CC BY
19
УДК 004.896
ТЕХНИКА
АВТОМАТИЗИРОВАННАЯ НАНООБРАБОТКА С ИСПОЛЬЗОВАНИЕМ РОБОТОВ НА НАНОУРОВНЕ: ОБЩИЙ ОБЗОР И СОВРЕМЕННОЕ СОСТОЯНИЕ
© С. Фатиков,
заведующий кафедрой робототехники и системы управления (AMIR), факультет микроробототехники и автоматизации,
отделение вычислительной техники, Ольденбургский университет, Германия
http://www.amir.uni-oldenburg.de/en/ эл.почта: fatikow@uni-oldenburg.de
Автоматизация наноообработки с использованием роботов представляет собой одну из ключевых задач на пути к промышленному внедрению нанотехнологий. Управляемые, воспроизводимые процессы манипулирования на наноуровне дают возможность начать крупносерийный выпуск инновационной продукции и открывают новые области ее применения. В статье дается краткий обзор состояния развития автоматизированного наноманипулирования и описывается сущность результатов, полученных за последнее время в нашей лаборатории в рамках нескольких научно-исследовательских проектов и представленных в различных международных журналах. Предложены некоторые решения ключевых исследовательских проблем в области автоматизации нанообработки.
Ключевые слова: наноманипулирование, микроробототехника, автоматизация нанообработки, роботоуправление
© S. Fatikov
ROBOT-BASED HANDLING AND AUTOMATION AT NANOSCALE: OVERVIEW AND CURRENT PROGRESS
Division of Microrobotics and Control Engineering, Department of Computing Science
University of Oldenburg, Germany http://www.amir.uni-oldenburg.de/en/ e-mail: fatikow@uni-oldenburg.de
Automated robot-based nanohandling is one of the key challenges on the way to commercialization of nanotechnology. Controlled, reproducible manipulation processes at the nanoscale will enable high-throughput manufacturing of revolu-tionary products and open up new application fields. This paper provides a short overview of the state of the art in automated nanohandling as well as the quintessence of the results achieved in several research projects recently implemented in our lab and presented in different international journals. Some solutions for key research problems in automated nanohandling are offered.
Key words: nanomanlpulatlon, microrobotics, automation of nanohandling, robot control
1. INTRODUCTION
The handling of objects at the nanoscale is important trend in robotics. It is referred to as na-nohandling, having in mind the positioning accuracy required, that is 10—20 nm and better. Nano-handling is normally understood as manipulation of nanoscale objects, which include their grasping, moving, releasing, positioning, pushing, pulling, cutting, bending, etc. How-ever, nanohandling may also include nanostructuring like indentation or scratching, measure-ments by a nanotool, characterization, etc. To give a general definition,
robot-based nano-handling is a technology embracing all kinds of operations at the nanoscale that require robotics. The fundamental concepts of automated robot-based nanohandling were introduced in [1—5].
As in the field of the regular industrial robotics, where humans leave hard, unacceptable work to robots, microrobots (drastically miniaturized robots) help humans to work at the na-noscale. Microrobots are able to operate in constricted workspaces, e.g. under a light microscope or in the vacuum chamber of a scanning electron
2013, том 18, № 4 llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll
microscope (SEM). In particular, companies looking into nano-electro-mechanical systems (NEMS) benefit from robot-based nanohandling, since batch technologies would not suffice for the fabrication of sophisticated devices.
The following chapter overviews the state of the art in automated handling. Chapter 3 addresses the recent developments in mobile nanohan-dling robotics. Advanced Ramona concept is introduced; it benefits from the existing experience in the field. Chapter 4 describes a novel approach for fast SEM vision feedback for real-time position control of the robots. The final chapter offers a brief introduction of most recently implemented nanohandling applications.
2. AUTOMATION OF NANOHANDLING
The following approaches are being pursued by the nanohandling research community:
• Top-down approach utilizing serial nanohandling by microrobots, especially inside SEM. The main research goals include miniaturization of robots and tools and adjustment of intrinsic robotic technologies (sensing, positioning, control, automation) to the challenges at the nanoscale.
