Optimal design and fabrication of three-dimensional calibration specimens for scanning probe microscopy

Xiaoning Liu, Tingting Luo, Yuhang Chen, Wenhao Huang, and Guido Piaszenski Citation: Rev. Sci. Instrum. 83, 053708 (2012); doi: 10.1063/1.4719661 View online: http://dx.doi.org/10.1063/1.4719661 View Table of Contents: http://rsi.aip.org/resource/1/RSINAK/v83/i5 Published by the American Institute of Physics.

Related Articles
Nanotube-based scanning rotational microscope Appl. Phys. Lett. 100, 173101 (2012) Methods and instrumentation for piezoelectric motors Rev. Sci. Instrum. 83, 033706 (2012) Note: Circuit design for direct current and alternating current electrochemical etching of scanning probe microscopy tips Rev. Sci. Instrum. 83, 036105 (2012) Tip geometry effects in dopant profiling by scanning microwave microscopy J. Appl. Phys. 111, 044314 (2012) Simple electronics for inertial and Pan-type piezoelectric positioners used in scanning probe microscopes Rev. Sci. Instrum. 83, 013708 (2012)

Additional information on Rev. Sci. Instrum.

Journal Homepage: http://rsi.aip.org Journal Information: http://rsi.aip.org/about/about_the_journal Top downloads: http://rsi.aip.org/features/most_downloaded Information for Authors: http://rsi.aip.org/authors

REVIEW OF SCIENTIFIC INSTRUMENTS 83, 053708 (2012)

Optimal design and fabrication of three-dimensional calibration specimens for scanning probe microscopy
Xiaoning Liu,1 Tingting Luo,1 Yuhang Chen,1,a) Wenhao Huang,1 and Guido Piaszenski2
1 Department of Precision Machinery and Instrumentation, University of Science and Technology of China, 230026 Hefei, China 2 Raith GmbH, Konrad-Adenauer-Allee 8, 44263 Dortmund, Germany

(Received 10 February 2012; accepted 3 May 2012; published online 23 May 2012) Micro-/nano-scale roughness specimens are highly demanded to synthetically calibrate the scanning probe microscopy (SPM) instrument. In this study, three-dimensional (3D) specimens with controllable main surface evaluation parameters were designed. In order to improve the design accuracy, the genetic algorithm was introduced into the conventional digital lter method. A primary 3D calibration specimen with the dimension of 10 m 10 m was fabricated by electron beam lithography. Atomic force microscopy characterizations demonstrated that the statistical and spectral parameters of the fabricated specimen match well with the designed values. Such a kind of 3D specimens has the potential to calibrate the SPM for applications in quantitative surface evaluations. 2012 American Institute of Physics. [http://dx.doi.org/10.1063/1.4719661]
I. INTRODUCTION

In various elds of science and technology, accurate surface nanometrology is considered to be of fundamental significance. Scanning probe microscopy (SPM) has the advantages of ultrahigh resolution, rich measurable quantities, and exible operation environment. It has been recognized as a powerful tool to characterize surface structures and numerous physic-chemical properties at the nanometer scale.1 In order to enable quantitative measurements but not qualitative observations only, reference materials and associated calibration methods are highly demanded. Nowadays, various calibration standards have been available, such as one-dimensional gratings for pitch determination,2 two-dimensional (2D) gratings for xy-plane orthogonality calibration,3 cylindrical artifacts for z-direction calibration,4 comb-shaped line structures for tip characterization,5 and non-periodic gratings for drift evaluation.6 These standards can help to calibrate the instrument behaviors in the xyz-coordinates. However, in practical quantitative surface measurement, the achieved surface parameters will closely depend on the intrinsic measurement principle, complex coupling of the specimen structure, and tip geometry,7 and also, the controller feedback parameters. Even with the precise calibrations of instrumental parameters in the three axes and the tip shape, the interpretation and comparison of the measured results remain very complicated. The regular calibration structures may not completely describe the real situations experienced in practical applications where the sample surfaces could have random structures. Toward the purpose of quantitative three-dimensional (3D) surface evaluations, the development of micro-/nano-scale 3D calibration specimens is thus necessary. For the theoretical simulation of 3D roughness surfaces with controllable statistic parameters, many methods have been proposed. In 1972, Shinozuka and Jan demonstrated
a) Electronic mail: chenyh@ustc.edu.cn.

