See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224322853

Field Programmable Analog Array (FPAA) based control of an Atomic Force

Microscope

Conference Paper in Proceedings of the American Control Conference · July 2008

DOI: 10.1109/ACC.2008.4586899 · Source: IEEE Xplore

CITATIONS READS
28 482

2 authors:

Georg Schitter Nghi Phan

TU Wien NHL Stenden University of Applied Sciences
332 PUBLICATIONS 6,909 CITATIONS 7 PUBLICATIONS 84 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Georg Schitter on 21 May 2014.

The user has requested enhancement of the downloaded file.

2008 American Control Conference ThB08.2
Westin Seattle Hotel, Seattle, Washington, USA
June 11-13, 2008

Field Programmable Analog Array (FPAA) based

Control of an Atomic Force Microscope
Georg Schitter* Nghi Phan
Delft Center for Systems and Control Veeco Metrology Inc.
Delft University of Technology 112 Robin Hill Rd.
Mekelweg 2, 8c-3-22 Santa Barbara, CA-93117
2628 CD Delft, The Netherlands USA

Abstract— For topography measurements and faster imaging

with the AFM a high control-bandwidth is required. This paper
presents an analog implementation of a model-based controller
for a high-speed Atomic Force Microscope (AFM) using a
new type of control hardware. The vertical positioning axis
of the AFM scanner is modeled, and the imaging bandwidth is
improved by means of model-based control. The new feedback
controller, which is designed in the H∞ -framework, is imple-
mented on a Field Programmable Analog Array (FPAA), which
enables operation of the model-based controlled AFM system
at a feedback bandwidth on the order of 100 kHz.
Measured results demonstrate that the closed-loop system
recovers from a step-like disturbance within 7 microseconds.
Recorded AFM images verify a significant performance im-
provement of the model-based controlled system over the analog
proportional-integral (PI) controlled AFM.

Fig. 1. Scheme of a high-speed AFM.

I. I NTRODUCTION
The continuing development and demand for nano-
fabrication as well as research in life science require better
processes, however, imaging at video-rates (25 frames
tools for nanotechnology. The atomic force microscope
per second) would be desirable. Recent advances in
(AFM)[1] is one of the most important tools to image and
AFM-instrumentation demonstrated feasibility of real-time
manipulate on the nanometer level. For some applications
imaging at the nano-meter scale with the AFM [3][4][5].
such as imaging biological systems in real-time, however,
Improvements in imaging speed have been reported by
current commercial AFMs are about two orders of
re-designing the scanning unit [4][6][7] to achieve a
magnitudes too slow, which puts a high demand on ongoing
faster response, as well as by applying modern control
instrument development.
methods. Improved control of the scanning motion has been
The principle of the AFM (see Figure 1) is that the vertical
demonstrated by damping of the actuator’s resonances [8]
(Z-direction) force interaction between the sample and a
as well as by applying model-based feedforward [9][10]
very sharp tip is held constant while either of them is
or feedback control [11][12] or iterative learning control
scanned relative to the other. This is done by feedback
[13]. Tracking of the sample topography has been improved
operation on the static deflection of the cantilever (in contact
by better control of the tip-position using model-based
mode or constant force mode) or on the amplitude of the
feedback [14], by an internal model control approach [15],
oscillating cantilever (in dynamic or tapping mode). For an
by a gain-scheduling approach [16], and by feedforward
overview see e.g. [2] and references therein.
compensation [17] of the estimated sample-profile.
In all measurement modes AFM operation can be split
Although the image acquisition and scanning speed of some
into two main tasks, namely i) the scanning motion and ii)
prototype AFMs is impressive, the feedback bandwidth for
controlling the tip-sample interaction force, given by the
tracking the sample topography with the AFM-tip still has
(static or dynamic) deflection of the cantilever.
to be improved for imaging soft biological tissue (see [3]).
One of the main issues in AFM instrumentation still
This paper presents an analog implementation of a model-
is the imaging speed [3]. Commercial AFM systems
based feedback controller on a field programmable analog
typically take between several seconds and a few minutes
array (FPAA), which enables a control bandwidth beyond
to acquire one image. For direct observation of molecular
100 kHz with tight control demands for high-speed AFM
* Corresponding author: Georg Schitter (phone: +31-15-278-6152; imaging at reduced variations of the imaging force. The
email: g.schitter@tudelft.nl) implementation of a model-based feedback controller,