• Bottom-up approach, or self-assembly, aiming at parallel assembly of micro/nano objects by autonomous organization into patterns or structures without human intervention.
• Nanohandling by a scanning probe microscope (SPM). Here, the (functionalized) tip of an atomic force microscope (AFM) or of a scanning tunneling microscope (STM) acts as a nanohandling tool affecting the position or the shape of a nanoscale object.
Several other approaches such as the use of optical tweezers or electrophoresis might also be adapted for automated handling. Their application is, however, restricted basically to the manipulation of biological samples requiring very low grasping forces in the pN range.
2.1 Self-assembly
Self-assembly draws its inspiration from nature; it can be defined as the spontaneous formation of higher order structures from basic units.
This approach is increasingly being studied by NEMS research communities as (parallel) self-assembly may be beneficial to a large number of parts involved. Current trends in self-assembly at the nanoscale were well reviewed e.g. in [6].
Generally, self-assembly includes recognition and making connections to other parts of the system. Each part must be able to recognize (self-assembly programming mechanism) and connect (self-assembly binding/driving force) to adjacent part or template. To guide the self-assembly, e.g. electrostatic, chemical or capillary forces can be utilized. Self-assembly of 1D materials like carbon nanotubes (CNT) and other nanowires has recently attracted significant attention. The reason is the interest in new applications in nanoelec-tronics and nanooptics as well as in new ways for interconnecting at the nanoscale. The handling of 1D materials is a challenging task due to their shape anisotropy that makes their proper integration into a device difficult. Robot-based handling of CNT is the alternative method; it will be introduced later in the paper.
Self-assembly has the potential to radically change the automated fabrication of nanoscale devices as it enables the parallel handling of arbitrarily shaped objects in a very selective way. However, despite promising results achieved up to now, this technology still remains on the level of basic research. Critical challenges are limited flexibility and the lack of interconnecting methods that enable automatic, site-specific localization and integration of parts into the chip. The implementation of fault-tolerant control methods like the ones in biological systems will also play a prominent role in transferring self-assembly from research labs to industry.
2.2 SPM as a nanohandling robot
SPM delivers high-resolution images of a wide class of samples. Additionally, this device can be used to interact with nanoscale objects in order to change their position or their shape, cf. [7; 8]. Especially AFM, applied as a nanohandling robot, has been actively investigated. The ultimate goal is automated handling in ambient conditions,
ВЕСТНИК АКАДЕМИИ НАУК РБ/
/ 2013, том 1B, № .....................................................................
aiming at rapid prototyping and even high throughput fabrication of nanodevices. Besides the positioning of nanoscale objects, the AFM tip can be used to modify surfaces with (sub-) nanometer resolution by scratching, indentation, deposition, cutting, dissecting, etc. This is primarily of interest for maskless lithography, characterization of biological samples and investigation of new materials.
The major problem on the way to automation is the lack of real-time visual feedback during manipulation by AFM. The same AFM tip cannot be simultaneously used for both imaging and handling. Due to the uncertainty of object behavior at the nanoscale, the handling process has to be frequently stopped in order to verify the current state by an AFM scan in dynamic mode. This makes the operation inefficient and slow. With a valid model describing all interactions between tip, object, and substrate it may be possible to simulate behavior of the nanoobjects during manipulation and to calculate the expected position of the object in realtime [9; 10]. This approach, however, requires full understanding of nanoscale phenomena. Another problem is spatial uncertainty in AFM due to thermal drift, creep, and hysteresis of piezoactuators.
From the automation point of view, a combination of AFM-based robotic operation with
other imaging techniques providing independent visual feedback during nanohandling is the most promising approach. For positioning accuracies down to 0.5^m (coarse positioning), the process can be monitored by an optical microscope. AFM—SEM hybrid system is the most powerful option for high-resolution visual feedback down to a few nanometers (fine positioning). Here, an AFM (normally custom-made) is integrated into the vacuum chamber of SEM, so that both technologies are used in a complementary fashion, building an effective nanohandling device [11]. SEM is used as a sensor for fast visual feedback during nanohandling by AFM tip.