that the power spectrum density function could be used in a random process.8 In 1990, Gu and Huang adopted autoregressive and moving average method combing with the Johnson transfer system to generate the non-Gaussian roughness surfaces with given skewness and kurtosis.9 Hu and Tonder proposed a rough surface generation procedure, which is based on fast Fourier transform (FFT) analysis and 2D digital lter technique.10 Wu used FFT method to generate non-Gaussian surfaces.11 In order to achieve better convergence and minimize the storage requirements, nonlinear conjugate gradient method was also adopted for non-Gaussian surface generation.12 Even with all these rapid progresses, the evaluation parameters of the generated surfaces are usually quite scattered when we are repeating the design procedure. In addition, the surface parameters suffer from undesirable large deviations as compared to the designed values, especially in the case of simulating non-Gaussian rough surfaces. Consequently, it is still a great challenge to design a 3D surface with its major surface evaluation parameters accurately controllable. As to the practical fabrication of the 3D calibration specimens, only a few investigations have been addressed. Nemoto et al. have fabricated a 3D roughness specimen by a super 5-axis nano machine with the dimension of 768 m 768 m for calibrating the surface roughness measured by some popular topography imaging instruments.13 However, the dimension is somewhat far beyond micro-/nano-scale. Thus, it may not be convenient for applications in SPM. Toward the purpose of accurate design and fabrication of micro-/nano-scale 3D specimens, the genetic algorithm (GA) was introduced into rough surface generations here. The surface design accuracy was veried to be improved without the sacrice of efciency. Based on such optimizations, a primary 3D roughness specimen with the dimension of 10 m 10 m was fabricated by electron beam lithography (EBL). Finally, the fabricated specimen was measured by an atomic force microscope (AFM) and the consequent surface

0034-6748/2012/83(5)/053708/5/$30.00

83, 053708-1

2012 American Institute of Physics

053708-2

Liu et al.

Rev. Sci. Instrum. 83, 053708 (2012)

evaluation parameters were calculated and compared with the designed data. All these investigations on the 3D specimens can help to establish the calibration and uncertainty analysis for various nanomeasurement instruments, including SPM.
II. CALIBRATION SPECIMEN DESIGN A. General design method

The ow chart of surface generation processes is shown in Fig. 1. In designing a non-Gaussian rough surface, the Gaussian input sequence z(i, j) should be transformed to another sequence z (i, j) , which is non-Gaussian with suitable skewness (Ssk) and kurtosis (Sku), through the Johnsons translator system.14 Then, this new sequence is ltered by a specially designed 2D digital lter h(k, l) , which is correlated with the autocorrelation function (ACF).10 In order to obtain the output sequence z(i, j) with the given skewness, kurtosis, and ACF, the skewness (Ssk ) and kurtosis (Sku ) of the modied input sequence z (i, j) can be obtained by the following equations:15, 16 Ssk =
q 2 3/2 i=1 i q 3 i=1 i

FIG. 2. Typical results of the general rough surface simulation method.

Ssk,
q 2 2 j=i+1 i j

(1)

Sku =

Sku

q 2 2 i=1 i

q1 i=1 q 4 i=1 i

(2)

deviations, which are almost larger than 200%. It means that the general method cannot efciently reach a satised accuracy, for example, the deviation within 5%. In order to obtain a surface with the prescribed parameters, we possibly have to repeat the design procedure many times and select the best surface. The optimal design by such an approach seems to be a little blind. To deal with the problem, GA is applied to optimize the design process and improve the accuracy.