978-1-4244-2079-7/08/$25.00 ©2008 AACC. 2690

designed in the H∞ -framework, for controlling the vertical
(Z) positioning axis of a high-speed AFM demonstrates
significant improvements over an analog proportional-
integral (PI) feedback controller. The high bandwidth of the
control system is demonstrated by recorded AFM images
as well as by demonstration of the fast recovery (within
7 µs) from a step-like disturbance, which emulates the
case of the sample topography when scanning at high-speeds.

II. M ODELING OF THE AFM DYNAMICS

The high-speed AFM scanner under investigation is
a prototype of a flexure-based scanner (cf. [7]), which
is designed for high resonance frequencies in order to
achieve a fast response. The piezoelectric stack actuators
are driven with custom made amplifiers. Coupling between
the individual positioning axis has been minimized during
the device design.
For modeling and control, the input to the power amplifier Fig. 2. Frequency response of the AFM in vertical direction, measured
for the vertical positioning axis is regarded as the system at various amplitudes (coloured lines) and simulated response of the fitted
model (solid black line).
input. The system output is given by the deflection of the
AFM cantilever sensing the position of the Z-piezo.

A. Vertical Dynamics
The scanner dynamics in the vertical direction have been
characterized by applying a sweep signal to the power
amplifier of the Z-piezo. The Z-piezo’s position, regarded
as the system output, is sensed by an AFM-cantilever
(resonance frequency > 500 kHz) and the AFM’s optical
deflection detection system. Generation of the input signal
and recording of the system’s response is done with a
network analyzer (HP35639A, Agilent, Palo Alto, CA,
USA).
Figure 2 shows the measured frequency response between
10 and 300 kHz for excitation amplitudes of 1 mV,
25 mV, and 50 mV at the input of the power amplifier,
demonstrating the linearity of the system’s dynamics.
Potential nonlinearities of the piezo actuator, such as Fig. 3. Step response of the AFM in vertical direction. Measured response
hysteresis and creep, are compensated by the feedback loop (dotted red); simulated response of the 10th order model (solid black).
and have to be considered at the data acquisition side when
recording and displaying the AFM signals.
The frequency response in Z-direction reveals two
decay of the oscillations in the Z-piezo’s position.
resonance–anti-resonance pairings due to the integration
of the piezo in the scanner structure (cf. [7]) at 70 and
100 kHz, respectively, and three resonances between 150
III. C ONTROLLER DESIGN
and 200 kHz, which can be attributed to the Z-piezo itself.
A fitted 10th -order mathematical model of the scanner is The oscillatory behavior of the AFM scanner imposes
shown by the solid black line in Fig. 2. limitations to the achievable imaging speed. In Z-direction
This model is also chosen as the nominal model of the the higher order oscillatory modes (see Figs. 2 and 3) cannot
scanner for the controller design (see Section III-A). be compensated by a simple PI controller, and therefore limit
Figure 3 demonstrates the good agreement between the the achievable closed-loop bandwidth for tracking the sample
simulated step-response of the modeled system (solid black topography with PI-feedback. For such oscillatory systems a
line) and the measured step-response (dotted red line). The model-based feedback controller, which is derived in the this
weakly damped multiple resonances of this system are section, enables to achieve a higher closed-loop bandwidth
obvious by the aggressive initial response and the slow (cf. [14]).

2691
A. Controller for Topography Tracking for the control error (We ) and for the output tracking
Figure 4 shows a block-diagram of the AFM topography (Wy ), as well as the achieved sensitivity function S and
measurement system. The blocks connected by solid lines complementary sensitivity T when closing the loop with
form the control-loop for tracking the sample topography. the H∞ -feedback controller, derived using Matlab (The
During this feedback operation two signals are recorded, MathWorks, Natick, MA, USA). The resulting controller is
forming the AFM images (i) the topography signal, which
is represented by the control action u, and (ii) the deflection
signal, which corresponse to the residual control error e.
The control goal is to keep the cantilever deflection close
to the deflection setpoint denoting the nominal imaging
force, which means to keep variations of the imaging force
small and to shift as much information as possible from the
deflection image to the topography image (cf. Fig. 11).
The frequency response (Fig. 2) and step response (Fig. 3)