2.3 Microrobot-based nanohandling
Microrobotics for handling nanoscale objects has been established as a self-contained research field for nearly 20 years. The transfer of classical industrial robotic 'know-how' from regular scale to nanoscale is a big R&D challenge. Novel sensor and actuator solutions as well as control and automation methods suitable for nanohandling are required. To ensure repeat positioning accuracy down to tens of nm, nano-handling robots use direct drives, implemented by piezoelectric, electrostatic, or thermal actuators.
Fig. 1. Generic concept of automated microrobot-based nanohandling station (AMNS)
ВЕСТНИК АКАДЕМИИ НАУК РБ/
/ 2013, том 18, № 4 llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll
The flexibility of such a robot is enhanced by dividing the actuator system into a coarse positioning module (micropositioning) and a fine positioning module or manipulator (nanopositioning) carrying an application-specific tool.
Fast and reliable process feedback, i.e. the transmission of sensor data from the workspace to the control system, is a crucial aspect of automation. Frequently, vision feedback is the only prac-tical way to control a nanohandling process. Optical microscopy is well-suitable for coarse po-sitioning of the robot, when accuracies in цш-range are required. For fine positioning, the vacu-um chamber of a SEM is the only solution for most applications. It provides ample and clean workspace, image resolution of down to 1 nm, and magnification of up to 500,000. With the development of nanotechnology, the interest in nanohandling inside SEM is rapidly growing [12—15].
Figure 1 presents a generic concept of the automated microrobot-based nanohandling station (AMNS), first introduced in [16] and further gradually developed by our team [3—5].
The microrobots are mostly driven by pi-ezoactuators allowing resolution down to sub-nm level. Both stationary robots and mobile robots are widely used in AMNS. The movement range is a few centimeters for stationary robots and limited by the size of the SEM vacuum chamber for mobile robots. The mobile robots have a nanomanip-ulator integrated into the mobile platform, which makes them capable both of moving over longer distances and of nanohandling. The mobility means more flexibility, as the robots can operate everywhere inside the chamber. Stationary robots are, on the other hand, easier to control, which makes them more suitable for high-throughput automation. The flexibility of AMNS can be enhanced by accommodating several robots that can cooperate and specialize in different handling steps. The robot manipulator can carry various tools like nanogrippers, nanoprobes or SPM tips.
Additional technologies and devices, i.e. focus ion beam system (FIB), AFM, or gas injection system (GIS) can be integrated into the SEM, which opens up a variety of new applications [17]. Such integration of different technologies allows
any combination of fabrication, manipulation and measurement/characterization, resulting in a powerful and versatile AMNS. The entire nano-handling process can be implemented and completed in the same vacuum chamber, which is a crucial pre-condition for automation.
SEM, video cameras, and — if available — position sensors integrated into robot axes build AMNS sensor system. The SEM delivers near-field sensor information for fine positioning of the robot, and video cameras provide the necessary far-field feedback for coarse positioning. The sensor data are sent to the low-level control system for real-time signal processing. The task is to detect and to track the position of robots and objects to be handled. The detected positions serve as input data for closed-loop control of robot position. The high-level control system is responsible i.e. for coordination of the station components, path planning, and error handling. This includes also graphical user interface (GUI) and haptic interface. The latter is used for teleoperation of AMNS robots, which is often the first step on the way to automation as it helps the user to learn more about the nanohandling task. An advanced control system architecture tailored for nanohandling automation in SEM was introduced in [18].
The following two chapters deal with the crucial issues to be taken care of on the way to high throughput nanohandling automation: design of fast and precise mobile nanohandling robots, and obtaining fast and reliable vision feedback from SEM to control the robot position.
3. NANOHANDLING ROBOTS
Stationary robots in AMNS normally contain actuator modules combined into a system with desired number of Cartesian degrees of freedom (DoF). The market for such robots is rapidly growing (cf. www.nanotechnik.com, www.smaract. de, www.physikinstrumente.de). Mobile robots are more difficult to design and control, but they offer several benefits. They have a large working range and the ability to approach the object from any direction. They do not need sophisticated integration; due to the compact design, mobile robots can be easily added to practically any nanohandling
013, TOM 18, № 4IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII llllllllllllllllED
setup (SEM, AFM, FIB, etc.) by simply providing a flat operating surface. Using this flexibility, versatile nanohandling tasks can be implemented in an effective way. There are, however, several research issues to be addressed. Mobile robots do not have internal position sensors, so that external sensors are required for closed-loop position control. Mobile robots are more sensitive to changes in environmental conditions such as working surface, humidity and dust. This can result in unreliable, non-linear and time-variant behavior. Last, effective control can become difficult as the locomotion and manipulation DoFs often have parasitic crosstalk.
