the flexible web is presented to provide an analytical framework for selecting parameters in the process. This versatile RBDT approach can be used to transfer a wide range of components to flexible substrates.

See supplementary material for details regarding the arrays of SiNMs that were used for evaluating transfer performance.

This work was supported by an STTR award to systeMECH, LLC (subcontractor: U. Wisconsin-Madison) from the Air Force under Contract FA9550-13-C-0041 (Program Manager: Gernot Pomrenke). The work used NSF-supported shared facilities at the University of Wisconsin-Madison. D.G. wishes to thank Erick Oberstar for technical assistance with motion control and electronics instrumentation.

1.
Y.
Zhan
,
Y.
Mei
, and
L.
Zheng
,
J. Mater. Chem. C
2
,
1220
(
2014
).
2.
G. P.
Crawford
,
Inf. Disp.
21
,
10
(
2005
); ISSN: 03620972.
3.
Y.
Jongseung
,
A. J.
Baca
,
P.
Sang-Il
,
P.
Elvikis
,
J. B.
Geddes
 III
,
L.
Lanfang
,
K.
Rak Hwan
,
X.
Jianliang
,
W.
Shuodao
,
K.
Tae-Ho
,
M. J.
Motala
,
A.
Bok Yeop
,
E. B.
Duoss
,
J. A.
Lewis
,
R. G.
Nuzzo
,
P. M.
Ferreira
,
H.
Yonggang
,
A.
Rockett
, and
J. A.
Rogers
,
Nat. Mater.
7
,
907
(
2008
).
4.
S. C. B.
Mannsfeld
,
B. C. K.
Tee
,
R. M.
Stoltenberg
,
C. V. H. H.
Chen
,
S.
Barman
,
B. V. O.
Muir
,
A. N.
Sokolov
,
C.
Reese
, and
B.
Zhenan
,
Nat. Mater.
9
,
859
(
2010
).
5.
A.
Nathan
,
B.-K.
Park
,
Q.
Ma
,
A.
Sazonov
, and
J. A.
Rowlands
,
Microelectron. Reliab.
42
,
735
(
2002
).
6.
D.-H.
Kim
,
N.
Lu
,
R.
Ghaffari
,
Y.-S.
Kim
,
S. P.
Lee
,
L.
Xu
,
J.
Wu
,
R.-H.
Kim
,
J.
Song
,
Z.
Liu
,
J.
Viventi
,
B.
De Graff
,
B.
Elolampi
,
M.
Mansour
,
M. J.
Slepian
,
S.
Hwang
,
J. D.
Moss
,
S.-M.
Won
,
Y.
Huang
,
B.
Litt
, and
J. A.
Rogers
,
Nat. Mater.
10
,
316
(
2011
).
7.
S.
Ryu
,
P.
Lee
,
J. B.
Chou
,
R.
Xu
,
R.
Zhao
,
A. J.
Hart
, and
S.-G.
Kim
,
ACS Nano
9
,
5929
(
2015
).
8.
C.
Pang
,
G.-Y.
Lee
,
T.-I.
Kim
,
S. M.
Kim
,
H. N.
Kim
,
S.-H.
Ahn
, and
K.-Y.
Suh
,
Nat. Mater.
11
,
795
(
2012
).
9.
G.
Schwartz
,
B. C. K.
Tee
,
J.
Mei
,
A. L.
Appleton
,
D. H.
Kim
,
H.
Wang
, and
Z.
Bao
,
Nat. Commun.
4
,
1859
(
2013
).
10.
K.
Dae-Hyeong
,
J.
Viventi
,
J. J.
Amsden
,
X.
Jianliang
,
L.
Vigeland
,
K.
Yun-Soung
,
J. A.
