Applications of Excimer Laser in Nanofabrication

Appl Phys A (2010) 98: 9–59 DOI 10.1007/s00339-009-5455-0

INVITED PAPER

Applications of excimer laser in nanofabrication

Qiangfei Xia · Stephen Y. Chou

Received: 7 October 2009 / Accepted: 15 October 2009 / Published online: 5 November 2009 © Springer-Verlag 2009

Abstract This paper addresses novel applications of an nanoparticles. Monolayers of particles are fabricated on var- excimer laser (308 nm wavelength, 20 ns pulse duration) ious substrates (silicon, fused silica and plastics) by expos- in nanofabrication. Specifically, laser assisted nanoimprint ing thin metal films to a single laser pulse. Periodic nanopar- lithography (LAN), self-perfection by liquefaction (SPEL), ticle arrays have been fabricated by fragmentation of metal fabrication of metal nanoparticle arrays, and the fabrication grating lines. The periodicity of these nanoparticles can be of sub-10-nm nanofluidic channels are covered. In LAN, a regulated by surface topography such as shallow trenches. polymeric resist is melted by the laser pulse, and then im- Finally, an excimer laser pulse has been used to melt the top printed with a fused silica mold within 200 ns. LAN has portion of 1D and 2D Si gratings to seal off the top surface, been demonstrated in patterning various polymer nanos- forming enclosed nanofluidic channel arrays. The channel tructures on different substrates with high fidelity and uni- width has been further reduced to 9 nm using self-limited formity, and negligible heat effect on both the mold and thermal oxidation. DNA stretching using 20 nm wide self- the substrate. SPEL is a novel technology that uses selec- sealed channels is also demonstrated. tive melting to remove fabrication defects in nanostructures post fabrication. Depending on the boundary conditions, PACS 06.60.Jn · 81.07.-b · 81.16.Rf · 81.16.Nd · 87.85.Rs SPEL is categorized into three basic types: Open-SPEL that takes place with surface open, Capped-SPEL where a cap plate holds the top surface of the nanostructures 1 Introduction and Guided-SPEL where a plate held a distance above the structure guides the molten materials to rise and form a Since its invention in 1960 [1, 2], laser has been widely used new structure with better profile. Using SPEL (in less than in many fields, such as scientific research, medical treat- 200 ns), we have achieved a reduction of line edge rough- ment, industry and military. Excimer lasers are an important ness (LER) of Cr lines to 1.5 nm (3σ ) (560% improve- family of gas lasers that were firstly commercialized in the ment from the original), which is well below what the pre- 1970s [3]. The most common excimer lasers use rare-gas vious technologies permit, and a dramatic increase of the monohalides such as KrF, XeCl, and have a wavelength in aspect ratio of a nanostructure. We have used SPEL to make ultraviolet (UV) range. sub-25-nm smooth cylindrical NIL pillar molds and smooth- The applications of excimer laser are very broad with ing Si waveguides. Excimer laser is also used to make metal new applications continuously coming out. This paper is not intended to give an overview on this expanding field but rather focuses on certain novel applications in nanofabrica- Q. Xia current address: Information and Quantum Systems Lab, tion that have been developed by the authors at Princeton Hewlett-Packard Laboratories, 1501 Page Mill Road, Palo Alto, CA 94304, USA. University as part of one of the author’s PhD thesis [4]. This paper is organized as follows: after a brief intro- · Q. Xia ( ) S.Y. Chou duction of our general experimental setup, we will discuss Nanostructure Laboratory, Department of Electrical Engineering, Princeton University, Princeton, NJ 08544, USA (1) laser assisted nanoimprint lithography (LAN), an ultra- e-mail: [email protected] fast imprint process that patterns nanostructures in polymers 10 Q. Xia, S.Y. Chou within 200 ns with negligible heating effect on both the mold imaging doublet, which consists of a set of objective lenses and substrate; (2) SPEL, which uses the excimer laser to re- that focuses the energy into a smaller spot so as to achieve move fabrication defects and to enhance the nanostructure the required energy intensity for melting different materials. profile; (3) the fabrication of metal nanoparticle monolayers An x–y–z stage is used to mount the samples. In most and periodic metal nanopaticle arrays using a single excimer of our experiments, the stage is moved manually although a laser pulse exposure in combination with nanoimprint litho- motor-driven stage can be used when precise alignment in graphy (NIL); and (4) the fabrication of sub-10-nm nanoflu- laser scanning is required. A continuous wave (cw) HeNe idic channels using laser melting and self-limiting thermal laser (λ = 633 nm, maximum output power 1 mW, beam oxidation, along with the application of the channels in diameter 0.5 mm) is used as a monitor laser. This monitor stretching λ-phage DNA molecules. laser serves several purposes: (1) To align the laser pulse and the sample; (2) To act as a light source for time-resolved reflectivity measurement and other in situ measurements as 2 Laser setup and molten time measurements described later in the text. One important parameter in laser melting is the molten We use a XeCl excimer laser in this study (Lambda Physik, time of a material for a given laser fluence. To measure the model number COMPex 102). The laser tube is filled to molten time, time-resolved reflectivity (TRR) measurements 3200 mbar with premixed gases. The gas composition (pres- were carried out for several semiconductors. The setup is sure percentage) is as follows: 0.13% HCl, 0.02% H2, 2.35% He, 1.87% Xe, and Ne as the balance. The maximum output showninFig.2. The HeNe laser beam is incident on a spot energy is 200 mJ/pulse and the maximum repetition rate is of the surface to be melted by the excimer laser. The re- 20 Hz. flected signal from the surface, which is much higher in Figure 1 shows the main experimental setup used through- molten material than its solid phase for semiconductors was out this study. After a laser pulse comes out of the laser tube, collected by a high speed Si p-i-n detector (Newport, 818- a beam delivery system (BDS) directs the laser beam to the BB-21A, rise time <500 ps, fall time <500 ps) with a band sample. This BDS consists of two mirrors, an attenuator, pass filter (Andover Corp, 633FS02-25, center wavelength a homogenizer, a condenser lens, a metal aperture, and an 633 nm, bandwidth 1.0±0.2 nm) mounted in front of the de- imaging doublet. A mirror at one end of the optical track tector. The signal is sent to an oscilloscope (Tektronix, TDS reflects the laser pulse coming out of the tube to an atten- 220, bandwidth 100 MHz, sample rate 1 GS/s) for analysis. uator, which is used to modulate the laser energy continu- The detector and oscilloscope together have a time resolu- ously from 10% to 90%. The homogenizer, together with tion better than 10 ns. the condenser, is used to generate a mesa-shaped spatial in- Figure 3 shows a typical reflectivity curve during melting tensity profile to achieve much more uniform distribution. silicon captured from the oscilloscope screen. In the solid The spatial plane of this intensity profile is called the ho- state, the reflectivity stays low and constant. When the UV mogenized plane. A metal aperture is mounted at the ho- laser pulse reaches the Si surface, it melts the surface layer mogenized plane to block off the edges of the beam and side immediately and the reflectivity jumps to a much higher lobes, resulting in a nearly flat top spatial intensity profile value within 1 ns [5, 6]. When the surface layer is at a molten for the laser pulse. Another mirror is placed at the other end state, the reflectivity stays constantly high before dropping of the optical track to direct the modified laser pulse to the to its previous level upon resolidification. Using this technique, the molten times for different materials at different excimer laser fluence have been measured. Figure 4 shows the molten time as a function of laser fluence for different types of semiconductors including a bare Si wafer (p-type, (100) orientation, resistivity 10–20 cm), an SOI wafer (SIMOX, 100 nm thick (100) silicon device layer and 380 nm thick buried oxide), a Ge wafer, and SiGe on Si or SiO2. The molten time increases with laser fluence for all cases. This is due to the fact that with a higher laser fluence (below the ablation threshold), a thicker layer is melted which needs a longer time to re-solidify. It is also interesting to note that under the same laser fluence, the molten time for SOI is Fig. 1 Excimer laser setup used for this paper. This setup includes an excimer laser, a beam delivery system, a cw HeNe monitoring laser longer than that for silicon. This trend is also observed for and a sample stage SiGe on SiO2 as compared to that on Si. This is believed to Applications of excimer laser in nanofabrication 11 Fig. 2 Schematic illustration of the TRR setup. A HeNe laser beam (λ = 633 nm) is incident at an angle on the UV laser exposed surface of a semiconductor. The reflected signal passes through a red filter and is collected by an ultrafast photodetector. The captured signal is analyzed by an oscilloscope

the reflectivity decreases from 92% to 85%, while that for Si increases from 34.7% to 73.4% [7]. Note that the reflectivity for molten metal is lower than that of its solid phase (rather than higher as in the case for semiconductors). This is probably due to the phonon scattering of the incident light in the metal examples, which is more noticeable when the temperature increases. This scattering results in a lower reflectivity [5].

3 Laser assisted nanoimprint lithography (LAN)

In traditional thermal NIL, both the mold and substrate are heated with the resists [8–10]. When the mold and substrates Fig. 3 A typical TRR curve captured from an oscilloscope screen. The are made from different materials, the thermal expansion reflectivity stays low when Si is at solid state, then jumps to a much difference will cause misalignment between the two. For ex- higher value within 1 ns when the surface is melted. After a certain ample, the thermal expansion coefficients of Si and SiO2 period of time, it drops to its normal value when the Si resolidifies. × −6 × −6 ◦ −1 The width of the curve is defined as the molten time, which is about are 2.6 10 and 0.5 10 ( C) , respectively [11]. If 70 ns in this case. The tail in the curve is probably due to the slow cool aSiO2 mold and a Si substrate are used during imprinting down process at the low temperature end in the solid state at 100◦C, the difference in thermal expansion would be 2.1 ×10−4, leading to a global misalignment of about 21 µm for a 4-inch wafer. Although a mold made from the same mate- result from the fact that the oxide layers have acted as ther- rial as the substrate has been demonstrated to minimize the mal barriers, making the heat dissipation speed into the bulk thermal expansion difference [12], this is not practical since substrate slower than the substrate without a SiO2 barrier. not every substrate material can be used for mold. A method In addition to semiconductors, the TRR technique is also that can greatly reduce the heating of the substrates and the suitable for measuring molten time of metals and other ma- molds is indispensable. terials which possess a difference in reflectivity between A new development in NIL is using a laser pulse as the their solid and liquid phases. For example, using a 633 nm heating source [13]. In this section, an excimer laser pulse wavelength HeNe laser, the molten times for Al, Al-based is used to melt the polymeric resists in NIL. Since there alloys, and TiN-based anti-reflective coatings have been is no direct heating of the underlying substrate, laser as- measured [7]. However, the difference in the reflectivity at sisted nanoimprint lithography (LAN) reduces the thermal solid and liquid states for metallic materials is not as signif- expansion effects. LAN is demonstrated to be capable of icant as for semiconductors. For example, upon melting Al, producing nanostructures in various polymers on different 12 Q. Xia, S.Y. Chou Fig. 4 Molten time measured by TRR technique for Ge, Si, SOI,andSiGeonSiandSiO2. (a) (100) Ge wafer. (b)Siand SOI, the SOI wafer has a device layer of 100 nm and buried oxide layer of 380 nm. The Si is a bare p-type (100) wafer. (c) SiGe on SiO2 and Si, the Si0.6Ge0.4 is 74 nm thick, and the oxide layer is 170 nm thick. Dots are measured data and lines are a linear fit for the data

substrates using only a single light pulse. The imprint time residual solvent. A fused silica (FS) mold with nanostruc- for LAN is less than 200 ns, which is measured using real- tures is applied against the resist film with some pressure. time monitoring by scattering of light (RIMS) [14]. Numer- A single XeCl excimer laser pulse passes through the mold ical simulation shows negligible heating of the substrate and and melts the polymer film, during which time the mold is mold, and little distortion of both. immediately imprinted into the resist. The mold is then sep- arated, leaving a negative pattern in the resist. 3.1 Principle of LAN Based on the working principle, it can be seen that LAN is an ultrafast process, taking place in nanoseconds. Since TheprincipleofLANisshowninFig.5. the laser pulse is so short, heating of the substrate is neg- A resist thin film is spin coated on a substrate (Si or ligible. The mold does not directly absorb the laser energy quartz). The resist-covered substrate is then baked at 60– because its bandgap (8 eV) [11] is much larger than the pho- 70◦C on a hot plate for about 30 minutes to drive out the ton energy of the 308-nm wavelength UV laser (4 eV). The Applications of excimer laser in nanofabrication 13

Fig. 5 Schematic illustration of LAN principle. (a) Bring mold and resist into contact under pressure, (b) UV laser irradiation; the resist melts upon laser exposure, (c) mold is pressed into resist layer, (d)re- Fig. 6 SEM image of the 200 nm period grating mold made from sist becomes rigid again; (e) mold separation. The mold is made from fused silica for LAN. The mold is made by NIL and RIE. The gratings fused silica, which is transparent to the laser pulse. A single laser pulse have a line width of 100 nm and height of 90 nm [16] is enough for the imprint process same principle may work for UV-NIL as well since many UV-curable resists will cross link at this wavelength.

3.2 Fabrication of imprint mold

Fused silica (FS) is used as the mold material for several reasons. It is transparent to the UV laser. The reflectivity at the FS/air interface is as low as 3% at normal incidence. FS is also mechanically strong, thermally and chemically stable, and does not react with the resists. The processing ability of FS using standard clean room techniques also favors it as an ideal candidate for mold material. In mold fabrication, the original master mold was pat- Fig. 7 SEM image of NPR-69 gratings on a Si substrate by LAN using terned using interference lithography, followed by pattern a single laser pulse with a fluence of 400 mJ/cm2 [16] transfer into silicon with thermally-grown oxide or bare silicon wafers [15]. 100 nm half-pitch gratings on a Si master mold were transferred to a 2-inch fused silica wafer using 3.3 Imprint results traditional thermal-NIL and reactive ion etching (RIE). The RIE was carried out in a Plasma Therm 2486 machine with A single laser pulse was sufficient to imprint the nanopat- 2 10 sccm CHF3, 1.5 sccm O2 at power density of 60 mW/cm terns of the mold into the resist with high fidelity. For ex- with a pressure lower than 3 mTorr. Cr was used as the etch- ample, with a laser fluence not lower than 350 mJ/cm2, uni- ing mask since it has very high selectivity to FS (>10:1). form gratings of the 200 nm period in the mold were trans- The etching rate was about 3–4 nm/min and the as-etched ferred completely to a thermoplastic resist, NPR-69, which ◦ FS trenches (depth = 90 nm) have vertical sidewalls and has a glass transition temperature of about 100 C. The im- flat bottoms. After stripping off the Cr using a CR-7 so- printed gratings had vertical sidewalls and flat surfaces and lution, the etched fused silica wafer was then diced into bottoms (Fig. 7) corresponding to those in the mold. Ex- 1 × 1mm2 pieces using a wafer dicing saw. The diced cellent pattern transfer by LAN was also observed in other molds were cleaned in a solution (NH4OH:H2O2:deionized resists, but at different laser fluences depending on the glass ◦ H2O = 1:1:5) at 80 C for 15 min (RCA #1 cleaning), fol- transition temperature and laser absorbance of the resist. For lowed by treatment with an anti-adhesion agent (1H, 1H, example, with NPR-46, a single laser pulse with a fluence of 2H, 2H-perfluorodecyltrichlorosilane, FDTS, Alfa Aesar) 560 mJ/cm2 was used to duplicate the pattern from the mold for better mold release during the separation. The finished to the resist (Fig. 8). molds have 200 nm period gratings (100 nm line/spacing, Resists on different substrates were also successfully im- 90 nm trench depth) over the entire mold area. A scanning printed by LAN. For NPR-69 on a quartz substrate, we electron microscope (SEM) image of the mold is shown in found that at least 500 mJ/cm2 laser fluence was needed for a Fig. 6 [16]. good imprint, higher than that on a Si substrate (Fig. 9). The 14 Q. Xia, S.Y. Chou difference in required fluence is attributed to the difference In LAN, the molds with anti-stick coating could be used in the laser absorbance of the resists on different substrates for several times without loss in grating quality during im- (21.4% for a 200 nm NPR-69 on a Si substrate, and 14.3% print. For example, the 1st and 3rd imprint with the same for that on a quartz substrate). This will be discussed in de- mold—without in between mold cleaning—resulted in the tail in Sect. 3.4. The other possible reason for the different same quality grating lines of NPR-46 (Fig. 10). fluences is that the quartz substrate does not absorb the 308- The imprinted area could be as large as that of the mold nm laser pulse, but a Si substrate does and transfers the heat and was quite uniform. An imprinted spot in NPR-69 was to the resist on top of it. Hence the incident energy was used examined under SEM after LAN at five different locations more efficiently with resists on the Si substrate. (four corners and one in the center), as shown in Fig. 11. From the top view SEM images, it can be seen that the imprint quality is high and uniform without noticeable defects. The pitch of the imprinted gratings was analyzed by fast Fourier transform (FFT) analysis using image processing software (Image Pro Plus, from Media Cybernetics, Inc., MD), as shown on the right panel of Fig. 11.Fromthe FFT images, it can be concluded that the pitch of the imprinted gratings at the five different spots is very uniform. This demonstrates both the uniformity of the imprint during LAN and the uniformity of the mold fabrication process for LAN.

Fig. 8 Atomic force microscope (AFM) image of NPR-46 gratings 3.4 Modeling of laser absorption in resists during LAN produced on a Si substrate. The height of the gratings is 90 nm, which corresponds exactly to the depth of the trench in the mold. A single laser pulse with a ﬂuence of 560 mJ/cm2 was used [16] Light absorption by the resist and substrate can be calculated using the following model. For example, with a SiO2/polymer/substrate sandwich, the light path in the system can be depicted as in Fig. 12. A simple case for this problem is when the substrate and the mold are both made of the same materials, such as SiO2. Consider a normal incidence (θi = 0), the refractive indices for both the mold and the substrate are equal and they are both real numbers. This is similar to a Fabry-Perot etalon [17] except that the system here absorbs energy in the intermediate layer. In this case, the fractional absorption IA Ii is [4]

∗ ∗ IA ARAR AT AT = 1 − ∗ − ∗ , (1) Ii AiAi AiAi

Fig. 9 SEM image of NPR-69 gratings made on a quartz substrate by where Ai , AR and AT are the amplitude of the incident, LAN with a single laser pulse of 500 mJ/cm2 [16] reﬂected and transmitted light, respectively.