Many labs have developed mobile robots capable of performing nanomanipulation with the assistance of high resolution imaging by SEM or SPM, e.g. [19—21]. Such robots normally utilize stick-slip actuation by piezoceramics; this driving principle is well-known. Several problems make its application challenging, however, if ultra-high resolution of down to a few nm is required, cf. [22] for in-depth analysis. In short, properties such as high stiffness, low driven mass and high acceleration are crucial for every robot actuated by a stickslip drive.
The recent development of the nanohan-dling robot Ramona (RApid Mobile platfOrm with Nanometer Accuracy) had to face these fundamental issues (Fig. 2). The most important design targets were avoiding surface abrasion, increasing actuator stiffness, and reducing accelerated masses during the slip phase.
Below only a short description of the
operating principle is given. A detailed analysis can be found in [23—24]. Fig. 2a and b show the working principle of the actuator. Each steel sphere (Ramona wheel) is driven by three ruby hemispheres (Fig. 2c). Each ruby hemisphere is epoxy-glued to three electrodes of the piezo plate. For small displacements, the comparably elastic epoxy film acts as a joint between ruby and ceramic. When electric potential is applied to an electrode, the corresponding segment expands or contracts (Fig. 2b). The three electrodes can be connected to any potential, so that the ruby hemisphere is actuated via three channels and can be rotated in two DoFs. Displacements in z are avoided by keeping the sum of all channels at 0V. The rotation is transmitted to the steel sphere using static friction.
The three steel spheres (three piezoactuators with nine ruby hemispheres) are necessary to support the platform and generate its motion. Overall this leads to 27 segments. Due to the redundancy, these segments can be combined into six independent channels. Segments with equal control signals are interconnected using a printed circuit board (PCB), which also integrates the actuators mechanically (Fig. 2d). The PCB is supplemented by a metal block, which provides the necessary mass as well as electrical shielding. The PCB has dimensions of 18x13 mm2; with steel spheres of 3 mm diameter the platform's minimum height is 5 mm.
The design of the ceramic actuator was developed and optimized using the results of numerous finite element simulations [24]. A very stable rotation around the ruby hemisphere's center
Fig. 2. Ramona design:
a) Working principle of the ruby hemisphere's and steel sphere's actuation in static condition and b) in working condition; c) Image of the laser-fabricated piezoactuators with glued ruby hemispheres, size: 4.8x5x0.5mm3; d) The mobile platform Ramona, bottom view
ВЕСТНИК АКАДЕМИИ НАУК РБ/
/2013. том 18, № 4
point with a maximum displacement of +/-75 nm at +/-150 V was achieved. [23] provi-des details of the experimental validation of Ramona performance, which is superior to other existing robots with respect to the key properties i.e. step size, step frequency, energy con-sumption, and fabrication costs. The next research step is to obtain a full model of the robots behavior on the basis of a novel stick-slip friction model. This is a challenging task due to the large number of relevant parameters. First implementation results were presented in [25—26].
Many nanohandling applications require integration of additional Z-axis into Ramona platform. Fig. 3 shows the design of the vertical axis that was developed for Ramona and successfully applied for various automated nanohandling tasks.
Fig. 3. Design and integration of the Z-axis Top view onto the mounted runner (l.); integration of the runner (m.); 4 DoF Ramona robot with a nanogripper (Femtotools) mounted on the tool holder (r.)
The runner carries a tool holder equipped with a mechanical and electrical interface to enable the use and exchange of different tools. The runner is guided and driven by six piezo-electric actuators using the stick-slip principle in the same way as in Fig. 2a,b. The preload spring integrated into the runner is implemented as a flexure hinge that generates the nece-ssary preload in order to exert a sufficient normal force between actuators and runner. Details to the integration of Z-axis
and all kinematic equations as well as the results of simulation and experimental validation of the slip-stick behavior are presented in [26].