Fig. 5. Weighting function 1/We (dotted green) and 1/Wy (dashed red)
for the H∞ controller design, and achieved sensitivity S (dash-dotted blue)
and complimentary sensitivity T (solid black).

of 12th order and can be balanced and reduced to 10th

order without significant loss of control performance. The
simulated response of the closed-loop controlled system
to a step-like input signal settles in less than 10 µs to
within two percent of the steady-state position, as shown
Fig. 4. Schematic of the AFM topography measurement system. The blocks
in Figure 6. In case case of the 12th order controller the
connected by solid lines form the control loop of the AFM in Z-direction. system settles without any oscillations. For the 10th order
The dashed lines show the signals recorded to display the AFM images. The controller the specifications are still met, and only very small
dash-dotted lines indicate the weighting functions of the extended model for
the design of the H∞ -controller.
oscillations in the Z-position can be observed (Fig. 6(b)).
Since the complexity of the 12th -order controller is close to

of the scanner in Z-direction shows several slightly damped

resonances of this positioner. To achieve a high closed-loop
bandwidth in combination with good robustness a model-
based feedback controller for the AFM-system in Z-direction
is designed. This controller is based on the fitted 10th -order
mathematical model presented in Fig. 2. This mathematical
model (comprised of the blocks between u and y in Fig. 4)
is extended by the weighting functions We , Wu , and Wy , as
indicated by the dash-dotted lines in Figure 4. The feedback
controller C(s) is designed in the H∞ -framework by a mixed
sensitivity design [18], as is also described in a similar
application in [17].
The weighting function Wu is set constant. The weighting
function Wy defines the upper bandwidth for the reliability
of the model G(s) and is fine tuned to shape the loop.
In the design step the weighting function We is increased Fig. 6. Simulated step-response of the closed-loop system, controlled with
in its bandwidth as much as possible in order to achieve the 12th -order (solid blue lines) and a 10th -order H∞ -controller (dashed
fast tracking of the closed-loop controlled system with the red lines), respectively. The step in the input signal (not shown) occurs at
time 0.1 10−4 s. (a) controlled position of the Z-actuator; (b) zoom onto the
requirement of no oscillations in the position of the Z-piezo. 2-percent-band around the steady-state position of (a); (c) control action u.
Figure 5 shows the singular values of the weighting functions