Fig. 10 SEM images of 200 nm NPR-46 period gratings produced on a Si substrate by a single laser pulse (560 mJ/cm2) using the same mold. (a) Imprint result after mold is used for the ﬁrst time and (b) after the third time. There is no obvious difference in grating quality in the two pictures [16] Applications of excimer laser in nanofabrication 15

Fig. 11 SEM images (left panel) and corresponding FFT images (right panel)for different locations of an imprinted NPR-69 spot on a Si substrate. The locations examined are: (a)Upperleft corner, (b) upper right corner, (c) center, (d) lower left corner and (e) lower right corner. The gratings are uniform over the whole imprinted spot which is about 1 mm by 1 mm 16 Q. Xia, S.Y. Chou 3.5 Simulation of the substrate/mold heating and deformation

For the molds and substrates made from SiO2, the temperature remains nearly unchanged during LAN because heat conduction is poor (two orders of magnitude less than Si) [20] and there is no direct absorption of laser energy. For Si substrates, the heating is also greatly reduced since LAN has a much shorter processing time (hundreds of nanoseconds) than traditional thermal NIL. To better understand the heating of Si substrates, numerical simulation using ABAQUS (ABAQUS Inc., RI [21]) was carried out. Also simulated was the deformation of the mold and substrate during the = Fig. 12 Schematic light path during the LAN process (θi 0 in LAN). LAN process. The incident light passes through the transparent mold and reaches the The case simulated was that of a 200-nm thick NPR- resist layer and the substrate. Part gets reflected back, part is absorbed by the resist layer, and part goes into the substrate 69 on a Si substrate exposed to a laser pulse with a fluence of 400 mJ/cm2. In this simulation, the thermal diffusion length for Si was calculated using tabulated physical properties [20] to be 1.3 µm for a 20 ns laser pulse. The temperature profile at 2 µs after the laser pulse reached the surface is shown in Fig. 14. It can be concluded that the thermally affected zone is mainly localized at the Si surface. This is reasonable since Si is a much better thermal conduc- tor than SiO2. Also, only a very thin surface layer in Si is thermally affected, while the bulk of the substrate remains at room temperature. This is beneficial to the inhibition of thermal expansion since the bulk Si wafer at room temperature serves as a constraint, limiting the thermal expansion. To better visualize the temperature distribution, temperature as a function of time for different distances from the surface is plotted in Fig. 15. It illustrates how the temperature drops quickly with distance from the surface. For example, after a certain period of time, the temperature at a depth of 14.3 µm is below 50◦C. Only a thin layer Fig. 13 Calculated fractional absorption of a NPR-69 resist with an −4 −1 near the surface (about 3% of a 500 µm thick Si wafer) was absorption coefficient of 8 × 10 nm on silicon and quartz (SiO2) substrates. On a Si substrate the resist absorbs more energy than on a thermally affected during the process, while the bulk of the quartz substrate under otherwise identical conditions Si wafer remained at nearly room temperature. The figure also illustrates how the Si surface temperature drops below the polymer glass transition temperature (about 100◦C) However, when the substrate is Si, the above equation within 500 ns, which suggests that the whole imprint process (1) does not apply since in this case, n1 = n3.Alsoatthe time was shorter than 500 ns. It should be pointed out here 308-nm wavelength, the refractive index for Si is a complex that the heating of the Si substrate could be further reduced number. In order to solve the problem, commercially avail- by tailoring the resist’s chemical composition to increase its able software such as GSolver (Grating Solver Development laser absorbance (see Sect. 3.6 for example). Company, TX [18]) was used for the calculation. Because the bulk of the Si wafer is at room tempera- For NPR-69 resist which has an absorption coefficient of ture during the imprint process, the lateral thermal expan- 8 × 10−4 nm−1, the fractional absorption as a function of sion of the surface layer is efficiently constrained. This sup- film thickness is plotted in Fig. 13 for both quartz and Si ports good overlay alignment. Numerical simulation of the substrates. For a 200-nm thick NPR-69 film, the fractional stress distribution shows that the stress is localized at the absorption on quartz and silicon is 14.3% and 21.4%, re- corner of the mold/substrate interface. The maximum prin- spectively. In this calculation, the indices at this wavelength cipal stress is about 15.3 MPa for an extreme high pressure are 1.49 for SiO2 [19]; 5.015 + 3.650i for Si; and 1.57 for case (2700 psi applied pressure). This is far less than the NPR-69. Young’s Modulus of Si (107 GPa) and SiO2 (70 GPa) [11]. Applications of excimer laser in nanofabrication 17 Fig. 14 Temperature profile 2 µs after the laser pulse in a typical LAN process. The laser fluence is 400 mJ/cm2.Fromthe cross section-temperature distribution, the highest surface temperature is about 60◦Cafter 2 µs, indicating a fast heating/cooling process

different substrates were tested for LAN. Although satisfac- tory results were obtained for all the resists, it is worthwhile noting that tailoring the chemical composition of the resists can improve the light absorption, hence further reduce the substrate heating. For this purpose, NPR-series resists with varying amounts of UV light absorber have been developed. One of the best UV absorbers is pyrene (98%, Aldrich), which shows strong absorbance from 200–350 nm as our UV light absorber. The absorption spectrum of pyrene is (98%, Aldrich) is shown in Fig. 17 [22], with the molecular structure depicted in the inset. To prepare the resist, the chemical ingredients (pyrene, commercial polymer powders such as PS, etc.) were dissolved in cholorobenze (HPLC grade, Aldrich). The amount of pyrene used for this study was 1–10% of the combinational weight of other solid Fig. 15 Simulation results of temperature evolution in the surface chemicals. The solution was then stirred with a magnetic layer of a Si substrate after the incidence of a single laser pulse spin bar at 300 rpm overnight. Thin resist films with differ- (400 mJ/cm2). Each curve represents the temperature as a function ent thicknesses were spin coated on fused silica (FS) wafers, of time at different distances from the Si surface. The inset shows ◦ the model geometry in which we assume a 200 nm polymer film on followed by baking on a 70 C hot plate for 15 min to drive a 500 µm thick Si substrate [16] out the residual solvent. The absorption properties of the resists were character- Numerical simulation of the deformation of a Si substrate ized by their absorption coefficients, which were measured shows that the maximum lateral displacement is also limited using a simple set-up illustrated in Fig. 18. In this set-up, to the corner area within 1 µm range (Fig. 16). The max- the laser pulse was introduced perpendicular to the FS/resist imum in-plane principal strain is about 5 × 10−4 (0.05%), sample from the wafer side. The intensity of the incident thus the local displacement of the substrate and the mold light (Ii ), and that of the transmitted light after a film of due to the pressure is less than a nanometer. thickness L (IL) were measured by averaging reading from 10 laser pulses. 3.6 Resist engineering for LAN The relationship between Ii and IL will give us information about the absorption coefficients (α) according to the Commercial polymers such as polystyrene (PS, molecular following equation: weight 67.5 K), poly (methyl methacrylate) (PMMA, Mw 15 K), and nanoimprint resists (homemade NPR series) on IL = Ii · exp(−α · L). (2) 18 Q. Xia, S.Y. Chou Fig. 16 Maximum in-plane principal strain distribution during the LAN process. It occurs at the corner of the mold/substrate interface. The maximum value is about 5 × 10−4, which is 0.05%

Fig. 17 The absorption spectrum of pyrene dissolved in cyclohexane. Fig. 19 Measured absorption coefﬁcients of polystyrene (Mw 67.5K) The inset shows the chemical structure of pyrene. Reproduced from doped with varying amount of pyrene. With more pyrene, the absorp- reference [22] tion coefﬁcient increases noticeably

The addition of pyrene greatly enhanced the absorption coefficients of the resist. IL/Ii as a function of thickness L is plotted in a semi-logarithmic chart for pyrene doped polystyrene (Mw 67.5 K) in Fig. 19. The slope of the linear fit line represents the absorption coefficient for each case, which increased significantly with the addition of pyrene. For example, with 5 wt.% pyrene, the absorption coefficient is about 4.32 × 10−4 nm−1, while increasing pyrene to 10 wt.% results in an absorption coefficient of about −4 −1 Fig. 18 Setup for measurement of resists’ absorption coefficients. The 8.66 × 10 nm . With the same method, the absorption 308-nm wavelength light goes in from the fused silica side with a nor- coefficients of other resists were also measured. Similarly, mal incident angle. The light intensity is measured after the resist thin the addition of pyrene has resulted in higher absorbance. film (IL) and directly from the light source (Ii ) for different resist film thicknesses (L), which are used for calculating the absorption coeffi- The improvement in the absorption coefficient resulted cient (α) in less heating of the substrate during LAN. With the dif- Applications of excimer laser in nanofabrication 19 resist filling of the grating trenches—is monitored continuously during the imprint process. Previous work has demonstrated that RIMS is capable of monitoring NIL processes which finish in less than 1 second [23]. Here, RIMS is used for measuring the processing time of LAN.

3.7.1 Principle of RIMS

In RIMS, the intensity of diffracted light depends on the filling ratio of the polymeric resist to the mold trench. As the resist flows into the mold trenches, the diffracted signal de- cays monotonically because the imprint resist has an index almost the same as that of the mold (1.46 at 633 nm in our experiments). The diffraction will disappear when the resist fills the mold grating trenches completely (Fig. 21). A rigorous model (scalar diffraction model) [24] has been Fig. 20 Absorbance of resists with different absorption coefficients appliedtoRIMS[14, 25], and it clearly shows the relation- (α) on SiO2 substrates ship between the normalized diffraction intensity and the filling ratio of the mold trench. It is also possible for RIMS to monitor the case where the refractive indices of mold and ference in the absorption coefficients due to the alternation resist are not the same. In this case, the intensity does not of the chemical composition of the resists, the percentage linearly decrease with the filling ratio, nor does it reach zero of the light that is absorbed by the resist layer is going to when the trench is fully filled with resist due to the refractive change. As a result, the percentage of light that goes into the index contrast. substrate and heats the substrate will change, too. As an example, Fig. 20 gives the absorption properties (that is, frac- 3.7.2 Experimental details tional absorption intensity, Ia/Ii , where Ia is the absorbed intensity by the resist films) for resists with different ab- A schematic of the RIMS experimental setup is shown in sorption coefficients on fused silica substrates (SiO2) calcu- Fig. 22. The substrate, resist and mold in our experiments lated using a model developed elsewhere [4]. For a 200 nm were prepared as follows: A clean silicon substrate with na- thick resist film on silica substrate, an increase of absorp- tive oxide was spin coated with a 210 nm thick resist that tion coefficient from 4 × 10−4 nm−1 to 10 × 10−4 nm−1 will result in a fractional absorption increase from 7.5% to 18.8%. The thickness of the imprint resists plays an important role. Under otherwise identical conditions, the absorption is improved with a thicker resist film. These conclusions also hold for Si substrates.

3.7 Measurement of imprint time for LAN

It is important to understand the speed of LAN. From the simulation results (e.g., Fig. 15), the Si surface temperature remained above 100◦C for 500 ns, which suggests that the imprint should be done within 500 ns since the glass transition temperature of the resist is about 100◦C. However, to get an accurate imprint time, direct measurement is necessary. Fig. 21 Schematic illustration for the operational principle of RIMS. Recently, real-time monitoring by scattering of light A laser beam is incident on the grating area and the intensity of −1st (RIMS) has been proposed and demonstrated for flow char- order diffraction is monitored continuously. Before imprint (a), the in- acterization of both thermal and photocurable polymers or tensity of the −1st order is strongest. During imprint when the mold is monomer mixtures in NIL [14, 23]. In RIMS, a surface relief pressed into the resist (b), the intensity is reduced, and when the mold is completely pressed into the resist (c), it falls to zero. In this figure, diffraction grating is created on the imprint mold and the dif- it is assumed that the refractive indices of the mold and resist are the fracted light intensity from the grating—which depends on same. Reproduced based on description in [14] 20 Q. Xia, S.Y. Chou Fig. 22 Schematic illustration of the RIMS experimental setup. AUVlaserspot(λ = 308 nm) covers the mold/resist/substrate sandwich, which was pressed by two parallel metal plates (not shown here). A HeNe laser beam (λ = 633 nm) is incident on the mold at an angle of 60◦ to the mold surface. The angle between the HeNe laser beam incident plane and the grating lines was 45◦. The diffracted signal (−1st order) passes through a red filter and is collected by an ultrafast photodetector. The captured signal is analyzed by an oscilloscope

was custom-made by modifying the commercially available Nanonex resist NXR-1023 with a UV light absorber. The resist film was then baked at 70◦C for 30 min to drive out the residual solvent. The mold was an optical grade fused silica wafer (0.5 mm thick, UV-transparent, refractive index 1.46) having parallel 300 nm wide grating lines of 950 nm pitch and 150 nm depth over the entire mold area (1.2 mm by 1.2 mm) fabricated by NIL and RIE. The main reason for etching the mold to 150 nm deep is that at this depth, the diffraction intensity will be monotonously reduced with the filling of the trench (Fig. 23). This makes the interpretation of experimental data straightforward. The finished mold was coated with FDTS monolayers for easy mold release after imprint. During the imprint, the mold was placed on top of the resist and pressed into the resist by two parallel plates. There Fig. 23 Calculated diffraction efficiency as a function of SiO2 mold was a hole in the center of the top plate that was large trench depth for a 950 nm pitch fused silica mold using GSOLVER enough for the laser beams to pass through. An excimer software. The protrusion width is 300 nm. Based on this plot, the mold laser pulse with a spot size large enough (3 mm by 3 mm) in our experiment was etched to 150 nm deep so that a monotonous to cover the mold area was used to melt the resist. A HeNe change in the diffraction signal could be observed laser (632.8 nm wavelength, continuous wave, beam diameter 0.5 mm) was used to monitor the resist imprint by the cilloscope have been described earlier in Sect. 2 during the RIMS process. The incident angle of the HeNe laser beam TRR measurement. was 60◦ relative to the surface of the silica mold. The angle ◦ between the incident plane and the grating lines was 45 . 3.7.3 Measurement results This arrangement ensured that the diffracted beam was di- rected along a different path than the reflected beam. The Figure 24 is a typical curve of the diffracted light intensity −1st order diffraction signal was collected by a high speed as a function of time obtained during the real-time moni- Si p–i–n detector with a band pass filter mounted in front toring of the ultrafast imprint process [26]. Before imprint, of the detector. The signal was sent to an oscilloscope for the grating mold and resist were in contact at room temper- analysis. The specifications for the detector, filter and os- ature and the resist was rigid with almost no deformation. Applications of excimer laser in nanofabrication 21 Fig. 24 Typical curve of normalized diffraction intensity vs time during RIMS. This curve was obtained when imprinting a 210 nm thick NXR 1023 thin film on Si substrate at a laser fluence of 380 mJ/cm2. Three zones are defined on the curve that represent the stages before, during, and after imprinting. The schematics of polymer filling into the grating trenches in each of the three stages are drawn as insets. Blue arrows indicate the start and end points of imprint. The whole imprint process takes about 200 ns [26]

No polymer filled into the trenches at this stage and the dif- better mold geometry and a thicker resist film may also be fraction intensity stayed constant (Fig. 24, zone I). As the helpful for polymer flow in NIL [29, 30]. resist was heated by an excimer laser pulse, it became soft and started to flow into the grating trenches under the ex- 3.7.4 Imprint fidelity during RIMS for LAN ternal pressure, causing a monotonic reduction of the effective trench depth and hence a reduced diffraction intensity To confirm that the trenches in the mold have been fully (Fig. 24, zone II). After completion of the imprint process, filled with resist during RIMS for LAN, both the mold and the trenches were completely filled with resist and the dif- imprinted resist were examined using SEM and AFM. Geo- fraction intensity remained constant near zero, due to the metrical features such as flat tops and vertical sidewalls were refractive index match between the mold and resist (Fig. 24, fully transferred from the mold into the resist (Figs. 25a, b). zone III). The entire imprint process was completed in about More importantly, the height of the resist lines was the same 200 ns. as the trench depth in the mold (Figs. 25c, d), which suggests Furthermore, a spike was observed in the diffracted light that the mold was completely pressed into the resist within 200 ns. Further SEM characterization (Figs. 25e, f) showed curve (between zone I and zone II in Fig. 24). Since the that the line width of the protruding part in the mold was spike had a pulse width almost the same as the excimer laser the same as the trench width in the resist (both are 300 nm). (20 ns), and the height could be reduced greatly by improv- This demonstrates the high fidelity of the ultrafast imprint ing the UV filtering, it was determined that it came from process. the scattering of the excimer laser. The spike, which marks the time of the excimer laser beam, is useful to establish the 3.7.5 Picosecond capability of RIMS time relationship of laser melt and imprint [26]. Since the diffracted light intensity drops immediately after the spike, it In addition to measurement of imprint time for LAN, the reconfirms our previous observation (that the resist melts in experiments here demonstrated the capability of RIMS for less than 20 ns after the excimer laser pulse), which was ob- monitoring nanosecond imprint processes. However, the ul- tained by TRR measurement of resists using the same pho- timate time resolution of RIMS could be even higher. This todetector and oscilloscope as in this study [27]. can be achieved using an ultrafast photodetector together Based on the measured imprint time, the average speed with an ultrafast oscilloscope. For example, a commercially of the imprint is estimated to be 0.5 m/s. The imprint speed available high-speed photodetector can offer as high as a is related to several factors such as viscosity and surface ten- 60 GHz bandwidth (PX-D7, Newport Corp., Impulse re- sion of the polymer melt, imprint pressure and mold geome- sponse 7 ps, wavelength range 400–900 nm) [31]. And, a try. Faster imprint speeds can be achieved by using polymers digital storage oscilloscope with bandwidth up to 15 GHz that intrinsically have low viscosity. Higher laser fluence can (TDS6154C, Tektronix Inc., 28 ps rise time, 40 GS/s sample be used to raise the imprint temperature and lower the vis- rate) is also on the market [32]. Using this or other advanced cosity of the polymer melt. Raising the imprint temperature equipments may extend the capacity of RIMS to monitor ul- also decreases the surface tension of the polymer melt [28], trafast processes to tens of picoseconds. resulting in increased imprint speeds. In addition, according In summary, we have proposed and demonstrated the to recent experimental and numerical simulation results, a laser assisted nanoimprint lithography in which 100 nm 22 Q. Xia, S.Y. Chou Fig. 25 AFM and SEM images of the mold used [(a), (c), and (e)] and the imprinted resist [(b), (d), and (f)]. The flat tops and bottoms and vertical sidewalls in the mold were fully transferred to the resist during imprint. Cross-sectional analyses of the mold (c)andthe imprinted resist (d) showed that mold grating height was the same as resist trench depth (both at 150 nm), indicating a full pattern transfer. Linewidth of the raised area in the mold (e) was also the same as the resist trench width (f) (300 nm), suggesting high process fidelity [26]

wide (200 nm pitch) grating lines have been fabricated upon (LER). However, the resist is only an intermediate material exposure to a single laser pulse. Since the pulse duration is for patterning that will be removed after a pattern transfer. very short, the heating of the substrate and mold is negli- And even with a perfected resist profile, the pattern transfer gible. Numerical simulations have confirmed that the tem- from the resist to a hard material will introduce new fabrica- perature increase in the substrate/mold and the deformation tion defects. For example, using isothermal heating, NPR-69 of both are minimal. This could be helpful in reducing the lines on a Si substrate were heated at 80◦C for 5 min, which misalignment due to thermal mismatch of the mold and sub- resulted in smooth polymeric lines (Fig. 26a). However, af- strate. The imprint time of LAN has been measured to be ter RIE (with 10 sccm SF6, 40 sccm CHF3, 10 sccm Ar at about 200 ns using real-time imprint monitoring by scatter- 15 mTorr, power density 425 mW/cm2, 2 min), the resulting ing of light (RIMS). The resist for LAN can be chemically Si lines have rough edges (Fig. 26b). This line edge rough- tailored, resulting in even less heating of the substrate. This ness is introduced by the RIE process, which suggests that technique could be used in direct patterning of electronic a smoothing process as the final stage in nanofabrication is and optical devices. indispensable. Recently, a drastically different approach that removes defects in nanostructure geometry within nanoseconds was 4 Self perfection by liquefaction (SPEL) proposed [33, 34]. The new approach, termed self-perfection by liquefaction (SPEL), removes defects by selectively melt- 4.1 Introduction ing nanostructures with defects for a short period of time (e.g., hundreds of nanoseconds) while guiding the flow of Fabrication defects are unavoidable in most nanofabrication the molten materials with a set of boundary conditions. and ultimately determine the fabrication limit. Although a Each of the molten structures reshapes itself into a desir- lot of research has been carried out to correct the defects in able geometry and then re-solidifies, and maintains the new the process, subsequent fabrications add new defects. Previ- shape. Using SPEL, we reduced the LER of Si and Cr grat- ously, isothermal heating to reflow and smooth polymeric ing lines as much as 5.6 fold in less than 200 ns. For ex- resist lines has been used to reduce line edge roughness ample, the 3σ LER of 70 nm wide Cr lines was reduced Applications of excimer laser in nanofabrication 23 Fig. 26 Even using smooth polymeric lines (a)asmasksfor RIE, the resultant Si lines can be very rough (b) due to the defects introduced by the RIE process. The 120 nm wide NPR-69 lines were smoothed by isothermal heating at 80◦Cfor5min