Here, i = h(i, j); i = (k 1)m + l; k = 1, 2, . . . , n; l = 1, 2, . . . , m. Typical results in designing non-Gaussian surfaces by the above general method are depicted in Fig. 2. Here, the prescribed skewness Ssk is 1 and kurtosis Sku is 5. The design procedure was repeated 50 times and the nal numerical evaluations were carried out. From the gure, the skewness and kurtosis of the generated surfaces are quite dispersive. Compared to the expected values, many of them even have great

B. Genetic algorithm optimization

GA is a method simulating natural evolution processes to search the optimal solution. It operates iteratively on a series of coded approximations to a solution, called as a population, by applying the principle of the ttest survival. The ow chart of the GA-based optimal design of a random 3D rough surface is schematically illustrated in Fig. 3. When the algorithm starts, the aforementioned parameters should be rst dened and several 3D surfaces are generated to compose the initial population. Then, the tness of every individual surface in the population is evaluated. In our design procedure, the tness is evaluated by the calculation of the error function, Ssk Ssk Ssk
2

tness =

Sku Sku Sku

(3)

FIG. 1. Flow chart of non-Guassian surface generation. Here, Ssk and Sku are the designed skewness and kurtosis, respectively. Ssk and Sku are the modied skewness and kurtosis. ACF denotes the autocorrelation function.

Here, skewness Ssk and kurtosis Sku are the designed parameters, Ssk and Sku are the skewness and kurtosis of the simulated surface, respectively. It should be mentioned that Eq. (3) is just the objective function to optimize the amplitude parameters (Ssk and Sku) of the 3D surface. However, other surface parameters can also be included. The individual with bigger tness means the simulated surface has smaller errors. Consequently, the individuals in each population are chosen with probabilities according to their tness. In each iterative generation, these survived individuals are applied with genetic operations, such as crossover and mutation. When a prescribed number of iteration is reached or the tness meets

053708-3

Liu et al.

Rev. Sci. Instrum. 83, 053708 (2012) TABLE I. Parameters used in the optimal design of 3D rough surfaces. Parameter Population dimension Crossover probability Mutation probability Iteration number Surface dimension Prescribed skewness Prescribed kurtosis Symbol N Pc Pm T M Ssk Sku Value 10 0.8 0.001 10 256 256 1 5

III. EXPERIMENT A. Fabrication

FIG. 3. Schematic illustration of the genetic algorithm based optimal design method.

the end criteria, the evolution terminates, and the individual with the best tness is exported. This exported individual represents the nal optimized rough surface. The typical optimization results are illustrated in Fig. 4. The main parameters adopted in optimization are shown in Table I. From Fig. 4(a), the tracked relative error decreases with the increase of iteration. At the same time, the skewness and kurtosis of the simulated 3D surfaces approach to the given values gradually. The variation tendency of the parameters during the genetic optimization is shown in Fig. 4(b). Only after 10 generations, the nal relative error can reach 1%, which is enough for practical applications. Results demonstrate that the optimization procedure is very efcient to design rough surfaces with different parameters, such as skewness and kurtosis. With the same design accuracy, the above GA method is demonstrated to be more efcient than the conventional approach.

After the optimal design, a primary 3D calibration specimen was fabricated with the RAITH150TWO EBL system.17 The designed pattern was rst exposed on 1.6 m thick PMMA 950 K resist, which was spin coated on a 1 cm 1 cm silicon substrate. The substrate was covered with native oxide. After spin coating, the sample was heated for 2 min at 180 C on a hot plate. Figure 5(a) gives an overview of the rough surface. The colored bitmap has been converted into a grayscale bitmap, as shown in Fig. 5(b). During the patterning, the gray scale level of each pixel corresponds to a specic dose of that pixel. However, the relationship between applied electron dose and resulting prole depth in the resist is not linear. It depends strongly on the contrast curve of the resist process as well as on proximity effect. Both inuences were taken into account by recalculating the applied dose for each pixel. The calculation was performed by using Raiths software package 3Lith, which is dedicated to the fabrication of 3D structures with EBL. After EBL exposure, the sample was developed in pure MIBK for 5 min at a temperature of 22.5 C before it was rinsed in pure IPA for 30 s. Subsequently, the surface was coated with 16 nm Pt in a sputter coating process in order to allow evaluation of the topography under scanning electron microscope (SEM) control.