4. FAST SEM VISION FEEDBACK AND ROBOT CONTROL IN AMNS
Due to uncertainties at the nanoscale the relation between robot position and values of the robots' internal position sensors is non-linear and time-variant. Therefore, the internal sensors are insufficient for reliable nanopositioning. Instead, SEM images can be directly used for closed-loop positioning by the so-called visual servoing or tracking [27]. However, the position-ing speed is limited as SEM image acquisition is slow. Even with well-selected region-of-interest (Rol) and coping with high image noise levels, 50 Hz can hardly be exceeded [28]. Image processing algorithms also take considerable time as they are computationally intensive. An elegant solution to this problem was offered in [29]. If the object of interest (pattern) is known, its position can be determined only relying on two dedicated line scans (Fig. 4). Two line scans are conducted exactly over the last known position of the pattern center (Fig. 4a, left). If the pattern moves, the line scans are no longer cross the center, and a deviation of the detector signal occurs during the line scan (Fig. 4b, left). This information is used to determine the new position of the pattern. The next pair of line scans is then adjusted to the new position of the pattern center. Thus, continuous tracking of a pattern over long distances is possible, with 1kHz update rate.
A precise calculation of the pattern position on each line scan is the key to robust and precise tracking by this approach. Novel method based on center of gravity calculation was develop, cf. [29] for details. To follow a robot position, a pattern has to be fabricated directly on the robot tool. Fig. 4 (right) shows the decoration of an electrostatic microgripper (courtesy of Univ. of Toronto) with four FIB patterns. The patterns 1, 3 and 4 are ro-tationally invariant, whereas pattern 2 can be used to detect rotation. Signal-to-noise ratio and well-defined shape of the pattern are crucial factors for this approach.
013, tom 18, № ............................................................E3
Fig. 4. Determining the position of a pattern using two orthogonal line scans
Servoing principle (l.); 1 |jm patterns made by FIB milling on the tip of a microgripper (r.)
For automated robot positioning closed-loop trajectory control of the robot was implemented based on the presented SEM line scan tracking. For simplicity, only the control of linear trajectories is briefly introduced below; the method is easily extendable for arbitrary trajectories. A linear trajectory is defined by its start s, destination d and duration T. The trajectory starts at the time t0 = 0 and finishes at the time tf = T. Thus, at an arbitrary time ti in this interval, the robot should be located at the position p(ti):
Each tracking update leads to an iteration of the control loop. The control time ti is determined and the control deviation eti is calculated comparing the current tracking result and p(ti). In order to eliminate this control deviation, the controller requires an integral part. Further-more, a derivative part can improve the response time. Thus, a PID controller is used to calculate the control value vector dt: и
dt = Kp • eti + Ki • ^ eti + Kd • (et. - et|_J ¿=o
Kp, Ki and Kd are the coefficients of the three PID controller components. The values dx and dy are then used to calculate the values px and py that determine the new robot direction:
As already mentioned, the 27 piezoelectric segments of Ramona are combined into six control channels. Using px and py, the amplitudes of saw-tooth shaped control voltage are calculated for all six channels. To support the update rates about 1Kz delivered by the line scan tracking, the signal generation is implemented on a specially designed hardware controller. Crucial parameters, signal amplitude and step frequency, can be changed within 80 ^s. Thus, the controller introduces very little latency into the control loop. Using the generated control voltages, the robot moves in an open-loop manner with the maximal speed in the direction given by px and py until a new position update is delivered by SEM line scan tracking; for further details cf. [29—30]. Current work aims at extending the SEM line scan approach to 3D tracking.
5. APPLICATION EXAMPLES OF AUTOMATED NANOHANDLING
The introduced technologies for automated nanohandling have been implemented by our lab in numerous research projects. This chapter offers a brief insight into two fields of application that we currently focus on and presents some of the recent implementation results.