from 8.4 nm to below 1.5 nm. This is well below the 3 nm During this process, the surface tension and the interaction “redzone limit” as shown in the International Technology between the molten material and the substrate cause the ma- Roadmap for Semiconductors (ITRS) [35]. We also dis- terial to flow to minimize the surface energy. As a result, covered, for the first time, that if a plate is placed a dis- the rough edges of a line are smoothed, and a circular dot is tance above the metal or semiconductor nanostructures dur- formed from a non-circular one. However, in O-SPEL, de- ing SPEL, the molten material self-rises to the top plate, and pending on the melting time and energy, the sidewall and the reshapes itself into new structures. These new structures not top surface can become rounded and the line width can be only have smooth edges, vertical sidewalls, and flat tops, widened. (2) Capped SPEL (C-SPEL): a flat plate is placed but also narrower width (>2.15-fold reduction) and greater in contact with the top of the to-be-perfected structures. This height (>2.10-fold) than the original (hence a >4.50-fold can be done by placing a single transparent place on top of aspect ratio improvement). The melting was achieved using all nanostructures, or one individual plate on the top of each a single nanosecond excimer laser pulse. structure. Under the boundary conditions set by the sub- In this section, the principle of SPEL is firstly introduced. strate and the top plate(s), the molten material reflows into Next, different forms of SPEL are demonstrated with exam- new shapes with smooth and vertical sidewalls, flat tops and ples. The mechanisms of SPEL are discussed according to the same height as originals. (3) Guided SPEL (G-SPEL): a different models. Finally, a few applications of SPEL are in- plate is placed a gap above the to-be-perfected structures. In troduced. this case, the interaction between the molten structures and the top plate can make the molten structure rise up to reach 4.2 Principle of SPEL the top plate. Consequently, a greater height and a narrower line width in addition to smooth edges, vertical sidewalls The principle of SPEL is based on the fact that surface ten- and flat tops are achieved. sion in a liquid will smooth out the rough edges of a liq- Based on the principle, it can be concluded that SPEL has uid line and change a non-circular shape of a liquid dot into the following novelties. a circle. This is because a smooth edge or circular shape represents an energy minimum in thermal equilibrium [36]. 1. It removes defects post fabrication. Although smoothing In SPEL, the defective nanostructures are selectively melted of etching barriers (e.g., resist lines) has been known for for a short period of time under a set of simple boundary many years [37], it is only intermediate material. Even conditions. In the molten state, the nanostructures reshape with a perfected resist profile, the transfer from the resist themselves into better geometries for energy minima. When to a hard material will introduce new fabrication defects, the structures re-solidify, the perfect shapes are maintained asshowninFig.26. From this point of view, SPEL is ca- and the set of boundary conditions can be removed. The pable of removing intrinsic fabrication defects that have melting can be achieved using a single nanosecond excimer no manufacturable solutions yet. laser pulse, which would melt only a thin surface layer of 2. It is an ultrafast, low-temperature process. Using of an a material in less than 1 ns [6] and keep other parts of the ultrafast laser pulse makes it possible that only the sur- material at a low temperature and in the solid phase. This al- face layer is thermally affected. The bulk of the mater- lows for modification of the surface layer while leaving the ial will stay at low temperature without being damaged. bulk unchanged. Isothermal heating has been demonstrated for polymer According to the differences in the boundary conditions, reflows and smoothing [37], but such an approach is not SPEL can be categorized into three basic forms (Fig. 27). encouraged for materials such as semiconductors, met- (1) Open space SPEL (O-SPEL): nanostructures are placed als, or hard dielectrics because these materials have high on a substrate without any additional boundary conditions. melting temperatures. This high temperature will destroy 24 Q. Xia, S.Y. Chou

Fig. 27 Working principle of three forms of self-perfection by liquefaction (SPEL). (a) Open SPEL, (b) Capped SPEL, and (c) Guided SPEL for (I) lines and (II) squares or dots. See text for more detail [33]

other parts of the devices as well as the substrates. Un- detrimental to nanofabrication and nanodevices. The use like other smoothing techniques, SPEL avoids a global of guiding conditions to achieve a desired self-perfection isothermal high temperature process. not only opens new avenues to overcome these problems, 3. It is a selective process. Selectivity of SPEL has a two- but also presents interesting physics at nanoscale. Fur- fold meaning. First, it is materials selective. The heat- thermore, different boundary conditions (such as combi- ing in SPEL is performed by a pulsed laser, which is nation of different surface properties) may lead to even absorbed by only a thin surface layer of the materials better self-perfection. with a bandgap smaller than the photon energy [38], so that other materials (such as SiO2) on the surface or un- 4.3 Open space SPEL (O-SPEL) derneath will be kept at much lower temperature during SPEL. Second, this process is area selective. For exam- 4.3.1 Experimental ple, in our study, the laser spot is about 3 mm by 3 mm and can be adjusted within a certain range. A selective In order to demonstrate the principle of SPEL in open space, melting of certain areas of a wafer can also be achieved several types of samples were prepared: (1) metal pads of with a mask (materials either reflecting or absorbing laser 200 to 225 nm pitch with rough edges; (2) 200 nm pitch energy) that is placed either directly on the wafer or a dis- metal lines with rough edges; (3) Si lines on SOI wafers or tance away. This allows us to selectively expose the area on Si wafers, with a zigzag profile or rough sidewalls. To with defects while leaving other components intact. make the metal nanostructures, NIL was first performed on 4. The perfecting is achieved by itself under simple bound- thermoplastic resists, followed by oxygen RIE to remove the ary conditions. The profile of the nanostructures can be residual resist layer. Metallization was then carried out using enhanced as well under certain conditions. Previous work an electron gun evaporator, followed by a liftoff process. The using laser heating to cause material reflow—such as roughness in the resultant nanostructures was mainly from planarization of phosphosilicate glass (PSG) [39] and the RIE and liftoff processes. smoothing of SiO2 optical disks with a continuous wave To fabricate the Si nanostructures, NIL was first used for (cw) CO2 laser [40], and metal planarization [41] and Si patterning the resists. Two sets of molds were used. One large grain growth promotion using pulsed lasers [42]— was a defective mold that has severe line edge roughness did not use any guiding in laser melting. As a result, the from failed interference lithography. Lines on the mold were smoothed structures usually have sloped and curved side- about 70 nm wide (200 nm pitch) and had a zigzag shape. walls, and a rounded line top, which are undesirable and The other was a mold having 950 nm pitch and 250 nm Applications of excimer laser in nanofabrication 25 wide lines with much better edge smoothness. After NIL with a single excimer pulse with a laser fluence of 545 and 2 and O2 RIE, SiOx was deposited on the samples using evap- 440 mJ/cm , respectively, the zigzag edges became smooth oration. After liftoff, chlorine based RIE with 10 sccm Cl2, (Figs. 28b, 30b). Using digitized SEM images and fractal 20 sccm Ar, 100 mW/cm2, 8 mTorr was used to intention- analysis [43], we found the 3σ line edge roughness (LER) ally create sidewall roughness for demonstration purposes. of the grating lines, which is defined as 3 times of the root Then samples were dipped in diluted HF briefly to remove mean square (RMS) of variation in fractal analysis, is re- the SiOx mask before being cleaned in RCA #1 solution. In duced from an original 19.5 nm to 3.6 nm for Fig. 28a and both cases, such a “bad” mold or RIE recipe offers an ex- from 14.4 nm to 3.3 nm for Fig. 30a, respectively, represent- cellent test of the effectiveness of SPEL in perfecting severe ing a LER improvement (simply defined as 3σbefore/3σafter) edge (sidewall) roughness. During SPEL experiments, sam- of 542% and 436% (Table 4). As shown in Table 4, the SEM ples were exposed to single laser pulses with proper laser image resolution is 1.7 nm/pixel for Fig. 28, hence the ac- fluence. The characterization was done using an SEM and tual LER after O-SPEL can be less than calculated. A close an AFM. examination of the LER in the SEM image (Fig. 29)shows that the maximum variation has been reduced from 16.1 nm 4.3.2 Results to 3.5 nm, and that there are high frequency digitizing noises from the SEM imaging process superimposed on the actual Si lines Before O-SPEL, the original 70 nm wide, 100 nm LER. high Si grating lines on a SOI wafer had a zigzag shape To demonstrate the ability of O-SPEL to remove the side- (Fig. 28). Similarly, zigzag shaped 70 nm wide, 70 nm high wall roughness, 250 nm wide Si grating lines in both Si and Si grating lines were made on a (100) bare Si wafer as SOI wafers were fabricated (Figs. 31a, 32a). well (Fig. 30). These Si lines had severe edge roughness, After O-SPEL with a single excimer pulse with a laser which came from the “bad” imprint mold and RIE in pat- fluence of 880 and 490 mJ/cm2, respectively, the sidewall tern transfer, as discussed earlier. However, after O-SPEL roughness in both cases was reduced (Figs. 31b, 32b). The

Fig. 28 70 nm wide and 100 nm high zigzag Si lines on SOI (a) were smoothed during O-SPEL using a single pulse of 545 mJ/cm2 (b)[33]

Fig. 29 Digitized edge of zigzag SOI lines before (a)andafter(b) O-SPEL. The edge proﬁles are from the high resolution SEM images of the 70 nm wide SOI lines shown in Figs. 28aandb[33] 26 Q. Xia, S.Y. Chou Fig. 30 70 nm wide, 70 nm high zigzag Si lines on Si (a) were smoothed during O-SPEL with a single pulse of 440 mJ/cm2 (b)[33]

Fig. 31 250 nm wide, 150 nm high Si lines with sidewall roughness (a) smoothed out by O-SPEL using a single laser pulse of 880 mJ/cm2 (b)[33]

Fig. 32 250 nm wide, 190 nm high Si lines on 380 nm thick buried oxide (SIMOX SOI) with sidewall roughness (a) smoothed out by O-SPEL with a single pulse of 490 mJ/cm2 (b)

Fig. 33 (a) 70 nm wide, 40 nm high rough edged Cr lines on 22 nm thermal oxide capped Si wafer. (b) After O-SPEL with a single laser pulse of 320 mJ/cm2, the lines become ultra-smooth [33]

difference in the laser fluence needed for Si was higher than For example, with a laser fluence of 1200 mJ/cm2 for the Si the SOI lines even though the lines in SOI were higher than linesshowninFig.31a, the resulting round cross-sectioned that on Si. This fact can be explained by the effect of SiO2 Si lines had a root size of about 520 nm, much larger than as a thermal barrier layer as discussed earlier in Sect. 2.It the original 280 nm. This is because high fluence results in should be noted that in these cases, the cross section of the longer molten time and hence longer flow distance, so larger lines changed from rectangular to semi-circular. roots are developed. The Si root sizes (line width on a top view image) of the Si lines after O-SPEL almost stayed at the same as be- Metal lines For metal lines, two types of nanostructures fore O-SPEL. However, if higher laser fluence was used, the were fabricated, namely, 70 nm wide Cr grating lines root size of the Si lines could be larger than the original. (Fig. 33a) and 75 nm wide Au grating lines (Fig. 35a). Both Applications of excimer laser in nanofabrication 27

Fig. 34 Digitized edge proﬁles of the Cr lines before (a)andafter that O-SPEL reduces 3σ LER of 70 nm Cr lines from 8.4 nm to below (b) O-SPEL. The edge proﬁles are from the high resolution top-down 1.5 nm, the same order of the noise from SEM imaging [33] SEM images of the 70 nm Cr lines shown in Figs. 33aandb.Note

Fig. 35 (a) 75 nm wide, 40 nm high rough edged Au lines on 22 nm thermal oxidized capped Si wafer. (b) After O-SPEL with a single laser pulse of 150 mJ/cm2, the lines become ultra-smooth

Fig. 36 (a)10nmthick,75nm side Cr pads with rough edges on a 300 nm SiO2 capped Si wafer. (b) After O-SPEL with a single laser pulse of 1.7 J/cm2, the original rough-edged pads were turned into smooth round dots of 50 nm in diameter and 22 nm in height [33]

are 40 nm high and on 22 nm thick thermal SiO2 capped Si (see Table 4). This is well below the 3 nm “red-zone limit” wafers. After O-SPEL with a single pulse of ∼320 mJ/cm2 and even smaller than the targeted LER for the year 2013. and ∼150 mJ/cm2 for Cr and Au, respectively, the rough lines were smoothed (Fig. 33,Fig.35). The edge proﬁles Metal dots To demonstrate the capability of O-SPEL in before and after O-SPEL of Cr are shown in Figs. 34a and turning the rough edged metal dots into nearly perfect round b, respectively. ones, Cr squares of 10 nm thick and 75 nm×75 nm area with It is worthwhile to note that for the Cr lines, the 3σ LER rough edges (Figs. 36a, 37a) were fabricated on a 300 nm before and after OSPEL was reduced from 8.4 nm to 1.5 nm thick thermal oxide capped Si wafer. A single laser pulse of 28 Q. Xia, S.Y. Chou Fig. 37 AFM images of Cr dots before (a)andafter(b) O-SPEL. These AFM images correspond to SEM images shown in Figs. 36a and b, respectively

Fig. 38 (a) Cr rectangles of 145 nm × 110 nm with rough edges. These Cr rectangles are 22 nm thick, 225 nm in pitch and were made on a 100 nm thick thermal SiO2 capped Si wafer. (b) After O-SPEL with a single pulse of 300 mJ/cm2,the rectangles became round shaped with a diameter of 90 nm and height of 47 nm. Laser melting did not change the center to center spacing

Fig. 39 FFT patterns for Cr rectangles before (a)andafter (b) O-SPEL. Both patterns showed that the periodicity stayed the same (225 nm) before and after O-SPEL

1.7 J/cm2 turned the squares into nearly perfectly round dots SPEL. The sides of the rectangles before O-SPEL were with a 50 nm diameter and 22 nm height (Figs. 36b, 37b). 143.7 ± 6.1 nm and 110.5 ± 3.8 nm, with a length/width O-SPEL changed the shape of rough-edged metal dots ratio of 1.29 ± 0.06. After O-SPEL, the circles had a di- into round ones and reduced their sizes. It also kept the ameter of 90.3 ± 1.8 nm, and the “length/width” ratio was periodicity of the dots before and after perfecting. The 1.08 ± 0.04, suggesting that the circles were in nicely round size of the nanostructures became more uniform as well. shape. From these data, it can be concluded that both the size For example, Cr nanodots of rectangular shape and size and size distribution became smaller after O-SPEL, with a approximately 145 nm × 110 nm were made on 100 nm standard deviation improved by a factor of ∼3.4. Figure 39 thick thermal oxide on a Si wafer (Fig. 38a). After O- shows the fast Fourier transform (FFT) of the Cr rectangle SPEL with a single laser pulse of 300 mJ/cm2, the rec- arrays. It was found that the pitch of the Cr rectangles before tangles turned into circles with a diameter of 90 nm and after O-SPEL did not change; both were still 225 nm, (Fig. 38b). suggesting that laser melting did not change the center to The size and pitch distributions before and after O-SPEL center spacing. were analyzed in Image Pro Plus [44]. A statistical size In addition to removing the defects, another important ad- analysis was carried out for 150 dots before and after O- vantage of open-SPEL is that reduces the original size. The Applications of excimer laser in nanofabrication 29 final dot diameter after SPEL can be estimated from the con- diameter, d, of the partial sphere is given by stant material volume V and contact angle θ (Figs. 40a, b). ⎧ 3V 1 ◦ ⎨2r = 2[ ] 3 sin θ, for θ<90 , If we assume the dot is a segment of a perfect sphere, its π(2−3cosθ+cos3 θ) d = (4) volume is ⎩ 3V 1 ◦ 2R = 2[ ] 3 , for θ ≥ 90 . π(2−3cosθ+cos3 θ) πh πR3 V = 3r2 + h2 = 2 − 3 cos θ + cos3 θ , (3) 6 3 The dependence of the effective metal dot diameter on the volume of the metal for materials with different contact where h is the height of the cap, r the diameter of the cap, R angles (we use 60◦ and 140◦ in the calculation) was plotted the diameter of the hosting sphere, and θ the contact angle. in Fig. 40c. Clearly, with less material and/or a larger contact Because the h of partial sphere is often many times larger angle, the dot size becomes smaller. than the initial pattern thickness, the diameter of the final Following the “scaling rule”, round Au dots of 10 nm partial sphere formed by melting the pattern will be smaller diameter were fabricated on a Si wafer capped with 200 nm than the lateral dimension of the initial pattern. The effective thick thermal oxide (Fig. 41). The structure before O-SPEL was made by NIL, RIE and an e-beam evaporator (4 nm Au was deposited on 1 nm thick Cr on the substrate), with 200 nm pitch and 46 nm width. After single pulse O-SPEL with a laser fluence of 640 mJ/cm2, the Au pads were turned into nanodots of 10 nm in diameter (Fig. 41).

4.3.3 Discussion on the O-SPEL speed

O-SPEL is an ultrafast process which ﬁnishes within hundreds of nanoseconds for Si and metals due the high surface tension and low viscosity of the molten materials. The self- perfection time in free space can be estimated using level- ing theory. In this theory, the smoothing time is proportional to the surface tension of the liquid, while inversely proportional to the viscosity. For a liquid layer with a thickness of h and a surface variation period of L, the time (T ) required for the amplitude of the surface wave to decay to 1/e of its initial amplitude is given by [45]

3ηL4 T = , (5) 16π 4γh3

where η and γ are the viscosity and the surface tension of the molten material, respectively. Fig. 40 Schematic of effective metal dot diameter when the contact ◦ The viscosity and surface tension data for different ma- angle between the dot and the substrate is smaller than 90 (a); and terials at their molten states are listed in Table 1.Fromthis larger than or equal to 90◦ (b). (c) Calculated dependence of metal dot diameter on the material volume for the materials with contact angles table, it can be noticed that the molten Cr and Si has a vis- (to the substrate) of 60◦ and 140◦, respectively [63] cosity about 1,000 to 1,000,000 times lower and a surface

Fig. 41 10 nm Au dots fabricated by single pulse O-SPEL of 640 mJ/cm2.The original structures were 46 nm in size fabricated by depositing 4nmthickAuon1nmCron the substrate patterned by NIL (a) and the final round dots’ size was 10 nm in diameter (b) 30 Q. Xia, S.Y. Chou Table 1 Viscosity and surface tension for molten materials controlled. However, O-SPEL still suggests a new approach Viscosity Surface tension References for removing nanofabrication defects. (mPa s, cp) (mN m−1) 4.4 Capped SPEL (C-SPEL) Silicon 0.58 720 [46, 47] Chromium 5.70 1,642 [48, 49] As discussed earlier, one limitation of O-SPEL is that Gold 5.13 1,145 [11, 49] the repaired structures have sloped and curved sidewalls PMMA (3K) 105–106 31 [50] and rounded tops. In some cases—such as coupling of Water 1 73 [11] waveguide to fiber optics—this is actually beneficial for the applications. However, a rectangular cross section with flat tops and vertical sidewalls is desired for pattern transfer fidelity (e.g., during RIE). Different from O-SPEL, the capped SPEL (C-SPEL) can keep the top surface of a to-be- perfected structure flat and the sidewall vertical. There are two types of C-SPEL (Fig. 27). One uses a FS plate on top of the nanostructures to be repaired (which is in contact with the structures to hold them during melting) and the other uses a Cr/SiO2 double layer on top of each structure to block the laser light from reaching the top of nanostructures.