B. Measurement and evaluation

One fabricated 3D specimen under 45 tilt through SEM is illustrated in Fig. 6(a). Compared with the designed surface

FIG. 4. Typical results of optimal design. (a) Tracked relative error during the iteration. (b) Variation tendency of the parameters during the iteration.

FIG. 5. One optimized surface. (a) Topography. (b) Corresponding grayscale bitmap.

053708-4

Liu et al.

Rev. Sci. Instrum. 83, 053708 (2012) TABLE II. Comparisons of surface parameters. Surface parameters Root mean square roughness Ten point height Skewness Kurtosis Fastest decay autocorrelation length Texture aspect ratio Density of summits Root mean square slope Arithmetic mean summit curvature Developed surface area ratio Surface bearing index Core uid retention index Valley uid retention index Symbol Sq (m) Sz (m) Ssk Sku Sal (m) Str Sds (m2 ) Sdq Ssc (m1 ) Sdr Sbi Sci Svi Designed 0.15 0.78 0.10 3.03 0.89 0.96 0.30 0.40 1.31 0.08 0.58 1.62 0.11 Fabricated 0.13 0.77 0.03 3.61 0.89 0.96 1.06 0.31 2.34 0.05 0.63 1.50 0.12

FIG. 6. Results of the fabricated specimen. (a) Scanning electron microscope image under 45 tilt. (b) Atomic force microscope image. (c) Comparisons of the sectional proles between the designed and measured surfaces along the arrows indicated in (b).

structure in Fig. 5, the peaks and valleys of the surface are coincident intuitively. In order to perform quantitative analyses of the fabricated specimen, an AFM (Bruker Innova) was used to measure the surface structure. Typical measured AFM image is demonstrated in Fig. 6(b). The image was taken in contact mode with a triangular silicon nitride cantilever having a nominal spring constant of 0.24 N/m (Bruker probes SNL10, cantilever C). The scan rate was chosen as 0.6 Hz. For clarity, the raw image was aligned and only the area containing the 3D surface information was extracted here. No other image processing methods have been applied. From the comparisons of the sectional proles, as shown in Fig. 6(c), the fabricated rough surface is in reasonable agreement with the designed one. The AFM data were then used for quantitative comparison with the designed data. The surface roughness parameters of the designed surface and the fabricated surface were calculated, respectively. The calculation results are listed in Table II. We mainly adopted 14 popular surface parameters, whose detailed denitions and calculations can be found in literature.18, 19 With all these parameters, the geometric properties of the 3D surface can be well characterized. In this special case, both the designed and the measured surfaces are nearly homogenous, with texture aspect ratio close to 1. The parameter texture direction Std becomes meaningless and it is omitted consequently. The comparisons verify that the measured results are generally in close agreements with the designed ones. No significant discrepancies are presented. However, relatively large deviations are observed for the parameters Sku, Sds, Sdq, and Ssc. Several aspects can explain the discrepancies. The limitation of the fabrication precision and the tip distortion in the AFM measurement may lead to the increase of Sku. Parameter Sds is the number of summits per unit area of the rough surface under inspection. The burrs and noise either

in the fabrication or in the measurement are unavoidable. These factors may contribute to the overestimate of summits and Sds increases consequently. The decrease of root mean square surface slope Sdq may also be induced by the tip dilations. For example, a peak in the real surface will be broadened and the average surface slope decreases. Parameter Ssc is the mean summit curvature, which is closely relevant with the calculations of the parameter Sds. Thus, it is quite sensitive to the image noise. Nevertheless, above quantitative evaluations demonstrate that the design and fabrication of micro/nano-scale 3D surface can be well controlled. With further optimizations of the fabrication processes, we believe that the geometrical characteristics of these specimens can be easily improved. Such a kind of reference specimen can enable accurate verications of the SPM instrument for measuring complex surface geometry. For example, if the surface structure is precisely controlled, we can use the specimen to investigate the inuence of feedback parameters on the SPM-based metrology or to examine the factor of drift. The evaluations of imaging parameters by experimental approaches seem to be semi-quantitative or even qualitative due to the lack of 3D reference structures, although large interests have been attracted on these issues by theoretical analyses and numerical simulations.20 Until now, a large amount of experimental verications have been mainly limited on periodic surfaces, such as 2D calibration artifacts or atomic lattices. These simple regular surfaces may not completely describe the real condition encountered in SPM applications, where the sample structures are random and complex. By further improving the wear-resistance of these digitally designed and fabricated specimens, they can be possibly served as a kind of valuable supplements to the conventional calibration gratings. These investigations are still undertaken for the time being. In addition to be the structural reference materials, the 3D calibration specimens can be adopted for understanding many other functional properties of certain surfaces, such as contact, lubrication, friction, and wear of surfaces at the micro-/nano-scale. The 3D reference specimens are expected to be able to provide powerful supports for all these relevant investigations.