5.1 Automated assembly of CNT-enhanced AFM probes
Standard AFM probes are not usable for imaging high aspect-of-ratio nanostructures (Fig. 5a). Ultra sharp silicon tips with radii of 5 nm and cone angles of 10° are available. However, their mechanical stability is limited due to brittleness of the sharp silicon tip. CNT-enhanced AFM probes with superior characteristics in stability, resolution, and lifetime are likely to overcome the limitations of micro-machined tips. Various approaches exist
2013, том 18, № 4 IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIH
Fig. 5. Automated nanorobotic fabrication of CNT-enhanced AFM probes:
a) motivation; b) AMNS structure; c) AMNS implementation in Lyra HR-SEM-FIB (Tescan); d) shadow-based depth detection; e) contact detection; f) EBiD bonding; g) resulting AFM tip
to fix a CNT on AFM tip (cf. the overview in [31]). We pursue the automated nanorobotic assembly and FIB modification. The AMNS for assembly of CNT-enhanced AFM probes includes a HR-SEM and FIB dual beam system (Fig. 5b,c). In addition, the station contains a GIS enabling the inlet of precursor gases for nanobonding by electron beam induced deposition (EBiD) and etching. The na-norobot combines a micropositioning (SmarAct) and nanopositioning (PI) units to facilitate operation over several orders of magnitudes. The accuracy of the nanopositioning amounts to 1.55 nm for all three XYZ axes in closed-loop mode. The robot carries an electro-thermal nanogripper with 2 ^m stroke (courtesy of DTU Nanotech, Denmark). The assembly sequence includes positioning microgripper and CNT, gripping and detachment of CNTs, and placing and releasing CNTs onto AFM tip. The SEM provides constant visual tracking of CNT and gripper. For the most difficult step for automation, the alignment of CNT between the gripper jaws, a novel depth detection method was developed. When CNT is on the same height with gripper, the gripper jaws block the secondary electrons (SE) emitted from the CNT and prevent them from reaching the SE-detector. The resulting "shadow" on the CNT is automatically detected by SEM imaging (Fig. 5d).
To pick up individual vertically aligned multi-wall CNTs grown onto a silicon substrate by CVD (courtesy of Univ. of Cambridge, UK), gentle shear gripping is used. FEM simulations were carried out for a detailed analysis of mechanical stress distribution along the CNT. For placing and fixing the CNT onto the AFM tip, EBiD-support-ed strategy was implemented. The contact between CNT and tip causes slight CNT bending; it is detected by SEM (Fig. 5e). Then, the electron beam is focused onto the contact area triggering EBiD of tungsten from W(CO)6 injected by the GIS. The result is a strong bond so that the gripper can be removed, and the fabrication of a CNT-enhanced AFM probe is completed (Fig. 5f-g).
Additional FIB processing of CNT-en-hanced AFM probes may open up special metrology applications such as the AFM analysis of sidewall nanostructures. Nanowire irradiation with high-energy ions triggers a self-organization mechanism which may result in a bending of the nanowires (Fig. 6a). The direction of the bending depends on the energy level of the ion beam, material composition, and other factors [18-19]. We implemented this technology for modification and customizing of CNT-enhanced AFM tips, using gallium ion liquid metal source. For CNT shortening the so-called filled circle with a radius
ВЕСТНИК АКАДЕМИИ НАУК РБ /
/ 2013, том 18, № 4
of 0.25 ^m is used as scan-ning area centered on CNT axis. The experiments indicated a nearly perfect linear removal.
Fig. 6. FIB modification of CNT-enhanced AFM
tips
a) concept; b) FIB-based shortening of As-grown MWCNT with a beam energy of 30 keV and probe current of 256 pA: 18.9 ^m (l.), 14.7 ^m after 80 s FIB irradiation (m.); c) FIB-based bending of the shortened MWCNT (top left) with a beam energy of 30 keV and probe current of 107 pA, after FIB irradiation for 10 s, 35 s, and 125 s (r., top to bottom)
CNT bending was implemented with the side orientation of FIB and a filled rectangle scan (18.5 ^m x 3m ^m). The CNT bends in the FIB direction (Fig. 6c). The bending radius can be well controlled by the time of FIB irradiation, resulting in repeat alignment accuracy of about 0.5°. For the experiment evaluation and more details to this work refer to [32].