2692
the maximum capability of the used control hardware [19], directly be implemented in the FPAA software using pre-
and the 10th -order controller also meets the specified control defined elements, where the user adjusted parameters are
performance, only the 10th -order controller is implemented the respective frequency and damping of the corresponding
and tested on the AFM setup. poles and zeros. The higher-order controller C( s) eventually
is realized by connecting the first and second order filters in
B. Controller Implementation on FPAA series
For an effective implementation of the new feedback KI Y Y
controller the control-hardware has to fulfill the following C(s) = G(s)bilin(i) G(s)biquad(j) , (3)
s i j
three requirements:
• For a bandwidth of more than 100 kHz and the tight where the factor KI corresponds to an integral gain and
control objectives (fast settling time without oscilla- i and j are the respective indices of the bilinear (1) and
tions), the sampling rate of a digital controller should biquadratic filters (2). The FPAA is eventually programmed
be at least 1 MHz. via a serial connection with the development-board.
• The resolution of the D/A-converter of a digital control Figure 7 shows the user interface for programming the
system has to be 16-bit or higher to achieve the required FPAA, showing the structure of the 10th -order feedback
accuracy over the entire positioning range of the Z- controller implemented at a switching frequency of 4 MHz.
actuator. The first element of the implemented controller is a
• The controller should be easy to implement and modify summing stage for calculating the error signal. The second
for varying applications of the AFM. element is an integrator, followed by four biquadratic and
The high demand on the fast sampling rate excludes most one bilinear filter.
digital signal processors (DSP) enabling floating point oper- In order to scale and interface the differential input of
ation from the choice for the implementation of the feedback
controller. Field programmable gate arrays (FPGA) may give
the high sampling frequency, but due to the fixed point
implementation programming is tedious and easy and fast
modification is not possible. A new type of hardware (see e.g.
[20]) that fulfills all the above listed requirements are field
programmable analog arrays (FPAA). These analog arrays
are based on switched capacitor filters with a switching
frequency up to 16 MHz [19]. The analog nature of these
circuits omits the quantization issue completely, and the
achievable resolution is determined only by the system’s
signal-to-noise ratio. Furthermore, programming of these Fig. 7. User interface for the implementation of the 10th order filter on the
circuits is fast and easy and can be updated on the fly. FPAA. Reprogramming the FPAA on the fly via a serial connection allows
The used FPAAs development boards [19] allow imple- for fast and easy changes of the implemented feedback controller.
mentation of first- and second-order elements (bilinear and
biquadratic filters) with left half-plane poles and zeros. the FPAA with the single-ended AFM signals, we used
s + ωz operational amplifiers AD-8132 (Analog Devices, Norwood,
G(s)bilin = , (1)
s + ωp MA, USA). The differential FPAA-output signal has been
s2 + 2ζz ωz s + ωz2 scaled and converted to a single-ended signal using an
G(s)biquad = , (2) operational amplifier AD-8130 (Analog Devices), and
s2 + 2ζp ωp s + ωp2
consecutively has been connected to the power amplifier
with frequencies fz = ωz /(2π) and fp = ωp /(2π) and driving the Z-piezo.
quality-factors Qz = 1/(2ζz ) and Qp = 1/(2ζp ) of poles In order to characterize the achieved closed-loop
and zeros, respectively. It should be noted that these FPAAs performance we measured the sensitivity function S
cannot implement non-minimum phase zeros, which may and complementary sensitivity T, as is shown in Figure 8.
be accounted for in the design and implementation of the The sweep-signals for measuring S and T are generated
controller. with the network analyzer and are applied to the closed-loop
For the implementation, the above designed 10th -order con- system by using an additional input of the FPAA, which
troller has to be converted into a biquadratic structure. This is connected with the summing stage for the error signal
is done by transforming the state space representation of generation (see Fig. 7). The corresponding response signal
the controller into a zero-pole-gain representation, using for S and T is recorded by the network analyzer via an
the Matlab command zpk, and by combining (pairs of) additional output of the FPAA. Tuning of the integral gain of
poles and zeros with neighboring frequencies to first- and the FPAA has been done during AFM imaging (see Figs. 10
second order subsystems. These first and second order sub- and 11) in order to increase the imaging bandwidth as much
systems, given by their transfer functions (1) and (2), can as possible, which explains the slightly higher bandwidth

2693
 (red line). Please note that the error signal returns to its
initial value (zero), whereas the cantilever deflection does
not since the disturbance step has been added to it after
the point where the deflection is recorded. The apparent
high noise-level on the FPAA output (blue signal) are
high-frequency oscillations due to the switching-frequency
of the capacitors, which get suppressed by the low-pass
characteristics of the power amplifier driving the Z-piezo
(green signal).

IV. I MAGING R ESULTS

For imaging applications with the high-speed AFM lateral
oscillations of the scanner are suppressed by model-based
filtering of the scanning signal in order to avoid imaging
artifacts due to turn-around ripples (see [10]).
Fig. 8. Measured complementary sensitivity function T and sensitivity To demonstrate the benefit of the new feedback controller
function S of the feedback loop in Z-direction. we imaged a silicon calibration grating as a test sample. Fig-
ure 10 shows deflection images recorded in contact mode at
a speed of 95 lines per second, recorded with the PI-feedback
of the measured curves of S and T as compared to the and the H∞ -feedback controlled AFM, respectively. To
simulation. One can observe some additional small peaks demonstrate the repeatability of the system response the slow
in the measured spectra of S and T (Fig. 8) as compared scan axis has been disabled, resulting in re-imaging the same
to the simulated responses (cf. Fig. 5), which may result scan-line over and over again and stacking the consecutive
from noise at the FPAA-output and from un-modeled higher responses of the same line together to one image. In case
modes of the AFM positioning system, but in general the
measurement reflects the predicted behavior very well.
To further characterize the closed-loop response a step-like