4.4.1 Experiments

To test the principle of C-SPEL, both Cr and Si grating lines with rough edges were fabricated using NIL, RIE, metallization and liftoff. The samples and their preparation pro- Fig. 42 Smoothing time as a function of viscosity for semiconductors, metals and polymers. A smaller viscosity at molten state favors a faster cedures are listed in Table 2. For schemes 1 and 2, a fused smoothing process. The film thickness is 50 nm and the surface vari- silica plate was placed in contact with the defective Si or Cr ation period is 1 µm for the calculations. Adapted from reference [51] nanostructures with some pressure. For scheme 3, Cr/SiO2 with extended data range double layer was used. This double layer was removed by CR-7, diluted HF (1:10) sequentially after C-SPEL. tension of 20–100 times higher than a molten polymer. This leads to a smoothing time 3–8 orders of magnitude shorter. 4.4.2 Results For example, for smoothing L = 1 µm in a 50 nm thick film, it would take 52 ns, 16 ns and 51 seconds for molten Cr, Si, Figure 43 shows the SEM images before and after C- and PMMA, respectively. SPEL of the 280 nm wide Si lines on a SOI substrate To better understand the difference in smoothing time guided by a single fused silica plate. The laser fluence for different materials, plots of smoothing time as functions used is 480 mJ/cm2. Comparison before and after C-SPEL of viscosity and surface tension are shown in Fig. 42 for a shows that the linewidth and height did not change, while 50 nm thick film with a surface variation period of 1 µm [51]. the roughness on the sidewall had been removed. Frac- The approximate ranges of smoothing time for metals, Si tional LER analysis showed that the 3σ LER had been re- and polymers are marked on the plots using ellipses. It is duced from 9.6 nm to 4.2 nm. Similarly, for Cr lines, the clear that the smoothing time needed for metals and semi- linewidth and height did not change before and after CSPEL conductors is much shorter than that for polymers. A shorter (406 mJ/cm2,Fig.44),witha3σ LER improvement from time in SPEL helps the self-perfection of desired structures 17.7 nm to 7.5 nm. without degrading other structures adjacent and underneath. For 200 nm wide Si lines guided by individual caps Limitations to O-SPEL do exist. For example, the cross (Cr/SiO2)—one on each line—a single laser pulse of sections of the nano-structures are altered after SPEL. The 390 mJ/cm2 was used. After C-SPEL, the top of these struc- line width after O-SPEL may change depending on the laser tures were flat and the sidewalls were vertical (Fig. 45). The fluence and melting time. Under certain conditions, the in- fractal analysis shows that the 3σ LER had been reduced stability in liquid may play some adverse roles and should be from 11.1 nm to 5.4 nm. Applications of excimer laser in nanofabrication 31 Table 2 Samples and fabrication procedures for C-SPEL

Scheme Structures Fabrication procedures

1. Single plate for all Si lines Si lines: 280 nm wide; 200 nm high; on (1) Thermal imprint NPR-69; (2) O2 RIE residual layer; 500 nm thick buried oxide (SIMOX SOI) (3) Evaporate SiOx ; (4) Liftoff; (5) RIE Si (10 sccm Cl2, 20 sccm Ar, 8 mTorr with 100 mW/cm2); (6) Diluted HF dip.

2. Single plate for all Cr lines Cr lines: 280 nm wide; 62 nm high; on (1) Thermal imprint NPR-69; (2) O2 RIE residual layer; 220 nm thick thermal oxide (3) Evaporate 62 nm Cr; (4) Liftoff.

3. Individual plate for each Si line Si lines: 200 nm wide; 140 nm high on Si (1) Grow 45 nm thick thermal oxide; (2) Thermal imprint NPR-69; (3) O2 RIE residual layer; (4) Evaporate 45 nm Cr; (5) Liftoff; (6) RIE SiO2(33 sccm CF4, 7 sccm H2 at 50 mTorr and 425 mW/cm2 for 1 min); (7) RIE Si (40 sccm CHF3, 10 sccm Ar, 10 sccm SF6 at 15 mTorr and 425 mW/cm2 for 2 min and 40 secs).

Fig. 43 Using a fused silica plate in contact with the Si lines of ing the line height, ﬂat top and vertical sidewalls. A single laser pulse 2 280 nm width, 200 nm height, 950 nm pitch on 500 nm thick SiO2 of 480 mJ/cm was used in C-SPEL. The 3σ LER was reduced from (SIMOX SOI), the rough lines in (a) were smoothed (b) without chang- 9.6nmto4.2nm[33]

Fig. 44 C-SPELofCrlines using a FS plate. (a)Initial structure is 280 nm wide, 62 nm high, with a 3σ LER of 17.7 nm. (b) After C-SPEL with a single pulse of 406 mJ/cm2, the width and height of the Cr lines was preserved but the 3σ LER was reduced to 7.5 nm [33]

Fig. 45 SEM images of 200 nm wide, 140 nm high Si lines before (a)andafter(b) C-SPEL using 45 nm Cr/45 nm SiO2 double layer as the cap and a single laser pulse of 390 mJ/cm2 [33] 32 Q. Xia, S.Y. Chou 4.4.3 Discussion to O-SPEL considering the energy minimum principle. For the nanostructures tested in our experiments, this criterion is The C-SPEL is made possible by the energy minimiza- satisfied for all cases (Figs. 43 to 45). tion principle. Consider a line with width w, height h and length l. The surface tension at the molten state is γ .The 4.5 Guided SPEL (G-SPEL) cross sections corresponding to the original, O-SPEL, and C-SPEL are illustrated in Fig. 46. In order to simplify the In guided-SPEL, local spacers are used to keep a plate fixed problem, let’s just consider a special case where the cross above the nanostructures (Fig. 27), the molten nanostruc- section after O-SPEL is a half-circle of radius r. This means tures rise against the surface tension until they reach the the contact angle of the molten materials with the substrate plate. This leads to smooth edges, vertical sidewalls and flat is 90◦, which is very close to that of molten Si on SiO 2 tops, and also to narrower linewidths and greater line heights (87◦ [52]). (and hence higher aspect ratios) than the original structures. The surface energy for the above three cases can be ex- The entire melting, rising-up and reshaping took less than pressed as 200 ns for silicon and chromium materials.

Ea = γ · (w + 2h)l, (6) 4.5.1 Experimental details Eb = γ · πrl, (7)

Ec = γ · 2hl. (8) Samples and fabrication Four types of nanostructures were fabricated for G-SPEL experiments: (1) Si dots; (2) Due to volume conservation, we have Cr dots; (3) Si lines; and (4) Cr lines. Their geometries are 1 listed in Table 3. The Si nanostructures were fabricated on whl = πr2l, (9) 2 an epitaxy SOI wafer which had a 50 nm thick Si device layer and 200 nm thick buried SiO2 by thermal NIL, O2 which gives the relationship between radius r and w, h as RIE, SiO etching mask deposition, liftoff and Cl-based RIE x with a ﬁnal quick HF dip to remove the residual SiO mask. 2wh x r = . (10) The Cr nanostructures were deposited on NIL patterned re- π sists by evaporation followed by liftoff. All the as-fabricated The equation for surface energy after O-SPEL (case (b) in nanostructures had rough edges which were introduced by Fig. 46) reduces to the fabrication process. 2wh √ E = γ · πrl = γ · π l = γ · 2πwhl. Spacer fabrication To control the gap between the top b π plate and the substrate, silicon oxide (SiOx ) spacers were Since (w + 2h)2 = w2 + 4h2 + 4wh ≥ 4wh + 4wh = deposited onto a fused silica (FS) wafer through a shadow 8wh > 2πwh, the surface energy after O-SPEL (case (b)) is mask using the electron beam evaporator. The SiOx islands smaller than that before O-SPEL (case (a)), which is reason- were about 500 µm by 500 µm in size and had different able for all the O-SPEL experiments. heights. The distance between each spacer was 700 µm in In order to favor C-SPEL over O-SPEL, γ · 2hl < γ · πrl one direction and 1200 µm in the other. should hold. This leads to the following geometrical condi- Another type of spacer was fabricated by etching into the tion: FS wafer using evaporated Cr as an etching mask, followed by CR-7 etching and RCA #1 cleaning. Although this fab- 2 w> h(= 0.64h). (11) rication process took more steps than the ﬁrst approach, the π spacers material is denser than the evaporated SiOx so they According to (11), as long as the width/height ratio of the could be cleaned under RCA #1 without degradation after original line is larger than 0.64, C-SPEL will be favorable each G-SPEL experiment and used repeatedly.

Fig. 46 Schematic cross section for original line (a) and those during O-SPEL (b) and C-SPEL (c). The surface energy is calculated using the ﬁgures here. For simpliﬁcation, the cross section after O-SPEL is considered as a semi-circle Applications of excimer laser in nanofabrication 33 Table 3 Samples and results for Si and Cr G-SPEL

Samples Geometries Gap Dose G-SPEL O-SPEL (mJ/cm2)

Si dots 90 nm × 100 nm 22 nm 400 78 nm diam. 85 nm diam. 50 nm height 73 nm height 62 nm height

Cr dots 140 nm × 110 nm 40 nm 420 70 nm diam. 80 nm diam. 20 nm height 62 nm high 48 nm height

Si lines 285 nm width 40 nm 595 175 nm width NA 50 nm height 90 nm height

Cr lines 280 nm width 70 nm 406 130 nm width NA 62 nm height 130 nm height

Fig. 47 SEM images of Si and Cr dots before and after O-SPEL and G-SPEL. As-fabricated Si squares (90 nm × 100 nm and 50 nm tall) (a) became semi-spheres (85 nm diameter and 62 nm tall) if O-SPEL was used (b), but became cylinders if G-SPEL (22 nm gap) was used (c). The cylinders had a 78 nm diameter and 73 nm height (150% of original height), in addition to ﬂat tops and vertical sidewalls. Similarly, the fabricated Cr squares (140 nm × 110 nm and 20 nm tall) (d) became semi-spheres (80 nm diameter and 48 nm tall) after O-SPEL (e), but became the cylinders that had a 70 nm diameter (50% of original in lateral size) and 62 nm height (310% of original height) in addition to ﬂat tops and vertical sidewalls (f) after G-SPEL (40 nm gap) [33]

G-SPEL experiments The substrate and the top plate were 4.5.2 Results sandwiched between two metal plates which applied pressure via a set of screws that kept a conformal contact. The The results for the G-SPEL of Si and Cr, together with the gap between the top plate and the surface of the nanostruc- experimental conditions are summarized in Table 3. tures was kept constant by the spacers. In all cases, a single laser pulse was sufficient to achieve G-SPEL. There was a Cr and Si dots An O-SPEL with a single laser pulse ex- laser fluence window for G-SPEL; for example, fluence of posure made the original dots round (Figs. 47b, e), while 370 to 450 mJ/cm2 worked best for G-SPEL of Cr dots. The a G-SPEL with the same laser fluence resulted in cylinders nanostructures before and after the laser pulse were charac- (Figs. 47c, f). For Si, with a single pulse of 400 mJ/cm2,the terized using both SEM and AFM. original pads of 90 nm by 100 nm and 50 nm tall (Fig. 47a) 34 Q. Xia, S.Y. Chou became (1) smooth semi-spheres of 85 nm diameter and for t ∼ 200 ns (observed in G-SPELs), d ∼ 7 × 10−8 m(Cr) 62 nm tall if an O-SPEL was used (Fig. 47b); and (2) cylin- and 4 × 10−8 m (Si), plugging in the viscosity and surface ders of smooth and vertical sidewall, flat top, 78 nm diam- tension date (in Table 1) shows that the required Φ is 22 eter, and 73 nm tall (150% of original height) if a G-SPEL and 7 V for G-SPELs in Cr and Si, respectively. These are was used (Fig. 47c). Similarly, a single pulse of 420 mJ/cm2 much larger than possible in the experiments since the work made the 140 nm by 110 nm by 20 nm Cr pads (1:7 as- function difference is only a few volts. This discrepancy sug- pect ratio) (Fig. 47d) into semi-spheres of 80 nm diameter gests that the model for polymers may not be applicable to and 48 nm tall (Fig. 47e) in O-SPEL; but a G-SPEL with Si and metals. the same laser fluence resulted in cylinders of 70 nm diame- The possible reasons for the discrepancy could be as fol- ter, 62 nm tall (310% of original height, and an aspect ratio lows. First, Si and metals have much higher surface tension of almost 1), in addition to flat tops and vertical sidewalls than molten polymers, hence require a much higher pulling (Fig. 47f). AFM characterization of the original Cr dots and force, which might be out of range for that model. Second, those after O-SPEL and G-SPEL, is shown in Fig. 48, which the original model was developed for a continuous film, in corresponds to the right column in Fig. 47. which case an assumption that the film thickness is much smaller than its lateral dimension was made [55]. However, Cr and Si lines Figure 49 shows SEM of Cr lines and Si in the G-SPEL experiments, the thickness and width are on lines before and after a single laser pulse G-SPEL (guided the same order. by a single quartz plate with 70 and 40 nm gaps above the Although other mechanisms at nanoscale need to be ex- original structures, respectively). Figures 49aandbshow plored in order to understand the underlying physics, the that the original 280 nm wide Cr lines of 62 nm thick on short self-perfection time makes SPEL very useful in many SiO2 surface became 130 nm wide (46% of original width) applications. It allows for the self-repair and self-perfection and 130 nm tall (210% of original height) after G-SPEL, of a desired structure without damage to degradation of and hence a 452% (4.5 times) increase in aspect ratio was other structures on the substrates or the substrate itself. achieved. The 3σ LER was reduced from 17.4 nm to 5.4 nm (322% improvement). For 285 nm wide silicon lines of Parameters of G-SPEL Several parameters are crucial to a 50 nm height, G-SPEL made them 170 nm wide and 90 nm successful G-SPEL. First of all, an appropriate gap size be- 2 2 high. A single laser pulse of 406 mJ/cm and 595 mJ/cm tween the guiding plate and the nanostructures is necessary. was used for Cr and Si lines, respectively. All G-SPEL oc- The gap size should be large enough to allow the molten ma- curred in a time frame of hundreds nanoseconds. terials to rise after being melted, since when a wide line/dot becomes narrower, the height always changes. However, if 4.5.3 Discussion the gap size is too large, there would not be enough force to pull the molten materials upwards. According to (12), Mechanism for self rising during G-SPEL Previously, a the rising time is proportional to the third power of the dis- self-rise-up of material under a plate placed a gap away tance (gap size), which means even with enough electrosta- was observed only in continuous polymer films [53, 54]. tic force, the rising time needed is very sensitive to the gap A model based on electrostatic interaction was developed size. which explained the experimental date well [53–55]. Here The geometries of the original nanostructures also mat- we are trying to fit that model to our observation with metal ter. Our experimental data showed that under certain circum- and Si in order to find out if the phenomena is governed by stances, the width and height of the lines played an impor- the same physics. tant role. For example, with 280 nm wide, 62 nm thick Cr According to reference [55], the time for the material lines, the resulting structures in G-SPEL were raised lines. (polymer in that case) to rise up can be calculated using the However, metal lines with the same height but narrower following formula: width (e.g., 90 nm, 50 nm, 40 nm) resulted in partial or total ηγ d3 fragmentation of the lines into nanoparticles. t = , (12) The thickness of the lines plays a similar role. For ex- ε2( Φ)4 0 ample, reducing the film thickness from 62 nm to 35 nm where t is the time for the material to rise up, d is the initial (as shown in Fig. 49) for Cr results in partial fragmentation gap between the material and the top plate (hence electric (Fig. 50) instead of G-SPEL. field ∼ Φ/d), η is viscosity, γ is surface tension, and ε0 is −12 2 2 vacuum permittivity (ε0 = 8.85 × 10 C /Nm ). 4.6 Line edge roughness (LER) analysis If the rising time is known, one can calculate the poten- tial difference between the molten material and the plate, One of the advantages of SPEL is its ability to smooth out Φ, using the same formula. Take Cr and Si for instance, LER at the final stage. To quantify the improvement for the Applications of excimer laser in nanofabrication 35

Fig. 48 AFM images of Cr dots before and after SPELs. (a)and(d), original Cr pads of 20 nm high. (b)and(e), 48 nm high spherical dots after O-SPEL of 420 mJ/cm2.(c)and(f), 62 nm high cylinders after G-SPEL of 420 mJ/cm2 36 Q. Xia, S.Y. Chou Fig. 49 In G-SPEL with a 70 nm gap and a ﬂuence of 406 mJ/cm2, the original Cr lines of 280 nm wide and 62 nm height on SiO2 (a) become 130 nm wide (46% of original) and 130 nm tall (210% of original height) due to material rise-up during G-SPEL), a 452% (4.5 times) increase in aspect ratio (b). In G-SPEL with a 40 nm gap and a ﬂuence of 595 mJ/cm2, the original Si lines of 285 nm wide and 50 nm tall (c) become 175 nm wide (61% of original) and 90 nm height (180% of original height) (d)[33]

Fig. 50 With the same laser ﬂuence and same line width as in Fig. 49 (i.e., 280 nm) and a 70 nm gap, 35 nm thick Cr lines on thermal oxide (a) became dots which are connected with each other rather than rising up into narrower lines (b)

LER of Si and Cr lines, a fractal analysis based on high- ξ defines a representative lateral dimension of a rough line resolution SEM images of the lines was carried out. The first edge. If the distance between two edge points is within ξ, step was acquiring high-resolution SEM images of the grat- the heights at these two points can be considered correlated. ing lines. The edges of the lines were then recognized by However, if the separation of two edge points is much larger Matlab which uses Otsu’s threshold selection method [56]. than ξ, then we can say that the heights at these two points The algorithm in the LER analysis is based on the work of are independent of one another. Constantoudis et al. in reference [43]. The basic idea is as The 3σ LER before and after SPEL was measured using follows. First, the position of each point along the edge is a MatLab program. The results, together with the improve- recognized (the interval is 1 pixel in our analysis) and a ment of ξ, are listed in Table 4 [33]. straight line for the average position of the points is con- From Table 4, the analysis of SEM images of the Cr and structed by the least square method. Next, the distance from each point on the edge to the linear fit line is measured, and Si grating lines before and after open-, capped-, and guided- the standard deviation of the measured distance is consid- SPEL show that SPEL can reduce 3σ LER by as much as a ered the LER (1σ). factor of 5.6, such as from 8.4 nm to far less than 1.5 nm for However, the σ value is related to the vertical dimension 70 nm wide Cr lines. For each case, the ξ value increased of roughness (in the direction perpendicular to the fit line) after SPEL, which means the high frequency LER was im- and gives no information about its spatial complexity. It is proved significantly. It has to be pointed out that the parame- suggested that parameters for the description of LER should ters depend heavily on the imaging processing, the quality include: (1) the σ value, (2) the correlation length ξ, and of the SEM image, and other factors. As a result, the real σ (3) the roughness exponent α [43]. The correlation length could be smaller. Applications of excimer laser in nanofabrication 37 Table 4 Measured line edge roughness (LER) before and after SPEL for Si and Cr lines [33]

Samples Image Res. 3σ LER Improvement ξ (nm) Fluence (nm/pixel) (nm) (mJ/cm2)

Open-SPEL Cr (Fig. 33) Original 0.4 8.4 5.6 281.7 After 0.4 1.5 5.6 387.6 320 Si (Fig. 28) Original 1.7 19.55.4 30 After 1.7 3.6 5.4 60 545 Si (Fig. 30) Original 0.4 14.44.4 19.6 After 0.4 3.3 4.4 216.6 440 Si (Fig. 31) Original 1.7 9.62.3 23.0 After 1.7 4.22.3 38.9 880

Capped-SPEL Si (Fig. 43) Original 1.7 9.62.3 17.7 After 1.7 4.22.3 35.4 480 Si (Fig. 45) Original 1.7 11.12.1 37.2 After 1.7 5.4 2.1 125.6 390 Cr (Fig. 44) Original 3.4 17.72.4 27.6 After 3.4 7.5 2.4 117.3 406

Guided-SPEL Cr (Figs. 49a, b) Original 3.4 17.43.2 24.1 After 3.4 5.43.2 79.3 406