5.2 Nanorobotic transfer and characterization of graphene flakes
Graphene has an exceptional potential for various applications. The long-term goal for a direct fabrication and integration of graphene is a CMOS compatible growth process on a silicon wafer. Nanorobotic handling is a promising approach to facilitate prototypic integration of graphene flakes onto a chip. This can speed up
the development of nanoelectronic devices. This section presents a recently developed nanorobotic technology for flexible transfer and mechanical characterization of CVD-grown graphene flakes.
The graphene samples used in our experiments are provided on a standard aluminum TEM grid with a mesh size of 50 ^m and with a supporting film of lacey carbon. The suspended graphene has one to six monolayers, which amounts to a thickness of 0.3-1.8 nm. The nano-handling station was introduced in section 5.1; it enables the preparation, structuring and characterization of graphene in the same workspace. The tool for graphene transfer is a tung-sten tip processed by an electrochemical etching in sodium hydroxide. Standard piezoresistive AFM probes are used to perform mechanical characterization on the suspended graphene.
The transfer process for the subsequent mechanical characterization is divided into six steps (Fig. 7). The graphene sample has to be undamaged, free of contaminations and of appro-priate size. Thus, the TEM grid is systematically inspected using the HR-SEM. The selected graphene fragment is separated (except a tiny bond bridge) from the lacey carbon film and pick-up holes are created by FIB milling. The etched tungsten tip is taken to graphene flake by the coarse positioning robot and inserted into the pick-up hole by the fine positioning robot. This handling sequence is automatically controlled using SEM imaging. After the bond bridge has been removed, the graphene flake is lifted by the fine positioning robot, transferred to a substrate and placed onto the prepared test bed for mechanical characterization.
Due to the small contact surface between tip and flake, the adhesive forces between sub-strate and flake exceed those between tip and flake as soon as the graphene flake touches the surface, and the tungsten tip can be easily removed. EBiD of platinum or tungsten is used to mechanically and electrically fix the graphene flake into the characterization test bed.
The mechanical characterization of the resulting graphene membranes is made by indentation using a self-sensing piezoresistive AFM probe (SEIKO PRC 400) with a tip radius smaller
ВЕСТНИК АКАДЕМИИ НАУК РБХ
/21
2013, том 18, № 4
Fig. 7. Nanorobotic transfer of a graphene flake:
A) SEM inspection/selection of a suitable graphene spot on the lacey carbon film; B) separation of the flake and milling pick-up holes by FIB; C) pick-up with a tungsten tip; D) transfer to the character-iza--tion test bed on a substrate; E) release onto the test bed; F) attachment to the substrate by EBiD
than 20 nm (Fig. 8).
The AFM probe is calibrated on a hard reference substrate in order to determine the bending of the cantilever vs. the applied force and to adjust the indentation measurements. The nanorobotic system is tilted up to 80° using the motorized SEM stage to provide free access of SEB beam to observe the tip-sample interaction. For a deflection of up to 300 nm the indentation force varies in the range of hundreds of nN and single pN (Fig. 8, r.). Fig. 6 shows the graphical representation of such a characteristic force-displacement curve. For details to the experiment verification and estimation of Young's modulus and pretension cf. [33]. After this proof-of-concept more systematic experi-
ments will be conducted in the near future on different graphene samples with well-known number of monolayers.
OUTLOOK
Nanohandling microrobots are able to operate in constricted workspaces, e.g. under a light microscope or in the vacuum chamber of a SEM. Automation of robot-based nanohandling will enable high-throughput manufacturing of sophisticated nanodevices and open up new application fields. It requires i.e. special robot design solutions, nanoscale-specific tools and handling techniques, novel technologies for sensory feedback from the nanoscale and advanced control methods. One of
2,0x10'7
Indentation depth 8 [m]
Fig. 8. Nanorobotic mechanical characterization of a suspended graphene membraneTest bed for application of a self-sensing piezoresistive AFM probe (l.); the probe tip is placed by the robot above the center of the graphene membrane (m.); afterwards a non-destructive deflection of the graphene membrane is performed and force-displacement curves are recorded (r.)
ВЕСТНИК АКАДЕМИИ НАУК РБ /
/ 2013, том 18, № 4