Fig. 9. The H∞ -controlled closed-loop system recovers from a step-like

disturbance within 7 µs. Control error signal (blue, 0.1 V/div) recorded at
Fig. 10. Comparison of the residual control error of the PI-controlled
an auxiliary FPAA output; Control action (green, 2 V/div) recorded at the
AFM (left image) and the H∞ -controlled one (right image). The lower
output of the Z-piezo’s power amplifier; Cantilever deflection signal (red,
panels show cross sections of single scan lines from the images above them,
0.5 V/div) at the AFM output; time-scale: 5 µs/div.
demonstrating the reduction of the cantilever deflection by a factor of 3. The
imaged sample is a silicon calibration grating with 100 nm deep squared
holes at a pitch of of 1 µm scanned at a speed of 95 lines per second.
disturbance signal has been applied to the AFM control
loop. This signal is added to the cantilever deflection, which
has the same effect to the closed-loop for tracking the of the H∞ -controller the cantilever deflection (denoting the
sample topography as a topographical step in the sample residual control error) is reduced by a factor of about 3,
surface. Figure 9, recorded with a digital oscilloscope, as is obvious when comparing the cross sections given in
shows that the closed-loop system fully recovers from the lower two panels of Fig. 10. This significant reduction
the step-like disturbance within 7 µs, displaying the of the cantilever deflection results in lower variations of
control error e measured at an additional FPAA output the imaging force and also enables further reduction of the
(blue line), the control action u measured at the output average imaging force, since one can image closer to the
of the power amplifier of the Z-piezo (green line), and minimum force point without loss of tip-sample contact.
the cantilever deflection measured at the AFM system Figure 11 shows the topography (left) and deflection (right)

2694
 VI. ACKNOWLEDGMENTS
The authors would like to thank Prof. Tsu-Chin Tsao from
UCLA for fruitful discussions.
This work is supported in part by TU-Delft 3mE faculty grant
PAL-614, by the Netherlands Organization for Scientific
Research (NWO) under Innovational Research Incentives
Scheme (VENI DOV.7835), and by the National Institutes
of Health under Award RO1 GM 065354-05.