4.7 Applications of SPEL size over large areas. The first step is to turn non-ideal shaped metal nanopads into round dots using SPEL with a 4.7.1 Sub-25-nm smooth cylindrical NIL molds single laser pulse. Next, RIE is carefully carried out into the substrate, achieving features as small as 25 nm over wafer Fabrication of NIL molds with very small feature sizes is scale. technically challenging. One traditional method involves the use of electron beam lithography (EBL) [8] to pattern a poly- Principle The principle of our method is shown in Fig. 51 mer thin film on a substrate. After developing the polymeric [63]. First, metal pads are patterned on a substrate using resist, a thin layer of metal is deposited, followed by a liftoff NIL, metallization and a liftoff process. The edges of the process. The patterned metal layer serves as a hard mask pads are usually rough (Fig. 51a), which results from the during a pattern transfer process using RIE. Although EBL has been successfully demonstrated in making NIL molds fabrication environment and fundamental fabrication prin- with lines of 10 nm width (35 nm pitch) [57] and dots of ciples. Next, we do SPEL for these metal pads using a 10 nm diameter (40 nm pitch) [58], it is a serial process laser pulse. The metal is selectively and rapidly heated that is time-consuming and expensive. In addition, it is dif- into a molten state, and the surface tension turns the pads ficult to control the critical dimension and structure profile into round droplets (Figs. 51b, c). The round shape of due to proximity effects and wet chemical processing (i.e., these droplets is preserved after the metal is resolidified development). Other methods such as interference lithogra- (Fig. 51d). They are then used as hard masks for RIE phy [59], proton beam writing [60], focused-ion-beam writ- (Fig. 51e). After stripping off the metal masks, round pillars ing [61], and X-ray lithography [62] have also been used in with smooth sidewalls remain on the substrate (Fig. 51f). NIL mold fabrication, but they encounter problems similar to EBL. Furthermore, for all the aforementioned mold fab- Experiment and results To test the principle, an NIL- rication methods, defects are intrinsically generated due to patterned array of Cr rectangles was deposited on a fused the noise in lithography and the use of sequential processes silica substrate. These rough-edged pads were about 140 nm such as developing and etching. by 110 nm in size, and 20 nm thick (Fig. 52a). During the In this section, SPEL was used to fabricate round and following SPEL process, a single excimer laser pulse of 2 smooth NIL pillar molds in SiO2 with sub-25-nm feature 420 mJ/cm was used to turn these pads into near-perfect 38 Q. Xia, S.Y. Chou circular dots with a final diameter of 80 nm (Fig. 52b). Fur- ther AFM measurements showed that the height of these dots was about 48 nm. These round Cr dots were then used as a hard mask during RIE of the fused silica, using a gas mixture of 10 sccm CHF3 and 1.5 sccm O2 under a pressure of <3 mTorr with a power density of 60 mW/cm2 in a Plasma Therm 2486 etcher. Round pillars with a height of about 200 nm were created after RIE and metal stripping using Cr-7 etchant (Fig. 52d). With this etching recipe, these pillars had smooth sidewalls. The periodicity of the metal nanodots was maintained as well (Fig. 52f). As a comparison, the pillars which were etched using the unsmoothed Cr pads as etching mask are also shown in Figs. 52c, e. They inherited the rough edges from the original Cr mask. Successful imprints were achieved using the as-fabricated round pillar molds (Figs. 53 and 54). In both UV and thermal imprint, the holes in resists are round in shape and uniform Fig. 51 Schematic of our process to make cylindrical pillar molds. (a) Metal nanostructures with geometrical defects. (b) The nanostruc- in diameter. tures in (a) are melted upon exposure with a single laser pulse. (c)The Large-area NIL mold have been made by step and repeat molten material reflows into round dots to minimize the surface en- exposure. In order to cover the whole area, a stitching area ergy. (d) After re-solidification, the rough-edged pads turn into smooth (200 µm wide) between each laser spot was intentionally e round dots. ( ) These round metal dots are used as a hard mask for RIE. left (Fig. 55a). We have successfully made a mold with an (f) The metal mask is stripped off before the pillar mold is ready for use [63] effective size of 1 inch by 1 inch in a FS wafer (Fig. 55b). It is worth noting that the Cr dots were of the same high

Fig. 52 (a)Crsquaresof 140 nm by 110 nm and 20 nm thickness on a fused silica substrate. (b) After SPEL, the irregular squares in (a) became nearly perfect round dots of 80 nm diameter and 48 nm height. (c) Rough pillars etched into FS using Cr pads in (a)as an etching mask. (d) Smooth round pillars etched into FS using round Cr dots in (b)as etching mask. (e)and(f)aretop view images of pillars in (c)and(d), respectively, which show the etching ﬁdelity during RIE [63] Applications of excimer laser in nanofabrication 39 quality and same size in the areas that experienced only one in addition to shrinking the size of the original nanostruc- or multiple pulses (Figs. 55c, d), suggesting high processing tures, the size variation of the metal dots is also reduced at tolerance of our process. The current area of the mold is not the same time while keeping the original periodicity. Fourth, the limit of our technology, as more step and repeat exposure the shape of the metal dots (masks) can be almost perfectly will result in a larger area. round due to the high surface tension of molten metal. Since the process is ultrafast, the ideal shape of a liquid droplet Discussions SPEL has several advantages in making is preserved after re-solidiﬁcation. This gives an ideal shape smaller feature size molds with fewer defects. First, a for the etched pillars. Fifth, the mold area is relatively large smaller feature size structure can be made from a structure and the process has high throughput. Sixth, as an ultrafast that is initially relatively large in size. This extends the ca- technology, the laser pulse can heat the materials on all kinds pability of lithographic tools into a smaller size regime. Sec- of substrate selectively, so the resolution of the approach is ond, the size can be tuned by tailoring the amount of materi- not limited by the substrate (for EBL, the electron charging als deposited on each site before SPEL, as well as the inter- effect is substrate-sensitive), hence one cannot make very facial properties between the metal and the substrate. Third, small feature sizes on non-conductive substrates.

Fig. 54 25 nm diameter hole array in thermal plastic resist (NPR-69) Fig. 53 Imprint in PUV-30 with the round FS pillar mold, resulting in imprinted using a Si pillar mold that was fabricated by SPEL on Cr round holes in the resist with uniform size dots and RIE [63]

Fig. 55 (a) Schematic of step and repeat exposure with overlap (shadowed area) in SPEL; (b) Photograph of a FS mold which is about 1 inch in size; (c)and(d)areSEM images from spot A and B in (a), which shows that the double exposure in the overlapped area during SPEL results in the same high quality round metal dots as etching masks [63] 40 Q. Xia, S.Y. Chou As discussed in earlier sections, the shape and size of the After removal of Cr, the silicon device layer was etched us- Cr dots can be changed; cylinders with a smaller diameter ing chlorine base RIE with the thermal SiO2 as a mask. can be created if G-SPEL has been conducted for the orig- The wafers were then briefly dipped in a diluted HF solu- inal Cr pads. As a result, smaller FS pillar with better side- tion to remove the top oxide layer. Finally, the samples were ◦ wall profile can be achieved after RIE. cleaned in RCA #1 solution at 80 C for 15 min before the In summary, SPEL has been used to make sub-25 nm samples were ready for SPEL experiments. round pillar molds with smooth sidewalls. The molds were used for successful nanoimprinting. The size of the final Results Multiple pulses were used for the best smooth- structure can be tuned by tailoring the amount of mask ma- ing results for the microscale waveguides. For example, the terial deposited at each site, and the interfacial properties rough sidewalls of a 4 µm wide waveguide (Fig. 56a) were 2 between the metal and the substrate. This method works for smoothed out after exposure to 20 pulses of 900 mJ/cm different substrates with high throughput and low cost, and with a repetition rate of 1 Hz (Fig. 56b). LER analy- it extends the capability of other lithographic methods into sis showed that roughness was reduced from 13 to 3 nm anewregime. (Fig. 57). It should be pointed out that the profile of the Si lines 4.7.2 Smoothing of Si waveguides changed from square to semi-circular. This might be helpful in optical coupling between the Si waveguides and round optical fibers. However, in the case when the square profile Integrating optical components on a chip with high packing needs to be preserved, C-SPEL can serve this purpose. density using existing Si fabrication technology is currently the topic of extensive research [64]. However, as device fea- Calculation of propagation loss The propagation loss is a ture sizes shrink, geometric defects such as sidewall rough- function of different sidewall roughness and roughness au- ness introduced in the fabrication process, will become more tocorrelation lengths. An SOI strip waveguide with a rec- prominent and will adversely affect light propagation prop- tangular cross section 500 nm wide and 200 nm high was erties. Previous methods for reducing sidewall roughness, considered. The propagation loss was then plotted as a such as anisotropic wet etching [65, 66], thermal oxidation function of σ and Lc (autocorrelation length). For exam- [67], or a combination of the two [65, 68], are either lim- ple, Fig. 58 shows the waveguide loss contours for a Si ited to certain crystalline facets of semiconductor materi- waveguide 200 nm high and 500 nm wide. It clearly shows als, or often involve harsh processing conditions and high that the waveguide transmission loss reduces with smaller temperatures. These methods may also affect other compo- correlation length and smoother surface. Figure 59 plots the nents and/or materials on the same chip that are inadver- loss as a function of waveguide sidewall roughness for a tently processed at the same time. In this section, the appli- 200 nm high Si waveguide with a width of half and quarter cation of SPEL in smoothing Si waveguides and its implica- wavelength used in telecommunication (i.e., 1.55 µm), as- tion in the reduction of optical propagation loss is discussed. suming an autocorrelation length of 5 nm. It indicates that for a 220 nm wide waveguide, a reduction of roughness Methods and experiments Micro-scale (4–10 µm) Si wave- from 13 nm to 3 nm can result in a decrease of propaga- guides were fabricated on a bonded SOI wafer (1.5 µm thick tion loss from 53 to 3 dB/cm. This means the transmitted Si layer and 1 µm thick buried oxide) [69]. A thin layer of power could increase by 5 orders of magnitude for a 1 cm thermal SiO2 was grown on the surface of the SOI wafer, long Si waveguide. followed by waveguide patterning using photo-lithography. As in the previous section, wafer scale SPEL has been A 20 nm thick Cr layer was evaporated using an electron achieved by a step and repeat exposure system for Si beam evaporator, followed by a liftoff process. The Cr pat- waveguides. An exposure overlap of about 200 µm wide terns were used as etching masks for RIE the thin SiO2 layer. was used between each pulse spot. The smoothing results

Fig. 56 (a) A 4 µm wide Si waveguide on SiO2 with rough sidewalls; (b) the waveguide was smoothed after exposure to 20 pulses with 900 mJ/cm2 at a repetition rate of 1 Hz [69] Applications of excimer laser in nanofabrication 41 Fig. 57 Line edge proﬁle along the length direction of a 4 µm wide Si waveguide on SiO2 before (a)andafter(b) SPEL. The 1σ LER in (a)and(b)are 13 and 3 nm, respectively [69]

Fig. 58 Waveguide loss contours for a Si waveguide of 200 nm high Fig. 59 Calculated waveguide transmission loss versus the roughness and 500 nm wide. The light wavelength used was 1.55 µm (in air) [69] for Si waveguides with a width of half (220 nm) and quarter (110 nm) wavelength. In this calculation, the waveguides are 200 nm high with an autocorrelation length of 5 nm. The light wavelength used was 1.55 µm (in air) [69] in the overlapping area had no difference with other areas, suggesting no rigorous alignment is required in SPEL for a large area. It was found that a fluence window was about In summary, we have used SPEL to smooth the side- 25% for smoothing the nanoscale Si gratings, indicating a wall roughness of Si waveguides. Our experimental results high tolerance for energy fluctuation from pulse to pulse for showed that waveguide sidewall roughness was reduced SPEL. The features make the SPEL process simple to im- from 13 to 3 nm using this technique. With this reduction in plement. the sidewall roughness, transmission loss in the waveguide 42 Q. Xia, S.Y. Chou (200 nm by 500 nm) decreased from 53 to 3 dB/cm ac- 5 Nanoparticle arrays fabricated by pulsed laser cording to our calculation. As an ultrafast, highly selective melting method, SPEL will be increasingly important when the size of silicon waveguides shrink for higher density nanopho- 5.1 Introduction tonics. The advantages of SPEL will become more prominent as waveguide sizes shrink to allow for single optical Metal nanoparticles have wide applications in catalysis [70], mode transmission and increased waveguide packing den- environmental remediation [71], DNA detection [72], high sity, since the effect of sidewall roughness in increasing density data storage [73], and electronic and optical devices transmission loss becomes greater with smaller waveguide [74, 75]. In certain applications, one and only one layer (i.e. sizes. In addition, as chips become hybridized with optical, monolayer) of metal nanoparticles is required. To achieve electronic, and other functions, chip fabrication technology metal nanoparticle monolayers, drop-casting [73] or spin- must keep pace. SPEL is an excellent candidate for highly coating [76] of nanoparticle colloidal solutions on a sub- selective fabrication of ultra-smooth surfaces. strate followed by solvent evaporation, or self-assembly using a Langmuir–Blodgett technique [77] have been used. However, these techniques involve wet chemical reactions 4.8 Summary and lengthy processing times that increase cost. Further- more, they often require thermal annealing to improve the Three forms of SPEL and their applications have been ex- adhesion of the particles to the substrate [73]. This not only perimentally demonstrated in this section. Using O-SPEL, increases the cost, but is also incompatible with substrates rough edges of nanoscale lines have been smoothed and non- like plastics that cannot withstand high temperatures. Previ- ideal shaped nanopads have been turned into a nearly perfect ously, an excimer laser has been used to change the shape shape. With O-SPEL, 3σ LER of rough edged 70 nm wide of noble metal islands on a quartz substrate into spherical Cr lines have been reduced from 8.4 nm to 1.5 nm, which is nanoparticles [78–81]. Usually, multiple laser pulses were well beyond the “red-zone limit” on the current technology used for this process. roadmap. With C-SPEL, the LER can be removed and the In this section, a simple method for manufacturing metal cross-section profile can be maintained. C-SPEL preserves nanoparticle monolayers using a single laser pulse to melt metal thin films and metal lines on various substrates is dis- the same height, flat tops and vertical sidewalls of the nanos- cussed. The method starts with a thin metal film deposited tructures. Furthermore, during G-SPEL the molten nanos- on a substrate or metal lines patterned by NIL, followed tructures rise upon their own and reach the plate, reshaping by melting using a single XeCl excimer laser pulse. Dur- themselves into structures that have not only smooth edges ing laser melting, the originally continuous metal thin film but also vertical sidewalls, flat tops and narrower width for or metal lines break into a monolayer of partial spherical both lines and dots. Applications of SPEL have been ex- nanoparticles or an array of metal particles due to the dewet- emplified by making smooth round NIL pillar molds and ting of the metal from the substrate surface. Various metals smooth Si waveguides that have propagation loss orders of (e.g., noble metals (Ag, Au, Pt), magnetic metals (Fe, Co, magnitude lower. Ni, Cr), refractory metals (W), etc.) on different substrates The novelty and major contribution of this work are (silicon, fused silica and plastics) were studied. The effect of as follows. (a) Different from previous work in isother- substrate material, film thickness, laser fluence, and ambient mal smoothing of photoresists, LER and other defects in conditions on the formation of nanoparticles were explored. nanostructures of high melting temperature materials (metals and semiconductor) are repaired directly. (b) Differ- 5.2 Metal nanoparticles from continuous films ent from other work such as lithographically induced self- assembly (LISA) [53] which works with a continuous film, 5.2.1 Experiments the current work uses of a second surface (the plate above) to guide the reshaping of pre-patterned structures. (c) It is ob- The substrates used in this study were silicon, fused silica served for the first time that liquid metals and semiconduc- and plastic (e.g., 1/16 thick polycarbonate sheets). Silicon tors flow upward against surface tension between the two and fused silica wafers were first cleaned using a RCA #1 plates. Previously, similar behavior was only observed in solution at 80◦C for 30 min. Some silicon wafers were then polymer systems. (d) The current work provides a method to dipped in diluted HF acid (1:50) for one minute to remove enhance the nanostructures profile into smaller feature size the native oxide before the metal deposition. The plastics and higher aspect ratio. This extends the lithography capa- were cleaned in boiling DI water with a small amount of bility to a new regime with only a set of simple boundary Micro-90 cleaning solution for 10 min, followed by a 30- conditions. min ultrasonic bath, and then rinsed in running DI water Applications of excimer laser in nanofabrication 43 for 10 min. Noble metals (Au, Ag, Pt), magnetic metals 740 mJ/cm2, respectively, monolayers of Pt nanoparticles and alloys (Fe, Co, Ni, Cr, permalloy), and refractory metals were achieved (Figs. 60a, b). These particles have almost (W) were investigated. Thin metal films (2 to 20 nm thick) the same size distribution (Figs. 60d, e), 9.8 ± 4.2nmfor were deposited on cleaned substrates using an electron beam (d) and 10.1 ± 4.4 nm for (e). However, for a 5 nm thick Pt − evaporator under a base vacuum better than 2 × 10 6 Torr. film on the silicon wafer without native oxide, the resultant All the as-deposited thin films were continuous except for structures are not particles (Fig. 60c). three cases: Ag and Au films on Si, and 2.1 nm thick Pt film on fused silica, as examined by an SEM. Metal nanoparticles on plastic substrate Metal nanopar- The excimer laser was used to melt the metal films on ticle monolayers on plastic substrates were also fabricated. substrates. In order to cover a large area, a step and repeat A single laser pulse of 160 mJ/cm2 incident on a 10 nm exposure scheme was used. A single pulse was found to be thick Ag film on a polycarbonate substrate produces a mono- sufficient to fragment a metal film into nanoparticles. Most layer of Ag nanoparticles (Fig. 61). There is no signifi- of the melting experiments were carried out in air, while oth- cant damage to the substrate due to the short pulse width ers were done in liquids such as acetonitrile to avoid oxida- of the laser and the fact that polycarbonate does not sub- tion of the nanoparticles and to study the effect of interfacial stantially absorb the 308 nm wavelength UV laser (<1% environments. Size distribution of the nanoparticles was an- for a 20 µm thick sheet [82]). Recently, low-temperature alyzed from SEM pictures using commercial software (Im- plasma enhanced chemical vapor deposition (PECVD) has age Pro-Plus) [44]. UV-Vis absorption spectra were mea- been introduced to grow carbon nanotubes (CNT) on various sured for Ag nanoparticle monolayers on UV-grade fused substrates [83, 84]. Our current result suggests that smaller silica using a UV-Vis spectrophotometer scanning from 200 metal catalyst particles can be made on plastic substrates. to 800 nm. The chemical composition of some nanoparticles This may create an opportunity for making smaller and more was analyzed using energy dispersive X-ray analysis (EDX). uniform CNTs for devices on plastics, such as flexible dis- plays. 5.2.2 Results

After a single laser pulse exposure, continuous thin films of Alloy metal nanoparticles In addition to pure metals, most metals (except Ti on a Si substrate) were turned into permalloy nanoparticle monolayers were also fabricated by nanoparticle monolayers. Part of the metallic thin films, ex- the same technique. These nanoparticles have been used as perimental conditions and results are tabulated in Table 5. catalysts for carbon nanotube growth by chemical vapor deposition, increasing the diameter uniformity and yield. Pure metal nanoparticles For almost all the pure metals Figure 62a shows a 2 nm thick permalloy film on a Si (Au, Ag, Pt, Fe, Co, Ni, W, etc.) tested in this study, sub- substrate after exposure to a single laser pulse with a fluence 100 nm diameter nanoparticles were achieved; provided that of 1 J/cm2. The particles have a diameter distribution cen- a proper substrate and laser fluence were used. For exam- tered at about 10 nm (10.4 ± 3.8nm)(Fig.62b). After CNT ple, 5 nm Pt was deposited on fused silica, Si (with na- growth with 1 LPM (liters per minute) ethylene at 700◦C tive oxide) wafers. After a single laser pulse of 680 and for 10 min, the typical TEM image shows uniform CNTs

Table 5 Partial list of metallic thin ﬁlms and resultant particle sizes

Category Film Thickness Substrate Fluence Particle size Material (nm) (mJ/cm2)(nm)

Nobel Aua 10 Fused Silica 194 106.6 ± 55.8 Agb 5 Siliconc 310 25.2 ± 12.6 Pt 5 Fused Silica 680 9.8 ± 4.2

Magnetic Co 5 Silicon 736 21.5 ± 5.3 Ni 5 Silicon 1000 23.1 ± 10.1 Permalloy 5 Silicon 1000 23.2 ± 6.8

Refractory W 6 Silicon 640 15.9 ± 8.9 aWith 2 nm Ti as sticking layer bAg ﬁlms are exposed to laser in acetonitrile cSi substrates in this table are all with native oxide 44 Q. Xia, S.Y. Chou

Fig. 60 SEM micrograph of the metal nanostructures produced from became a monolayer of Pt nanoparticles, while in (c) there were no 5 nm thick Pt film on (a) fused silica substrate, (b) silicon with na- particles formed. (d), (e) are the diameter histograms for nanoparticles tive oxide, and (c) silicon without oxide, with a laser fluence of 680, in (a)and(b), respectively. Note they are both centered at 10 nm with 740 and 740 mJ/cm2, respectively. Only in (a)and(b), the metal film similar distribution Applications of excimer laser in nanofabrication 45 Large-area nanoparticles for optical applications Al- though the laser spot is only several mm in size, a larger coverage area can be achieved by a step and repeat exposure system. A 5 nm thick Ag film was deposited on a piece of 0.5 mm thick UV grade fused silica (2 cm by 1 cm). Step and repeat exposure was carried out for Ag films immersed in acetonitrile, a solvent that does not absorb 308 nm UV light. Each spot was exposed to a single pulse of ∼310 mJ/cm2. The average particle size was about 25 nm. A typical UV-Vis spectrum of the resulting nanoparticle monolayers (Fig. 64) exhibited an absorption peak at 395 nm.