Blanco
,
B.
Panilaitis
,
E. S.
Frechette
,
D.
Contreras
,
D. L.
Kaplan
,
F. G.
Omenetto
,
H.
Yonggang
,
H.
Keh-Chih
,
M. R.
Zakin
,
B.
Litt
, and
J. A.
Rogers
,
Nat. Mater.
9
,
511
(
2010
).
11.
K.
Dae-Hyeong
,
L.
Nanshu
,
M.
Rui
,
K.
Yun-Soung
,
K.
Rak-Hwan
,
W.
Shuodao
,
W.
Jian
,
W.
Sang Min
,
T.
Hu
,
A.
Islam
,
Y.
Ki Jun
,
K.
Tae-il
,
R.
Chowdhury
,
Y.
Ming
,
X.
Lizhi
,
L.
Ming
,
C.
Hyun-Joong
,
K.
Hohyun
,
M.
McCormick
,
L.
Ping
,
Z.
Yong-Wei
,
F. G.
Omenetto
,
H.
Yonggang
,
T.
Coleman
, and
J. A.
Rogers
,
Science
333
,
838
(
2011
).
12.
S.
Khan
,
L.
Lorenzelli
, and
R. S.
Dahiya
,
IEEE Sens. J.
15
,
3164
(
2015
).
13.
M. A.
Meitl
,
Z.-T.
Zhu
,
V.
Kumar
,
K. J.
Lee
,
X.
Feng
,
Y. Y.
Huang
,
I.
Adesida
,
R. G.
Nuzzo
, and
J. A.
Rogers
,
Nat. Mater.
5
,
33
(
2006
).
14.
T.-H.
Kim
,
A.
Carlson
,
J.-H.
Ahn
,
S. M.
Won
,
S.
Wang
,
Y.
Huang
, and
J. A.
Rogers
,
Appl. Phys. Lett.
94
,
113502
(
2009
).
15.
M. B.
Tucker
,
D. R.
Hines
, and
T.
Li
,
J. Appl. Phys.
106
,
103504
(
2009
).
16.
B.
Kang
,
J.
Yun
,
S.-G.
Kim
, and
M.
Yang
,
Small
9
,
2111
(
2013
).
17.
F.
Cavallo
and
M. G.
Lagally
,
Soft Matter
6
,
439
(
2010
).
18.
H. C.
Ko
,
G.
Shin
,
S.
Wang
,
M. P.
Stoykovich
,
J. W.
Lee
,
D.-H.
Kim
,
J. S.
Ha
,
Y.
Huang
,
K.-C.
Hwang
, and
J. A.
Rogers
,
Small
5
,
2703
(
2009
).
19.
J. A.
Rogers
,
M. G.
Lagally
, and
R. G.
Nuzzo
,
Nature
477
,
45
(
2011
).
20.
A.
Carlson
,
H.-J.
Kim-Lee
,
J.
Wu
,
P.
Elvikis
,
H.
Cheng
,
A.
Kovalsky
,
S.
Elgan
,
Q.
Yu
,
P. M.
Ferreira
,
Y.
Huang
,
K. T.
Turner
, and
J. A.
Rogers
,
Appl. Phys. Lett.
98
,
264104
(
2011
).
21.
H.
Cheng
,
J.
Wu
,
Q.
Yu
,
H.-J.
Kim-Lee
,
A.
Carlson
,
K. T.
Turner
,
K.-C.
Hwang
,
Y.
Huang
, and
J. A.
Rogers
,
Mech. Res. Commun.
43
,
46
(
2012
).
22.
H. J.
Kim-Lee
,
A.
Carlson
,
D. S.
Grierson
,
J. A.
Rogers
, and
K. T.
Turner
,
J. Appl. Phys.
115
,
143513
(
2014
).
23.
S.
Kim
,
A.
Carlson
,
H.
Cheng
,
S.
Lee
,
J.-K.
Park
,
Y.
Huang
, and
J. A.
Rogers
,
Appl. Phys. Lett.
100
,
211904
(
2012
).
24.
S.
Liu
,
A.
Peyronnel
,
Q. J.
Wang
, and
L. M.
Keer
,
Tribol. Lett.
18
,
505
(
2005
).
25.
J.
Hu
,
L.
Li
,
H.
Lin
,
P.
Zhang
,
W.
Zhou
, and
Z.
Ma
,
Opt. Mater. Express
3
,
1313
(
2013
).

Supplementary Material