R EFERENCES
[1] G. K. Binnig, C. F. Quate, and C. Gerber, “Atomic force microscope,”
Phys. Rev. Lett., vol. 56(9), pp. 930–933, 1986.
[2] D. Abramovitch, S. Andersson, L. Pao, and G. Schitter, “A tutorial on
the mechanisms, dynamics, and control of atomic force microscopes,”
in Proceedings of the 2007 American Control Conference, AACC.
Fig. 11. Topography (left) and deflection (right) images of a silicon New York, NY: IEEE, July 11–13 2007, pp. 3488–3502.
calibration grating with a pitch of of 1 µm, recorded with the H∞ - [3] P. Hansma, G. Schitter, G. Fantner, and C. Prater, “High speed atomic
controlled AFM at 47 lines per second. The lower panels show cross sections force microscopy,” Science, vol. 314, pp. 601–602, 2006.
marked by the white line in the images. [4] T. Ando, T. Kodera, E. Takai, D. Maruyama, K. Saito, and A. Toda,
“A high-speed atomic force microscope for studying biological macro-
molecules,” Proc. Nat. Acad. Sci., vol. 98, pp. 12 468–12 472, 2001.
[5] A. Humphris, M. Miles, and J. Hobbs, “A mechanical microscope:
High-speed atomic force microscopy,” Appl. Phys. Lett., vol. 86(3), p.
image of a calibration grating, recorded with the H∞ - 034106, 2005.
controlled system at a scan-speed of 47 lines per second [6] J. Kindt, G. Fantner, J. Cutroni, and P. Hansma, “Rigid design
of fast scanning probe microscopes using finite element analysis,”
and a resolution of 512 by 512 pixels. The fast scanning Ultramicroscopy, vol. 100(3-4), pp. 259–265, 2004.
direction is aligned with the horizontal axis of the images, the [7] G. Schitter, K. Åström, B. DeMartini, P. Thurner, K. Turner, and
recording direction is from left to right. The high feedback P. Hansma, “Design and modeling of a high-speed afm-scanner,” IEEE
Trans. Contr. Syst. Technol., vol. 15(5), pp. 906–915, 2007.
bandwidth is obvious by the small (<150 mV) deviations [8] A. Fleming and S. Moheimani, “Sensorless vibration suppression
from the nominal cantilever deflection of 0 V. In this imaging and scan compensation for piezoelectric tube nanopositioners,” IEEE
example the feedback-system has to handle over 2400 step Trans. Contr. Syst. Technol., vol. 14(1), pp. 33–44, 2006.
[9] D. Croft, G. Shed, and S. Devasia, “Creep, hysteresis, and vibration
responses per second, and still manages a fast return to the compensation for piezoactuators: Atomic force microscopy applica-
nominal cantilever deflection, given by the dominant brown tion,” ASME J. Dyn. Syst. Meas. Contr., vol. 128(35), pp. 35–43, 2001.
color in the deflection image. [10] G. Schitter and A. Stemmer, “Identification and open-loop tracking
control of a piezoelectric tube scanner for high-speed scanning-probe
microscopy,” IEEE Trans. Contr. Syst. Technol., vol. 12(3), pp. 449–
454, 2004.
V. C ONCLUSION [11] S. Salapaka, A. Sebastian, J. P. Cleveland, and M. V. Salapaka, “High
bandwidth nano-positioner: A robust control approach,” Review of
Scientific Instruments, vol. 73, no. 9, pp. 3232–3241, 2002.
This contribution presents the successful implementation [12] A. Sebastian and S. Salapaka, “Design methodologies for robust nano-
of a new feedback controller for a high-speed AFM. Field positioning,” IEEE Trans. Contr. Syst. Technol., vol. 13(6), pp. 868–
programmable analog arrays enables an easy, fast, and 876, 2005.
[13] K. Leang and S. Devasia, “Design of hysteresis-compensating iterative
re-programmable implementation of higher-order feedback learning control for piezo-positioners: Application to atomic force
controllers for control applications at high bandwidths. microscopes,” Mechatronics, vol. 16, pp. 141–158, 2006.
A model of the AFM dynamics in the vertical positioning [14] G. Schitter, P. Menold, H. Knapp, F. Allgöwer, and A. Stemmer, “High
performance feedback for fast scanning atomic force microscopes,”
direction is derived, and a model-based feedback controller Rev. Sci. Instrum., vol. 72, no. 8, pp. 3320–3327, 2001.
for tracking the sample topography at a control bandwidth [15] N. Kodera, H. Yamashita, and T. Ando, “Active damping of the scanner
on the order of 100 kHz is designed in the H∞ -framework. for high-speed atomic force microscopy,” Rev. Sci. Instrum., vol. 76,
p. 053708, 2005.
A measured step response demonstrates that the closed-loop [16] N. Kodera, M. Sakashita, and T. Ando, “Dynamic proportional-
system fully recovers from a step like disturbance within integral-differential controller for high-speed atomic force mi-
7µs. The successful implementation of the 10th -order H∞ croscopy,” Rev. Sci. Instrum., vol. 77, pp. 083 704–1–083 704–7, 2006.
[17] G. Schitter, A. Stemmer, and F. Allgöwer, “Robust two-degree-of-
feedback controller on the FPAA demonstrates a significant freedom control of an atomic force microscope,” Asian Journal of
improvement over the PI-controlled AFM system. Control, vol. 6(2), pp. 156–163, 2004.
The fast imaging capabilities and high feedback bandwidth [18] S. Skogestad and I. Postlethwaite, Multivariable Feedback Control.
Wiley, 1996.
of the prototype AFM are demonstrated by imaging a test [19] Field Programmable Analog Array: AN231K04-DVLP3 -
specimen at high speed. The significant reduction in the AnadigmApex Development Board, Anadigm Inc., Campbell,
residual cantilever deflection by means of the new feedback CA, USA, 2006.
[20] L. Znamirowski, O. Palusinski, and S. Vrudhula, “Programmable
system is an important step towards gentle imaging of analog/digital arrays in control and simulation,” Analog Integrated
fragile biological specimens in real-time. Circuits and Signal Processing, vol. 39, pp. 55–73, 2004.

2695

View publication stats