5.2.3 Discussion

Effect of substrate Nanoparticle monolayers were ob- 2 Fig. 61 Ag nanoparticles fabricated by a 160 mJ/cm laser pulse inci- tained for metal thin films on fused silica and silicon sub- dent on a 10 nm thick silver film on a polycarbonate substrate strates covered with native oxide. However, this is not the case for silicon substrate that had been dipped in diluted HF (Fig. 62c). One hundred arbitrary selected CNTs from the acid. This suggests that molten Pt de-wets with SiO2,but TEM images are measured, and the diameter distribution is not with Si, which might be a result of reactions between centered around 10 nm (10.9 ± 1.9nm)(Fig.62d). For com- the molten Pt and Si substrate (Fig. 60). parison, the typical image and diameter distribution of CNTs grown in the unexposed area are also shown (Figs. 62e, f). Dependence of particle size on film thickness The parti- CNTs grown directly on the as-deposited film had a diame- cle size is dependent on the film thickness. Under otherwise identical conditions, a thicker film usually results in larger ter distribution centered at 15 nm (14.8 ± 2.5 nm). Compar- particles. In order to explore the correlation between film ison between Figs. 62d and 62f shows that CNTs grown on thickness and particle size, Pt thin films with thickness of the nanoparticles produced by laser exposure have smaller 2.1, 5.1, 10.0, 16.1 and 20.8 nm were deposited on fused diameters and a narrower size distribution. Laser exposure silica substrates and exposed to single laser pulses with the Fe films also enhanced the yield for SWNTs growth when same laser fluence (∼650 mJ/cm2). There was a linear rela- methane was used as the precursor gas. tionship between the average particle size and the film thickness except for the 2.1 nm case (Fig. 65). It was suggested Particle formation and alloying with a single pulse Bi- that the particle formation is a spinodal dewetting process layer metal films can be turned into alloy nanoparticles with [86] in which the particle size should be a monotonic in- single pulse laser exposure as well. For example, a 4 nm creasing function of film thickness. The disruptive case for thick Cu film was deposited on a silicon substrate, followed the 2.1 nm thick film might be a result of the fact that the bya5nmthickNifilmdeposition. A laser pulse with flu- film is actually isolated islands or networks and not contin- 2 ence of 1130 mJ/cm breaks the thin film into nanoparticles uous. Upon laser exposure, the islands shrink into spheres with an average size of about 48 nm. To determine the chem- rather than breaking into even smaller ones. The typical laser ical composition of the nanoparticles from bi-layer metals, induced shape/size change described in references [78–81] EDX was used on single nanoparticles. In order to avoid occurs. the strong Si background signal in EDX, larger particles of heavier metals were made from thicker metal films. For ex- Effect of laser fluence The laser fluence required to ob- ample, 6.6 nm Ni and 4.4 nm W thin films were deposited on tain nanoparticles varies from metal to metal. However, for a Si substrate, followed by exposure to a single laser pulse each particular metal (alloy), there is a laser fluence win- (500 mJ/cm2). A typical EDX spectrum of a single particle dow in which the resultant nanoparticles have similar size isshowninFig.63, which exhibits both Ni and W peaks, distributions. The fluence window can be as large as 100% indicating that the particles are NiW alloys rather than pure for many metals or alloys on both silicon (with native ox- Ni or W. This suggests that nanoparticle formation and al- ide) and fused silica substrates (Fig. 66). This means that loying can be achieved at the same time during a single laser as long as the laser fluence is high enough to melt the thin pulse exposure. Similar results were achieved for Pt–Co and film without noticeable evaporation, varying the laser flu- Pt–Ni systems, which are widely used as catalysts for fuel ence does not change the size distribution of the nanopar- cell applications [85]. ticles. This finding indicates that our approach has a high 46 Q. Xia, S.Y. Chou

Fig. 62 (a) SEM image of nanoparticles resulting from a 2 nm thick catalysts shown in (a), and (d) the corresponding MWNT diameter his- permalloy film after exposure to a single pulse laser with a fluence togram; (e) a typical TEM image of MWNTs grown using as-deposited of 1 J/cm2;(b) histogram of the diameter for permalloy nanoparticle 2 nm thick permalloy film as a catalyst, and (f) their diameter distribu- in (a); (c) a typical TEM image of MWNTs grown with nanoparticle tion histogram Applications of excimer laser in nanofabrication 47

Fig. 65 Dependence of the average Pt particle diameter on the initial film thickness for fused silica substrate. Particle diameter is propor- Fig. 63 Typical EDX spectrum of a single particle made from tional to the film thickness from 5.1 to 20.8 nm, but not at 2.1 nm Ni/Wbi-layer metal thin film, suggesting that the particle is an alloy. The inset SEM image shows the NiW particles used for the analysis which were intentionally made large for the ease of EDX analysis. Scale bar: 200 nm

Fig. 66 The average particle diameter as a function of laser fluences for different metal thin films on silicon or fused silica substrates. Note that particle diameter is insensitive to laser fluence within a certain Fig. 64 UV-Vis absorption spectrum for a Ag nanoparticle monolayer range on a fused silica surface with an absorption peak at 395 nm. Inset is the SEM image of the nanoparticles which have an average size of 25 nm of a 2 nm thick film, confirming that the particle size is related to the original film thickness. tolerance for energy fluctuation from pulse to pulse, typi- cally a problem for gas lasers. Two other important points 5.3 Nanoparticle arrays by laser fragmentation of metal can be drawn from Fig. 66. First, under otherwise identical lines conditions, a 5 nm thick Ag film results in larger particles than those for a 5 nm thick Co or permalloy film. This might As shown in Table 5, the nanoparticles formed by expos- be attributed to the difference in the surface tension between ing a continuous metal film to laser pulse have a wide size Ag and Co (or permalloy). Since Co has a higher surface distribution, and exhibit almost no periodicity. In order to tension than Ag [87, 88], the liquid droplets have a stronger control the particles periodicity and size distribution, NIL is tendency to shrink into a smaller size. Second, a 5 nm thick first used to pattern the metal films. Since the lines are iso- permalloy thin film results in a larger particle size than that lated from each other, this method is expected to improve 48 Q. Xia, S.Y. Chou

Fig. 67 Au nanodots formed by fragmentation of a blank thin film is 65.2 ± 10.4 nm; (f) FFT image for (d), showing a periodicity of (left column) and a 200 nm pitch, 100 nm line width grating (right 220 ± 70 nm along the original grating line direction and the original column) on fused silica substrates. (a) SEM image of nanodots from a 200 ± 9 nm grating period in the orthogonal direction. In both cases continuous film; (b) histogram of the particle size distribution for (a), the films are 10 nm thick (with 2 nm Ti adhesion layer), and the laser the size is 106.6 ± 55.8 nm; (c) FFT image for (a), showing no regular fluences are 194 mJ/cm2. Insets in (a)and(d) are schematics of the periodicity for the particles in (a); (d) SEM image of nanodots from starting structures (blank thin film and nanogratings) on substrates [89] Au lines; (e) histogram of the particle size distribution for (d), the size both the periodicity and the size distribution of the resultant ration and liftoff. Single laser pulses of 194 mJ/cm2 were nanoparticles. used for exposing both samples. With a continuous film, To demonstrate the idea, 10 nm thick Au film was de- the particle size is 106.6 ± 55.8 nm with a broad distrib- posited on a fused silica substrate, with a 2 nm thick Ti film ution of period (Figs. 67a, b, c). However, with the same as the sticking layer [89]. Meanwhile, 10 nm Au/2 nm Ti laser fluence, the diameter of the nanodots formed from the metal lines of 200 nm pitch and 100 nm width were fab- 200 nm pitch gratings (10 nm Au/ 2nm Ti on fused silica) ricated on fused silica substrates using NIL, metal evapo- was 65.2 ± 10.4 nm, both the size and size distribution have Applications of excimer laser in nanofabrication 49 been improved. The period along the original grating line direction is 220 ± 70 nm and that perpendicular to the grating line is 200 ± 9 nm (the same as the original grating period) (Figs. 67d, e and f). The relative standard variations (1 sigma) of the size and pitch distribution were significantly better than those of nanodots from a blank thin film. The fragmentation of molten metal lines into dot arrays can be explained using the Rayleigh instability theory [90]. According to this theory, a liquid cylinder of a radius R will become unstable and starts to break into periodic droplets of a critical wavelength, λc. If it is perturbed along the cylinder longitudinal direction, the critical wavelength can be calculated using λc = 2πR√ and the maximum (equilibrium) wavelength is λm = 2 2πR [91]. We find the pitch of the nanodots along the original grating lines agrees with Rayleigh instability model. For example, 80 nm wide Cr lines of 200 nm pitch broke into the dots with an average period of about 250 nm along the original grating line direction. The ratio of the nanodot pitch to the original grating line width is 3.12, which is close to the predicted value (π).

5.4 Periodicity engineering of nanoparticle arrays Fig. 68 Principle of periodicity engineering using difference of wettability. (a) Patterning lines of adhesion material are deposited on a To further improve the periodicity along the nanograting di- substrate; (b) Lines of particle material are deposited with an angle (adjustable) to the first layer of lines; (c) During the fragmentation rection, two approaches were proposed. The first approach process, the molten particle material directly in touch with the non- utilizes differences in wettability and the second approach wetting substrate flows to the cross points of the two materials due to takes advantage of surface topography. Figure 68 shows the the difference in the wettability, maintaining a regular periodicity de- principle of the first approach. Lines of adhesion material termined by the two layers of lines (A) are patterned and deposited on a substrate (Fig. 68a). Then, another set of lines are patterned at an angle to the trenches were used. Then 2 nm thick Ti and 10 nm thick Au first ones. The angle can be adjusted according to the appli- lines were deposited using NIL, O2 RIE, metal evaporation cation. The particle material (B) is then deposited, forming and liftoff. Similarly, 7 nm thick permalloy lines were de- the crossbar structure (Fig. 68b). This set of lines is either posited on 210 nm thick SiO2 capped Si wafer with 10 nm in contact with the substrate, which is non-wetting to mate- deep trenches that were 70 nm wide and had 200 nm pitch. rial B, or the adhesion lines, which are wetting to material The metal lines are nearly perpendicular to the trench di- B. During the fragmentation process (e.g., pulsed laser melt- rections in both cases. After exposing the samples to a sin- ing), the molten material B will preferably flow to the cross gle laser pulse of 325 and 790 mJ/cm2 for Au and permal- points where A and B join due to the better adhesion at those loy, respectively, regular arrays of metal nanoparticles were areas, hence forming nanoparticle arrays with regular peri- made on the substrates (Figs. 70a, c). The periodicity along odicity predetermined by the lines of A and B (Fig. 68c). the original metal lines and trenches is shown in the FFT Another approach is to use surface topography differ- images (Figs. 70b, d). A closer look at the SEM images ences, as shown in Fig. 69. Shallow trenches are first pat- (insets in both Figs. 70a and 70c) clearly indicates that the terned and etched into the substrate (Fig. 69a), followed by resultant nanoparticles are sitting in the cross points of the metal line deposition at an angle to the trenches (Fig. 69b). original metal lines and the shallow trenches, favoring min- During the fragmentation process, the molten material tends imum system energy. The fast Fourier transfer (FFT) analy- to flow to the trench to minimize the system energy and stays sis showed the periodicity of nanodots in the original metal at the cross points of the trench and the original metal lines, line direction was 200.0 ± 11.3 nm and in the normal di- resulting in a regular metal nanoparticle array with period- rection was 200.0 ± 6.9 nm (Fig. 70b) for Au, and those icity determined by the shallow trenches and the pitch of the for permalloy were 200.0 ± 7.4 nm and 200.0 ± 5.8nm, original metal lines. respectively (Fig. 70d). Compared with the nanodot arrays To test the principle, both Au and permalloy nanograt- by fragmentation of a blank Au film (Fig. 67a, c) (which ings were fabricated. For Au lines, fused silica substrate had no certain periodicity), and those by fragmentation of with 10 nm deep, 70 nm wide and 200 nm pitch shallow Au nanogratings on a flat surface (Fig. 67d, f) (which had a 50 Q. Xia, S.Y. Chou √ than the maximum wavelength 2 2πR, one shall expect that there will be multiple dots formed at one intersection point. This is important because it suggests that complex nanodot arrays can be formed by engineering the surface topography. In summary, we have used a single laser pulse to fabricate various kinds of metal nanoparticle monolayers on silicon, fused silica and plastic substrates. The particle size was mainly determined by the film thickness, while other factors, such as the laser fluence, have little effect on the size distribution. Nanoparticle formation and nano-alloying can be achieved during a single laser pulse exposure of bi- layer metal thin films. A large coverage area was achieved using a step and repeat method, and optical properties of Ag nanoparticles on fused silica were studied. To get metal nanoparticles with better periodicity, metal nanogratings were used as the starting structure. This approach not only improves the size distribution, but also introduced a certain degree of periodicity along the original grating direction according to Rayleigh instability. The periodicity of the nanodot arrays was further engineered using a surface relief Fig. 69 Principle of periodicity engineering using surface topography. structure, resulting in a regular 2D array of nanodots with (a) Patterning of shallow trenches on a substrate; (b) Lines of particle regular periodicity along both directions. As a simple fabri- material are deposited with an angle (adjustable) to the trench direc- cation method, it could be extended to other metals and has tion; (c) During the fragmentation process, the molten particle material flows into the trenches and resolidifies at the cross points of the trench wide applications in many areas such as magnetics, plas- and the metal lines due to a lower energy on those sites, maintaining a monics, surface enhanced Raman scattering and other pho- regular periodicity determined by the original metal lines and the shal- tonic devices. low trenches [89] 6 Sub-10-nm self-enclosed nanofluidic channels pitch of 220 ± 70 nm in the original grating line direction), the pitches of the nanodot arrays in Fig. 70 were much more 6.1 Introduction uniform because they were predetermined by the pitches of the shallow trenches on the substrate surface and the origi- Nanofluidic channels are important tools in the emerging nal metal nanogratings. It is also interesting to note that in bionanotechnology field for manipulation and analysis of Fig. 70, the permalloy has a more uniform periodicity than biomolecules at the single molecule level [92]. Fabrication Au in both directions. There are also some missing dots in of extremely small channels (and arrays) requires state-of- the Au nanodots array and some cloud in the FFT image the-art nanopatterning techniques. Equally important is the (Figs. 70a, b), a possible reason for this is that the migration sealing of trenches into functional channels. For the pat- of Au at molten state because it has higher mobility. terning of nanoscale trenches, EBL [93, 94] or focused ion- The results in Fig. 70 suggest that although the pitch of beam (FIB) milling techniques [95] have been widely used. nanodots depends heavily on the original nanowire width However, these methods suffer from low throughput and according to Rayleigh instability theory, they can be regu- are expensive. The advent of NIL [8–10] has greatly alle- lated using substrate surface topography. As a result, the fi- viated some of these problems. Using NIL, low-cost wafer nal nanostructure will be determined by the interplay of the scale fabrication of high density nanofluidic channel arrays natural instability process and the substrate surface topogra- for DNA manipulation and analysis has been demonstrated phy. Figure 71 summarized the possible nanostructures af- [96–98]. ter fragmentation of nanogratings on different surface relief The other challenge, i.e., sealing of nanochannels is not structures. When the pitch of the shallow trenches√ is smaller as easy as it sounds. Soft elastomers such as PDMS were than the maximum instability wavelength (2 2πR), there used to seal the channels [99], resulting in a uniform sealing will be only one dot at each trench/grating intersection point. over a larger area. However, channels might become clogged And the pitch of the nanodots along the original grating di- since the soft material can be easily pressed into the channel. rection is determined by the original trench pitch. On the Wafer bonding techniques [100], on the other hand, provide other hand, if the pitch of the surface relief structure is larger a rigid seal, but are limited because they require defect free Applications of excimer laser in nanofabrication 51

Fig. 70 SEM and FFT images for regular Au and permalloy nanodot odot arrays fabricated by fragmentation of permalloy gratings (200 nm arrays on pre-patterned surfaces with shallow trenches. (a)Aunan- pitch, 70 nm width, 7 nm thick) on a pre-patterned 210 nm thick odot arrays fabricated by fragmentation of Au gratings (200 nm pitch, SiO2 capped Si wafer. The dot size is 72.5 ± 6.8 nm. Laser ﬂuence: 100 nm width, 10 nm thick, with 2 nm thick Ti as adhesion layer) 790 mJ/cm2.(d) FFT image for (c). The substrates have trenches of on pre-patterned fused silica substrate. The dot size is 80.1 ± 8.1nm. 200 nm pitch, 70 nm linewidth which are 10 nm deep [89] Laser ﬂuence: 325 mJ/cm2.(b) FFT image for (a). (c) Permalloy nan-