053708-5

Liu et al.
2 I.

Rev. Sci. Instrum. 83, 053708 (2012) Misumi, S. Gonda, T. Kurosawa, and K. Takamasu, Meas. Sci. Technol. 14, 463 (2003). 3 G. Dai, F. Pohlenz, T. Dziomba, M. Xu, A. Diener, L. Koenders, and H. U. Danzebrink, Meas. Sci. Technol. 18, 415 (2007). 4 F. Marinello and E. Savio, Meas. Sci. Technol. 18, 462 (2007). 5 H. Itoh, T. Fujimoto, and S. Ichimura, Rev. Sci. Instrum. 77, 103704 (2006). 6 D. Niu, J. Li, Y. Chen, and W. Huang, J. Vac. Sci. Technol. B 28, 1070 (2010). 7 Y. Chen and W. Huang, Meas. Sci. Technol. 15, 2005 (2004). 8 M. Shinozuka and C. M. Jan, J. Sound Vib. 25, 111 (1972). 9 X. Gu and Y. Huang, Wear 137, 275 (1990). 10 Y. Z. Hu and K. Tonder, Int. J. Mach. Tools Manuf. 32, 83 (1992). 11 J. J. Wu, Tribol. Int. 37, 339 (2004). 12 K. K. Manesh, B. Ramamoorthy, and M. Singaperumal, Wear 268, 1371 (2010). 13 K. Nemoto, K. Yanagi, M. Aketagawa, I. Yoshida, M. Uchidate, T. Miyaguchi, and H. Maruyama, Meas. Sci. Technol. 20, 084023 (2009). 14 N. L. Johnson, Biometrika 36, 149 (1949). 15 W. Watson and T. A. Spedding, Wear 83, 215 (1982). 16 S. K. Chilamakuri and B. Bhushan, Proc. Inst. Mech. Eng., Part J: J. Eng. Tribol. 212, 19 (1998). 17 See http://www.raith.com/?xml=solutions%7CLithography%26 nanoengineering for RAITH150TWO EBL system. 18 W. P. Dong, P. J. Sullivan, and K. J. Stout, Wear 178, 29 (1994). 19 W. P. Dong, P. J. Sullivan, and K. J. Stout, Wear 178, 45 (1994). 20 J. Legleiter, Nanotechnology 20, 245703 (2009).

IV. SUMMARY

Genetic algorithm has been applied to the design of 3D roughness specimens based on FFT analysis and 2D digital lter technique. Results indicate that the accuracy is improved without the sacrice of design efciency. After the optimal design, the 3D roughness calibration specimens were fabricated by EBL. Surface roughness parameters of the fabricated specimens were characterized by AFM and compared with the values of designed surface data. In general, all the 14 surface parameters were in close agreement with the prescribed ones without signicant discrepancies. Such a kind of controllable micro-/nano-scale rough specimens can have a great potential in the precise calibration of various metrological instruments, including SPM instruments.
ACKNOWLEDGMENTS

This work was supported by the National Natural Science Foundation of China (No. 91023021), and 973 Project (No. 2011CB932801).
1 D.

Fujita, H. Itoh, S. Ichimura, and T. Kurosawa, Nanotechnology 18, 084002 (2007).