√ Fig. 71 Summary of the relationship between the surface trench pitch ing with a pitch Λ<2 2πR, λ is determined√ by the trench pitch Λ, and the resulting nanodot pitch. (a) For a flat surface, Λ =∞, the pitch λ = Λ;(c) for the surface trenches with Λ>2 2πR, λ<Λ, resulting of the nanodot arrays along the original grating direction (λ) is about in particle doublet (or triplet, etc.) arrays [89] 2πR, agreeing well with Rayleigh instability; (b) for the surface grat- surfaces. Cracks might be introduced when there is a mis- techniques such as non-uniform deposition [96] and imprint match in the thermal expansion between the substrate and of trench molds into thin polymer films [98] address some of the cover material. Sealing of nanochannels has also been these concerns, although they may not be suitable for com- done by deposition of sealing materials over sacrificial lay- plicated biochips with different functional devices. ers such as polysilicon or polymers followed by wet etch- In this section, a self-sealing technique to make enclosed ing or thermal decomposition [101, 102]. These methods nanofluidic channel arrays from NIL patterned silicon nan- usually require long etching times to remove all the sacri- otrenches and pillar arrays is introduced. During the sealing ficial material and become increasingly difficult for smaller process, a single laser pulse is used to melt the top layer nanochannels where the flow rates are very low. Sealing of the nanostructures, inducing lateral flow which seals the 52 Q. Xia, S.Y. Chou nanostructures before resolidification. The channel dimen- Petersburg, FL) using a CHF3/O2 chemistry, with 10 sccm 2 sions can be controlled by the laser fluence with fine size CHF3, 1.5 sccm O2, 5 mTorr base pressure and 70 mW/cm control by secondary thermal oxidation. power density. The patterned thermal SiO2 served as a hard mask for silicon etching using deep RIE (STS, Newport, 6.2 Self-sealing of Si trenches UK). The Si etching had several cycles depending on the desired depth. Each cycle started with an etch step (30 sccm The principle of our method is shown in Fig. 72 [103]. Sil- SF6, 6 sccm O2, 600 W ICP power, 12 W platen power, icon nanostructures (made by NIL) (Fig. 72a) are exposed 15 mTorr base pressure, 7 sec), followed by a passivation to a UV laser pulse, which melts the top portion of the step (85 sccm C4F8, 600 W ICP power, 15 mTorr base pres- structures (Fig. 72b). Due to the surface tension, the orig- sure, 5 sec). The etched samples (200 nm pitch trenches with inal rectangular cross section of the etched nanostructures different widths and depths, and 700 nm diameter pillars of becomes circular in shape during the reflow process. With 970 nm pitch and 2 µm height) were dipped into diluted HF sufficient melting, adjacent structures join together to form to strip the SiO2 mask, followed by a further cleaning using an enclosed structure (Fig. 72c), and re-solidify (Fig. 72d). RCA #1 for 15 min. As an option, thermal oxidation can be carried out on these The fabricated structures were then exposed to the ex- structures to further shrink the size (Fig. 72e), and to make cimer laser with a spot size of 3 × 3mm2. In each case, the cover optically transparent. a single laser pulse with an appropriate fluence was used The fabrication of the self-enclosed Si channels included to melt the top of the structures just enough to result in a the following main steps: First, fabrication of the nanoscale uniform seal over the exposed area. Some channel samples trenches (or pillars) on Si using NIL and RIE; second, laser were then put into a Tystar oxidation tube (Tystar Corp, CA) melting to seal the trenches and form channels; third, size for a standard wet thermal oxidation process at 1,000◦C. Af- reduction using thermal oxidation. Fabrication of nanoscale ter oxidation, the samples were cleaved 1 mm away from the trench and pillar arrays started with clean p-type (100) sili- edge of the sealed area and perpendicular to the channels to con wafers capped with 30 nm thermal oxide. After RCA #1 expose the channel openings for SEM imaging and DNA cleaning, the wafers were first spin coated with a 180 nm stretching experiments. thick thermoplastic imprint resist (NXR 1020, Nanonex A single laser pulse was found sufficient to melt the top ◦ Corp, NJ) and then baked at 70 C for 15 min to drive out the portion of the fabricated structures and form enclosed 1D residual solvent. Silicon master molds (made by interference or 2D nanochannels. Figure 73a shows the starting struc- lithography and RIE) having parallel lines of 200 nm pitch ture, which consists of 100 nm wide trenches of 200 nm or pillars of 970 nm pitch over 4 inch wafers were used to pitch and 640 nm depth. After exposure with a single laser imprint the resists. The patterns on the imprinted resist were pulse of 790 mJ/cm2, the top liquid layer joins together to then transferred to the thermal oxide layers by RIE in a Plas- form a cap, and forming enclosed channels of 100 nm width, matherm SLR 720 RIE system (Plasma-Therm/Unaxis, St. ∼250 nm height (Fig. 73b). Similarly, 970 nm pitch pillars

Fig. 72 Schematics of self-sealing process for enclosed nanochannels. Fig. 73 (a) 100 nm wide, 200 nm pitch Si lines (640 nm high). (b)A (a) Si nanostructures (lines of pillars) made by NIL; (b) the top portion single laser pulse of 790 mJ/cm2 turns nanolines in (a) into enclosed of the Si structures is melted by laser; (c) the molten Si flows sideward Si channels. (c) 700 nm diameter Si pillars (970 nm pitch, 2 µm high) and joins the neighboring lines (pillars); (d) after resolidification, en- were turned into channels upon exposure to a single laser pulse (d)of closed channels are formed; (e) channel size shrinking using thermal 765 mJ/cm2. Insets in (b)and(d) are magnified images of the cross oxidation [103] section of the nanochannels [103] Applications of excimer laser in nanofabrication 53

Fig. 74 Effect of laser fluence on channel sealing quality. (a)With 560 mJ/cm2, only the pillar tips are melted. Without enough molten Fig. 75 With narrower trenches in Si, the top surface is flat after laser material, the tips are turned into individual round dots. (b)With sealing. (a) 70 nm wide, 840 nm deep trenches of 200 nm pitch in Si. 635 mJ/cm2, the top is partially sealed but with several holes on the (b) With a single laser pulse of 656 mJ/cm2, the trenches are sealed into surface. (c) A completed sealing is achieved with a laser fluence of enclosed channels with a flat top. (c) Tilt view of the sealed sample in 765 mJ/cm2. The starting structures are 700 nm diameter Si pillars of (b), which clearly shows a flat and smooth top surface [103] 2 µm height, 970 nm pitch and 270 nm spacing 2.25 m/s for the 270 nm spacing Si pillar structures where (270 nm spacing, 2 µm high) (Fig. 73c) are sealed, form- Si from each side of the gap needs to travel 135 nm in 60 ns. ing an interconnected network of 2D nanochannels (270 nm The surface morphology after laser sealing was found wide and 400 nm high) (Fig. 73d) upon exposure to a sin- to be related to the trench size. For example, the surface gle pulse of 765 mJ/cm2. The laser fluence was optimized of sealed 100 nm wide/200 nm pitch channels was rugged to achieve successful nanochannel sealing. Laser fluences (Fig. 73b). However, with the same pitch, laser sealing of ranging from 250 to 1100 mJ/cm2 were tried for the 200 nm 70 nm wide Si trenches (Fig. 75a) resulted in a flat and pitch Si trenches. smooth top surface (Figs. 75b, c). This is believed due to We found that laser fluence below 600 mJ/cm2 did not the fact that in the latter case the distance needed for the result in neighboring molten silicon lines joining together, molten Si to flow is shorter. In that case, a self-planarization while fluence higher than 880 mJ/cm2 resulted in partial ab- process of molten Si under surface tension (to minimize the lation of the lines (Fig. 74). An optimal processing window surface area) before it re-solidifies made the top surface flat. for the 200 nm pitch silicon lines was found to be 690– 790 mJ/cm2. The molten time vs laser fluence for a flat 6.3 Shrinking the channel sizes (100) silicon wafer is plotted in Fig. 4. For a laser pulse of 6.3.1 By laser fluence 750 mJ/cm2, the molten time is about 60 ns. Since full sealing is completed while the silicon is in a liquid state, the es- One way to control the channel size is by using different timated flow speed of the molten silicon in this case is about laser fluences, which will result in different molten times 54 Q. Xia, S.Y. Chou

Fig. 76 Different laser fluences result in different channel sizes. With 665 mJ/cm2, the channel size by self-sealing of pillars (in Fig. 73c) is 270 nm by 750 nm (a), while that for a 765 mJ/cm2 laser pulse is 270 nm by 400 nm (b)[103] and different melting depths [104]. Higher laser fluence will result in a molten layer that penetrates deeper into the Si surface. As a result, a smaller sized channel (reduced height) can be made with a higher fluence. For example, for the starting pillar structures shown in Fig. 73c, the resulting channel size was 270 nm by 750 nm (Fig. 76a) after being exposed with a laser fluence of 665 mJ/cm2, while for a 765 mJ/cm2 laser pulse, the channel was 270 nm by 400 nm (Fig. 76b). A similar effect was observed with 1D Si trenches. For example, with 840 nm deep (70 nm wide, 200 nm pitch) Si trenches, a laser pulse of 656 mJ/cm2 resulted in nanochannels of 450 nm height, while increasing the fluence to 756 and 856 mJ/cm2 reduced the channel height to 400 and 300 nm, respectively (Fig. 77).

6.3.2 By thermal oxidation

For fine size control, thermal oxidation can be used to further reduce the size of the nanochannels. Turning the sealing layer to transparent silicon oxide also makes the channels suitable for fluorescence analysis of biomolecules such Fig. 77 Starting with 70 nm wide, 840 nm deep trenches in Si (see as DNA. Self-sealed silicon nanochannels were put into the Fig. 75a for the original structure), a fluence of 656 mJ/cm2 results in 2 oxidation furnace for wet oxidation at 1,000◦C for differ- channel height of 450 nm (a); while with a laser fluence of 756 mJ/cm and 857 mJ/cm2, the channel height is 400 nm (b) and 300 nm (c), ent lengths of time. After oxidation, the samples were cut respectively [103] 1 mm apart from the laser spot edge for SEM cross-sectional imaging. Channel widths were measured after 30 to 180 min oxidation (Fig. 78). The 100 nm wide channels shrunk to (Fig. 78) for the 100 nm wide enclosed channels. In or- 50, 20, and 9 nm after 30, 60 and 90 min oxidation, respec- der to understand this “self-limiting” behavior, the samples tively. The size reducing rate from 30 min to 90 min slowed that have been oxidized for different lengths of time were down. Further increasing the oxidation time to 180 min did cleaved and dipped into diluted HF (1:100) for 2.5 min to not continue to reduce the size. delineate the SiO2/Si interface, as shown in Fig. 79.Note that the channel sizes shown in Fig. 79 were a little bit larger than their corresponding ones in Fig. 78 due to the HF etch. 6.4 Discussion on channel oxidation After 30 min thermal oxidation, there was a Si core existing in between each channel (Fig. 79a). With the oxidation time 6.4.1 Mechanism of the “self-limiting” oxidation increased to 60 min, the Si core size was reduced significantly (Fig. 79b). After 90 min, the Si in between the chan- It is very interesting that the channel size did not change nels was consumed completely, and the enclosed channels with oxidation being prolonged from 90 min to 180 min are all above the SiO2/Si interface line (Fig. 79c). Further Applications of excimer laser in nanofabrication 55 6.4.2 Geometrical factors for the Si channels oxidation

The oxidation behavior of the enclosed Si channels depends heavily on their geometries. One example is the channel width. For channels of 70 nm by 350 nm (cap thickness 270 nm), 90 min wet oxidation at 1000◦C has resulted in smaller channels <12 nm in size (Fig. 80a). However, there was still Si around the channels, which was further oxidized when the oxidation time was increased to 180 min. As a result, the channels were filled and did not exist any more (Fig. 80b). These results confirmed that the Si supply controls the final size/existence of the channels. The cap layer which encloses the channels also matters for the oxidation behavior. Without the cap layer, the open channels were oxidized much faster and in a less controllable fashion. For example, 70 nm wide, 200 nm pitch Si trenches of 500 nm ◦ Fig. 78 Plot of channel size as a function of wet oxidation time. Insets deep were oxidized at 1,000 C (wet oxidation) (Fig. 81a). are the SEM cross-section images for 100, 50, 20 and 9 nm channels, After only 5 min, the trenches were almost filled with SiO2, respectively. Note that increasing the oxidation time from 90 min to 180 min does not reduce the channel size further [103] leaving 15 nm wide channels at the bottom (Fig. 81b). The 15 nm channels disappeared when the samples were oxidized for 60 min (Fig. 81c). Similar phenomenon was observed for the 100 nm wide open trenches in Si as well. The cap layer has provided an enclosed environment in which the oxygen supply is lower than the outside surface, thus the oxidation rate is reduced. Other geometrical factors such as the thickness of the cap, the aspect ratio of the channel, etc. also play important roles in oxidation. Exploration in these parameters will not only help understanding the process, but also produce useful transparent structures/devices for different applications.

6.5 DNA stretching demonstration

To demonstrate the continuity and optical transparency of the channels, we tested the stretch of λ-phage DNA, sus- pended in a buffer solution into the 20 nm wide, 60 nm tall nanochannels fabricated by self-sealing and oxidation (Fig. 78, 1 h oxidation). Oxidized samples were cleaved at

Fig. 79 Cross-sectional images of laser-sealed channels after oxidation of different durations (a)30min;(b)60min;(c)90min; (d) 180 nm, and 2.5 min diluted HF etch of oxide. The structure before oxidation is shown in the ﬁrst inset in Fig. 78.TheSiO2/Si interfaces are clearly shown after HF dip. Note that the channel sizes were a little bit larger due to the HF etch. Increasing the oxidation time from 90 min to 180 min has pushed the SiO2/Si interface further into the Si bulk. Scale bars: 100 nm [103]

Fig. 80 (a) 90 min oxidation for the 70 nm by 350 nm enclosed Si increasing the oxidation time from 90 min to 180 min did trenches results in <12 nm wide channels. Note that there is still Si around the channel. (b) Further increasing the oxidation time to 180 nm not further reduce the channel diameter, but only grew an makes the channels disappear. These results confirmed that the Si sup- additional layer of oxide below the channels (Fig. 79d). ply controls the final size/existence of the channels [103] 56 Q. Xia, S.Y. Chou Fig. 81 (a)70nmwide, 200 nm pitch open trenches of 500 nm deep in Si. (b) After 5 min wet oxidation at 1000◦C, the trenches are almost filled with oxide except for 15 nm wide channels at the bottom. (c) After 60 min oxidation, the trenches are totally filled without any channels left [103]

wavelength are not damaged. The process is area selective. The laser spot size is adjustable, so we can expose only those areas that need to be sealed while leaving other areas untouched. This flexibility is necessary for complicated biochips that have different functional devices on one chip. Channel size can be controlled by different laser fluence or by oxidation. And lastly, this method can be extended to channels made from other materials using a laser of a different wavelength. For example, a CO2 laser can be used for sealing SiO2 trenches since it absorbs the laser pulse [105]. In summary, a simple method that can self-seal nanoflu- Fig. 82 CCD image of λ-phage DNA stretching inside of nanochan- idic channels was developed. Using a single laser pulse, 1D nels which have oxidized for 1 h. The channel size is about 20 nm. and 2D enclosed silicon nanochannels were fabricated. The The DNA molecules have been effectively stretched. This experiment size could be controlled by using different laser fluence or demonstrates that the nanochannels are continuous [103] thermal oxidation. Nanochannel arrays with a feature size down to 9 nm were prepared using this method. The oxi- least 1 mm away from the edge of the sealed area and per- dation for certain structures was found to be self-limiting, pendicular to the channels to expose the channel openings. and the mechanism was explored. DNA stretching using the λ phage DNA (48.5kbp, New England BioLabs) at a concen- oxidized channels (20 nm wide) was also demonstrated. tration of 1 µg/ml in 0.5 × TBE buffer (0.045 M tris-base, 1 mM EDTA with 0.045 M boric acid) were loaded into the channels by capillary action. The DNA was labeled with 7 Concluding remarks TOTO-1 (Molecular Probes, 1 dye molecule per 10 bp) and imaged using a Pentamax ICCD camera (Roper Scientific, In this paper, a host of novel technologies have been de- NJ) on a Nikon Eclipse TE2000 microscope using laser flu- veloped using a pulsed excimer laser together with NIL. In orescence microscopy (488 nm excitation/514 nm emission) all the cases, the laser pulse has been used as an ultrafast with a Nikon 100× (oil) objective (NA = 1.4). Figure 82 heating source which melts only a surface layer of a mater- shows a typical image for the DNA. It can be concluded that ial, leaving negligible thermal effect on the substrate and/or the DNA molecules have been successfully loaded into the other components on the same chip. channel and effectively stretched, which means the channels With laser assisted NIL (LAN) process, 100 nm wide are continuous. (200 nm pitch) grating lines have been fabricated upon ex- The self-sealing method has several advantages. It is ul- posure to a single light pulse. Since the pulse duration is trafast, simple, and cost effective. It does not involve a lot very short, the heating of the substrate and mold is greatly of complicated equipment and processing, extra material for reduced. This has been verified by numerical simulations. the seal or exceptionally flat surfaces. Due to its high-speed The imprint time of LAN has been measured to be about nature, it is regarded as a low-temperature sealing process, 200 ns. so the sealing method is viable for substrates that do not Self-perfection by liquefaction (SPEL) has been pro- withstand high temperature processes. It brings negligible posed and demonstrated as a new paradigm to remove fab- thermal effect to other components on the same chip, too. rication defects and enhance nanostructure profiles. Three Our method is highly selective to materials and only melts forms of SPEL, namely, O-SPEL (in open space), C-SPEL those which absorb the specific wavelength of a UV laser. (with a top plate in contact) and G-SPEL (with a top plate Other materials on the same chip which do not absorb that a distance above), have been discussed. Using O-SPEL, Applications of excimer laser in nanofabrication 57 rough-edged Si and metal lines have been smoothed out and 5. J. Boneberg, J. Bischof, P. Leiderer, Nanosecond time-resolved non-ideal shaped nanopads have been turned into nearly per- reflectivity determination of the melting of metals upon pulsed fect shapes. From this point of view, the most important laser annealing. Opt. Commun. 174, 145–149 (2000) 6. J. Liu, H. Kurz, N. Bloembergen, Picosecond time-resolved achievement of O-SPEL is that 3σ LER of rough edged plasma and temperature-induced changes of reflectivity and 70 nm wide Cr lines was reduced from 8.4 to 1.5 nm, transmission in silicon. Appl. Phys. Lett. 41, 643–646 (1982) which is well beyond the capability of current semiconduc- 7. P.G. Carey, B.J. Woratschek, F. Bachmann, Progress toward ex- tor industry. With C-SPEL, the cross-section profile can be cimerlaser metal planarization and via hole filling using in situ monitoring. Microelectron. Eng. 20, 89–106 (1993) maintained while the LER is removed. The first observa- 8. S.Y. Chou, P.R. Krauss, P.J. Renstrom, Imprint of sub-25 nm tion and exploration of G-SPEL has major impact both in vias and trenches in polymers. Appl. Phys. Lett. 67, 3114–3116 scientific research and in industrial applications. Applica- (1995) tions of SPEL in making smooth round NIL pillar molds, 9. S.Y. Chou, P.R. Krauss, P.J. Renstrom, Imprint lithography with 25-nanometer resolution. Science 272, 85–87 (1996) and smooth Si waveguides that have propagation loss orders 10. S.Y. Chou, P.R. Krauss, W. Zhang, L.J. Guo, L. Zhuang, Sub- of magnitude lower after SPEL, have been demonstrated. 10 nm imprint lithography and applications. J. Vac. Sci. Technol. Excimer laser was also used to make various kinds of B 15(6), 2897–2904 (1997) metal nanoparticle monolayers on different substrates. For 11. D.R. Lide, CRC Handbook of Chemistry and Physics, 79th edn. (CRC Press, Boca Raton, 1998) metal thin film on a blank substrate, the particle size is 12. W. Zhang, S.Y. Chou, Multilevel nanoimprint lithography with mainly determined by the film thickness. For better period- submicron alignment over 4 in. Si wafers. Appl. Phys. Lett. 79, icity, metal nanogratings were used as the starting structure. 845–847 (2001) A surface relief structure with shallow trenches has been 13. S.Y. Chou, C. Keimel, J. Gu, Ultrafast and direct imprint of nanostructures in silicon. Nature 417(6891), 835–837 (2002) introduced to regulate the periodicity of the nanoparticles 14. Z.N. Yu, H. Gao, S.Y. Chou, In situ real time process characteri- along the direction of the original metal lines. zation in nanoimprint lithography using time-resolved diffractive Finally, a simple method that can self-seal nanofluidic scatterometry. Appl. Phys. Lett. 85, 4166–4168 (2004) channels has been developed. Using a single laser pulse, 1D 15. Z.N. Yu, Subwavelength Gratings and Applications (Princeton and 2D self-enclosed silicon nanochannels have been fabri- University, Princeton, 2003) 16. Q.F. Xia, C. Keimel, H.X. Ge, Z.N. Yu, W. Wu, S.Y. Chou, Ul- cated. The size can be controlled by using different laser flu- trafast patterning of nanostructures in polymers using laser as- ence or fine tuned using thermal oxidation, leading to SiO2 sisted nanoimprint lithography. Appl. Phys. Lett. 83, 4417–4419 nanofluidic channels with 9 nm in. The self-limiting oxida- (2003) tion behavior of enclosed Si channels was studied, and DNA 17. A. Yariv, Optical Electronics, 4th edn. (Saunders College Pub- lishing, Philadelphia, 1991) stretching using 20 nm wide channel arrays has been demon- 18. GSolver, http://www.gsolver.com/ strated. 19. E.D. Palik, Handbook of Optical Constants of Solids (Academic It has to be pointed out that the current paper only demon- Press, New York, 1985) strates a limited number of applications of excimer laser in 20. S.A. Campbell, The Science and Engineering of Microelectronic Fabrication, 2nd edn. (Oxford University Press, Oxford, 2001) nanofabrication. As a powerful tool, excimer laser has been 21. ABAQUS, http://www.abaqus.com/ demonstrated in other fields and broader applications are yet 22. H. Du, R.C.A. Fuh, J.Z. Li, L.A. Corkan, J.S. Lindsey, Pho- to be developed. tochemCAD: A computer-aided design and research tool in pho- tochemistry. Photochem. Photobiol. 68, 141–142 (1998) Acknowledgements This work was partially supported by the US 23. Z.N. Yu, H. Gao, S.Y. Chou, New developments in real-time im- Office of Naval Research (ONR) and the Defense Advanced Research print monitoring by scattering-of-light (RIMS), in The Third In- Projects Agency (DARPA). We appreciate contributions from many ternational Conference on Nanoimprint and Nanoprint Technol- former group members in the NanoStructure Laboratory (NSL) at ogy (NNT04). Vienna, Austria, 2004 Princeton University, including but not limited to Chris Keimel, Dr. 24. R.L. Morrison, Symmetries that simplify the design of spot array Zhaoning Yu, Dr. Wei Wu, Dr. Keith Morton, Dr. Haixiong Ge, Dr. phase gratings. J. Opt. Soc. Am. A 9, 464–471 (1992) Xingyu Huang, Dr. Dr. Zengli Fu, Dr. He Gao, Dr. Shufeng Bai, Dr. 25. H. Gao, Nanoimprint Methods, Monitoring and Processes Xiaogang Liang, Dr. Patrick Murphy. We would also like to thank Ms. (Princeton University, Princeton, 2007) Dandan Zhang for some of the schematic illustrations. 26. Q.F. Xia, Z.N. Yu, H. Gao, S.Y. Chou, In situ real time monitoring of nanosecond imprint process. Appl. Phys. Lett. 89, 073107 (2006) 27. Q. Xia, TRR measurement for polymers. Unpublished, 2003 References 28. D.W. van Krevelen, Properties of Polymers, 3rd edn. (Elsevier Science, Amsterdam, 1990) 1. T.H. Maiman, Stimulated optical radiation in ruby. Nature 187, 29. H.C. Scheer, H. Schulz, A contribution to the flow behaviour of 493–494 (1960) thin polymer films during hot embossing lithography. Microelec- 2. A.L. Schawlow, C.H. Townes, Infrared and optical masers. Phys. tron. Eng. 56, 311–332 (2001) Rev. 112, 1940–1949 (1958) 30. H.D. Rowland, A.C. Sun, P.R. Schunk, W.P. King, Impact of 3. J.J. Ewing, Rare-gas halide lasers. Phys. Today 31, 32–39 (1978) polymer film thickness and cavity size on polymer flow during 4. Q.F. Xia, Nanostructure Engineering Using Pulsed Laser embossing: toward process design rules for nanoimprint lithog- and Nanoimprint Lithography (Princeton University, Princeton, raphy. J. Micromech. Microeng. 15, 2414–2425 (2005) 2007) 31. Newport high-speed detector, http://www.newport.com 58 Q. Xia, S.Y. Chou

32. Tektronix digital storage oscilloscope, http://www. 56. N. Otsu, A threshold selection method from gray level his- tektronix.com tograms. IEEE Trans. Syst. Man Cybern. 9, 62–66 (1979) 33. S.Y. Chou, Q.F. Xia, Improved nanofabrication through guided 57. W. Zhang, Development of Large-Area and Multilevel Nanoim- transient-liquefaction. Nat. Nanotechnol. 3, 295–300 (2008) print Lithography and the Applications in MOSFETs (Princeton 34. S.Y. Chou, Q.F. Xia, Self-repair and enhancement of nanos- University, Princeton, 2001) tructures by liquificaiton under guiding conditions. US Patent 58. S.Y. Chou, P. Krauss, Imprint lithography with sub-10 nm fea- 7282456, 2007 ture size and high throughput. Microelectron. Eng. 35, 237–240 35. International Technology Roadmap for Semiconductors, 2006 (1997) Update. http://www.itrs.net/Links/2006Update/FinalToPost/08_ 59. S. Park, H. Schift, H.H. Solak, J. Gobrecht, Stamps for nanoim- Lithography2006Update.pdf, 2006: p. Table 77b print lithography by extreme ultraviolet interference lithography. 36. C. Herring, Some theorems on the free energies of crystal sur- J. Vac. Sci. Technol. B 22(6), 3246–3250 (2004) faces. Phys. Rev. 82, 87–93 (1951) 60. K. Ansari, J.A. van Kan, A.A. Bettiol, F. Watt, Fabrication of 37. Z.N. Yu, L. Chen, W. Wu, H.X. Ge, S.Y. Chou, Fabrication high aspect ratio 100 nm metallic stamps for nanoimprint litho- of nanoscale gratings with reduced line edge roughness using graphy using proton beam writing. Appl. Phys. Lett. 85(3), 476– nanoimprint lithography. J. Vac. Sci. Technol. B 21(5), 2089– 478 (2004) 2092 (2003) 61. K. Watanabe, T. Morita, R. Kometani, T. Hoshino, K. Kondo, K. 38. P.L. Silvestrelli, A. Alavi, M. Parrinello, D. Frenkel, Ab initio Kanda, Y. Haruyama, T. Kaito, J. Fujita, M. Ishida, Y. Ochiai, molecular dynamics simulation of laser melting of silicon. Phys. T. Tajima, S. Matsui, Nanoimprint using three-dimensional mi- Rev. Lett. 77, 3149–3152 (1996) crolens mold made by focused-ion-beam chemical vapor deposi- 39. M. Delfino, T.A. Reifsteck, Laser activated flow of phosphosil- tion. J. Vac. Sci. Technol. B 22(1), 22–26 (2004) icate glass in integrated circuit devices. Electron Device Lett. 62. I. Junarsa, P.F. Nealey, Fabrication of masters for nanoimprint, 3(5), 116–118 (1982) step and flash, and soft lithography using hydrogen silsesquiox- 40. D.K. Armani, T.J. Kippenberg, S.M. Spillane, K.J. Vahala, Ultra- ane and x-ray lithography. J. Vac. Sci. Technol. B 22(6), 2685– high-Q toroid microcavity on a chip. Nature 421(6926), 925–928 2690 (2004) (2003) 63. Q.F. Xia, S.Y. Chou, Fabrication of sub-25 nm diameter pillar 41. D.B. Tuckerman, A.H. Weisberg, Planarization of gold and alu- nanoimprint molds with smooth sidewalls using self perfection minum thin-films using a pulsed laser. IEEE Electron. Device by liquefaction (SPEL) and reactive ion etching. Nanotechnology Lett. 7(1), 1–4 (1986) 19, 455301 (2008) 42. T. Sameshima, Self organized grain growth larger than 1 µm 64. L.C. Kimerling, L.D. Negro, S. Saini, Y. Yi, D. Ahn, S. Akiyama, through pulsed-laser-induced melting of silicon films. Jpn. J. D. Cannon, J. Liu, J.G. Sandland, D. Sparacin, J. Michel, Appl. Phys. 2 32(10B), L1485–L1488 (1993) K. Wada, M.R. Watts, Monolithic silicon microphotonics. Top. 43. V. Constantoudis, G.P. Patsis, A. Tserepi, E. Gogolides, Quan- Appl. Phys. 94, 89–119 (2004) tification of line-edge roughness of photoresists. II. Scaling and 65. K.K. Lee, D.R. Lim, L.C. Kimerling, J. Shin, F. Cerrina, Fabri- fractal analysis and the best roughness descriptors. J. Vac. Sci. cation of ultralow-loss Si/SiO waveguides by roughness reduc- Technol. B 21(3), 1019–1026 (2003) 2 tion. Opt. Lett. 26(23), 1888–1890 (2001) 44. Image Pro Plus, http://www.mediacy.com 66. N. Tsumita, J. Melngailis, A.M. Hawryluk, H.I. Smith, Fabri- 45. S.E. Orchard, On surface levelling in viscous liquids and gels. cation of x-ray masks using anisotropic etching of (110) Si and Appl. Sci. Res. A11, 451–464 (1962) shadowing techniques. J. Vac. Sci. Technol. B 19, 1211–1213 46. Y. Sato, Y. Kameda, T. Nagasawa, T. Sakamoto, S. Moriguchi, (1981) T. Yamamura, Y. Waseda, Viscosity of molten silicon and the factors affecting measurement. J. Cryst. Growth 249(3–4), 404– 67. J. Arentoft, T. Sondergaard, M. Kristensen, A. Boltasseva, 415 (2003) M. Thorhauge, L. Frandsen, Low-loss silicon-on-insulator pho- 47. P.H. Keck, W. Vanhorn, The surface tension of liquid silicon and tonic crystal waveguides. Electron. Lett. 38, 274–275 (2002) germanium. Phys. Rev. 91(3), 512–513 (1953) 68. W.H. Juan, S.W. Pang, Controlling sidewall smoothness for mi- 48. R.F. Brooks, A.T. Dinsdale, P.N. Quested, The measurement of cromachined Si mirrors and lenses. J. Vac. Sci. Technol. B 14, viscosity of alloys—a review of methods, data and models. Meas. 4080–4084 (1996) Sci. Technol. 16(2), 354–362 (2005) 69. Q.F. Xia, P.F. Murphy, H. Gao, S.Y. Chou, Ultrafast and selective 49. B.J. Keene, Review of data for the surface tension of pure metals. reduction of sidewall roughness in silicon waveguides using self- Int. Mat. Rev. 38(4), 157–192 (1993) perfection by liquefaction. Nanotechnology 20, 345302 (2009) 50. J. Brandrup, E.H. Immergut, Polymer Handbook, 3rd edn. (Wi- 70. G. Schmid, Large clusters and colloids-metals in the embryonic ley, New York, 1989) state. Chem. Rev. 92, 1709–1727 (1992) 51. S.Y. Chou, Q.F. Xia, Ultrafast selective self-repair of nanostruc- 71. W.X. Zhang, Nanoscale iron particles for environmental remedi- tures by liquefaction, in The 49th International Conference on ation: An overview. J. Nanopart. Res. 5, 323–332 (2003) Electron, Ion and Photon Beam Technology and Nanofabrication 72. W. Fritzsche, T.A. Taton, Metal nanoparticles as labels for het- (EIPBN) (AVS, Orlando, 2005) erogeneous, chip-based DNA detection. Nanotechnology 14, 52. E. Yablonovitch, T. Gmitter, Wetting angles and surface tension R63–R73 (2003) in the crystallization of thin liquid films. J. Electrochem. Soc. 73. S. Sun, C.B. Murray, D. Weller, L. Folks, A. Moser, Monodis- 131, 2625–2630 (1984) perse FePt nanoparticles and ferromagnetic FePt nanocrystal su- 53. S.Y. Chou, L. Zhuang, Lithographically induced self-assembly perlattices. Science 287, 1989–1992 (2000) of periodic polymer micropillar arrays. J. Vac. Sci. Technol. B 74. H.-M. So, J. Kim, W.S. Yun, J.W. Park, J.-J. Kim, D.-J. Won, 17, 3197–3202 (1999) Y. Kang, C. Lee, Molecule-based single electron transistor. Phys- 54. S.Y. Chou, L. Zhuang, L.J. Guo, Lithographically induced self- ica E 18, 243–244 (2003) construction of polymer microstructures for resistless patterning. 75. Z.C. Wang, G. Chumanov, WO3 sol–gel modified Ag nanopar- Appl. Phys. Lett. 75, 1004–1006 (1999) ticle arrays for electrochemical modulation of surface plasmon 55. L. Wu, S.Y. Chou, Electrohydrodynamic instability of a thin film resonance. Adv. Mater. 15, 1285–1289 (2003) of viscoelastic polymer underneath a lithographically manufac- 76. G. Leo, Y. Chushkin, S. Luby, E. Majkova, I. Kostic, M. Ul- tured mask. J. Non-Newtonian Fluid. Mech. 125, 91–99 (2005) meanu, A. Luches, M. Giersig, M. Hilgendorff, Ordering of Applications of excimer laser in nanofabrication 59

free-standing co nanoparticles. Mater. Sci. Eng. C, Biomim. 93. H. Asoh, K. Nishio, M. Nakao, A. Yokoo, T. Tamamura, H. Ma- Supramol. Syst. 23, 949–952 (2003) suda, Fabrication of ideally ordered anodic porous alumina with 77. S.W. Chen, Langmuir-blodgett fabrication of two-dimensional 63 nm hole periodicity using sulfuric acid. J. Vac. Sci. Technol. robust crosslinked nanoparticle assemblies. Langmuir 17, 2878– B 19(2), 569–572 (2001) 2884 (2001) 94. C.K. Harnett, G.W. Coates, H.G. Craighead, Heat- 78. J. Bosbach, D. Martin, F. Stietz, T. Wenzel, F. Trager, Laser- depolymerizable polycarbonates as electron beam patternable based method for fabricating monodisperse metallic nanoparti- sacrificial layers for nanofluidics. J. Vac. Sci. Technol. B 19(6), cles. Appl. Phys. Lett. 74, 2605–2607 (1999) 2842–2845 (2001) 79. T. Wenzel, J. Bosbach, A. Goldmann, F. Stietz, F. Trager, Shap- 95. D.M. Cannon, B.R. Flachsbart, M.A. Shannon, J.V. Sweedler, ing nanoparticles and their optical spectra with photons. Appl. P.W. Bohn, Fabrication of single nanofluidic channels in Phys. B 69, 513–517 (1999) 80. M. Kawasaki, M. Hori, Laser-induced conversion of noble metal- poly(methylmethacrylate) films via focused-ion beam milling island films to dense monolayers of spherical nanoparticles. for use as molecular gates. Appl. Phys. Lett. 85(7), 1241–1243 J. Phys. Chem. B 107, 6760–6765 (2003) (2004) 81. D.Q. Yang, M. Meunier, E. Sacher, Excimer laser manipula- 96. H. Cao, Z.N. Yu, J. Wang, J.O. Tegenfeldt, R.H. Austin, E. Chen, tion and pattering of gold nanoparticles on the SiO2/Si surface. W. Wu, S.Y. Chou, Fabrication of 10 nm enclosed nanofluidic J. Appl. Phys. 95, 5023–5026 (2004) channels. Appl. Phys. Lett. 81(1), 174–176 (2002) 82. S. Gupta, D. Choudhary, A. Sarma, Study of carbonaceous 97. W. Reisner, K.J. Morton, R. Riehn, Y.M. Wang, Z.N. Yu, M. clusters in irradiated polycarbonate with UV-VIS spectroscopy. Rosen, J.C. Sturm, S.Y. Chou, E. Frey, R.H. Austin, Statics and J. Polym. Sci. Part B, Polym. Phys. 38, 1589–1594 (2000) dynamics of single DNA molecules confined in nanochannels. 83. S. Hofmann, C. Ducati, J. Robertson, B. Kleinsorge, Low- Phys. Rev. Lett. 94(19) (2005) temperature growth of carbon nanotubes by plasma-enhanced 98. L.J. Guo, X. Cheng, C.F. Chou, Fabrication of size-controllable chemical vapor deposition. Appl. Phys. Lett. 83, 135–137 (2003) nanofluidic channels by nanoimprinting and its application for 84. S. Hofmann, C. Ducati, B. Kleinsorge, J. Robertson, Direct DNA stretching. Nano Lett. 4(1), 69–73 (2004) growth of aligned carbon nanotube field emitter arrays onto plastic substrates. Appl. Phys. Lett. 83, 4661–4663 (2003) 99. H.P. Chou, C. Spence, A. Scherer, S. Quake, A microfabricated 85. M.K. Min, J.H. Cho, K.W. Cho, H. Kim, Particle size and alloy- device for sizing and sorting DNA molecules. Proc. Natl. Acad. ing effects of Pt-based alloy catalysts for fuel cell applications. Sci. USA 96(1), 11–13 (1999) Electrochim. Acta 45, 4211–4217 (2000) 100. M. Stjernstrom, J. Roeraade, Method for fabrication of microflu- 86. J. Bischof, D. Scherer, S. Herminghaus, P. Leiderer, Dewetting idic systems in glass. J. Micromech. Microeng. 8(1), 33–38 modes of thin metallic films: nucleation of holes and spinodal (1998) dewetting. Phys. Rev. Lett. 77, 1536–1539 (1996) 101. S.W. Turner, A.M. Perez, A. Lopez, H.G. Craighead, Monolithic 87. J. Lee, W. Shimoda, T. Tanaka, Temperature dependence of sur- nanofluid sieving structures for DNA manipulation. J. Vac. Sci. face tension of liquid Sn–Ag, In–Ag and In–Cu alloys. Meas. Sci. Technol. B 16(6), 3835–3840 (1998) Technol. 16, 438–442 (2005) 102. W.L. Li, J.O. Tegenfeldt, L. Chen, R.H. Austin, S.Y. Chou, 88. W.J. Yao, X.J. Han, M. Chen, B. Wei, Z.Y. Guo, Surface tension P.A. Kohl, J. Krotine, J.C. Sturm, Sacrificial polymers for of undercooled liquid cobalt. J. Phys., Condens. Matter 14, 7479– nanofluidic channels in biological applications. Nanotechnology 7485 (2002) 14(6), 578–583 (2003) 89. Q.F. Xia, S.Y. Chou, The fabrication of periodic metal nanodot 103. Q.F. Xia, K.J. Morton, R.H. Austin, S.Y. Chou, Sub-10 nm self- arrays through pulsed laser melting induced fragmentation of enclosed self-limited nanofluidic channel arrays. Nano Lett. 8, metal nanogratings. Nanotechnology 20, 285310 (2009) 90. L. Rayleigh, On the instability of jets. Proc. Lond. Math. Soc. 10, 3830–3833 (2008) 4–13 (1878) 104. D.H. Auston, J.A. Golovchenko, A.L. Simons, C.M. Surko, 91. F.A. Nichols, W.W. Mullins, Surface- (interface-) and volume- T.N.C. Venkatesan, Dynamics of Q-switched laser annealing. diffusion contributions to morphological changes driven by Appl. Phys. Lett. 34(11), 777–779 (1979) capilliarity. Trans. Metall. Soc. AIME 233, 1840–1848 (1965) 105. A.D. McLachlan, F.P. Meyer, Temperature-dependence of the 92. R.H. Austin, J.P. Brody, E.C. Cox, T. Duke, W. Volkmuth, extinction coefficient of fused-silica for Co2-laser wavelengths. Stretch genes. Phys. Today 50(2), 32–38 (1997) Appl. Opt. 26(9), 1728–1731 (1987)