2010 American Control Conference FrC08.

6
Marriott Waterfront, Baltimore, MD, USA
June 30-July 02, 2010

Robust H∞ Control of a Scanning Tunneling Microscope

under Parametric Uncertainties

Irfan Ahmad, Alina Voda and Gildas Besançon

Abstract— This paper is devoted to the control system are done with simple classical PI (PID) control technique
design for high performance scanning tunneling microscope with a simplified version of the system model. There is
(STM). A common approach by scanning probe community is no discussion about noise in that work as well. A step
to use conventional proportional integral (PI) control design
to control the vertical movement of STM tip (z-direction). variation in sample surface is studied in [5] and a VSC
In this article, a modern H∞ control design is analyzed in (variable structure control) design methodology in the
order to obtain the dual purpose of ultrahigh positioning presence of PI control is proposed in order to avoid
accuracy with high bandwidth. Uncertainty model, based STM tip collision with sample surface. A control design
on experimental analysis of tunneling characteristics and methodology based on pole placement with sensitivity
parametric description of the STM, and norm-bounded
real perturbations are considered, and an H∞ controller function shaping using second order digital notch filter
is designed by following the desired control objectives. A is proposed in [6] for the feedback control system of STM
performance and robustness analysis is finally performed to (the general description of this control design methodo-
test robust stability and performance of STM. logy is given in [7], [8]) but the proper tuning of the control
parameters in order to follow the design constraints can
Keywords : Scanning tunneling microscope, Nano-technology,
Precision positioning, H∞ control, Uncertainty model, be a difficult task for STM operators. According to author’s
Parametric variations, Robustness. knowledge, no one else before analyzed the performance
of STM in vertical z-direction with modern H∞ control
I. INTRODUCTION design framework, although it has been discussed in
Scanning Tunneling Microscope (STM) is an important case of Atomic Force Microscope (AFM) control [9]. The
tool in obtaining images of the material samples with control of AFM for vertical z-direction is discussed in
atomic resolution which was first invented by Gerd Binnig [10], [11] with classical PI (PID) control technique as well.
and Heinrich Rohrer in early 1980s [1]. STM works on So, we first tried to investigate the performance of STM,
tunneling phenomenon which occurs when an extremely keeping in mind the dual purpose of ultrahigh positioning
sharp metallic electrically charged tip approaches at the accuracy with high bandwidth, with modern H∞ approach
vicinity of the conductive sample surface (the distance and we concluded better performance than classical PI
between tip apex and sample surface in the range of control design methodology [12]. But still there was a need
0.1 − 1 × 10−9 m) [2]. Because of the atomic resolution, of further analysis with uncertainty model designed by
the STM has vast applications in different domains and considering variations in all parameters of STM nominal
the ultrahigh positioning accuracy with high bandwidth model. Then, robust stability and performance must be
are the great challenges. analyzed in addition to nominal stability and performance
As the distance between STM tip apex and surface of the system which can be critical depending on the
is less than 1 × 10−9 m to get the tunneling effect, the range of parametric variations. The presence of noise,
electronic control in vertical z-direction is very critical non-linearities and physical limitations in the control loop
in order to get a good image quality of the surface, are always the limiting factors to be considered in order
in the presence of external disturbances. Presently, in to get desired performances.
most commercial equipments of STM, only simple types
of controllers (proportional-integral (PI) or proportional- The goal of the present work is thus to analyze the
integral with derivative (PID) control) are implemented to scanning tunneling microscope (STM) performance with
control the movement of STM tip in vertical z-direction nominal and uncertainty model in the H∞ framework. A
where parameters of such controllers are fixed manually complete system overview with corresponding simulation
by the operator. In such operation modes, the imaging model and the tunneling characteristics of STM experi-
process can not be optimum and the image does not mental setup is given in Section II. The nominal and
correspond necessarily to the reality [3]. The feedback uncertainty model are discussed in Section III. Section
loop of STM in vertical z-direction with some stability IV then presents the control problem formulation with
conditions has been presented in [4]. All such analysis desired performances. H∞ control design applied to STM
is discussed in section V. Simulation results to validate
The authors are with GIPSA-lab, Control System De- the controller and to analyze the robust stability and
partment, ENSE3, BP 46, 38402 Saint Martin d’Hères,
France. (email: irfan.ahmad, alina.voda, performance of STM are presented in Section VI. Finally,
gildas.besancon@gipsa-lab.inpg.fr) Section VII draws some conclusions.

978-1-4244-7427-1/10/$26.00 ©2010 AACC 6555

 Fig. 1. Complete simulation model for STM closed-loop system

II. STM SYSTEM DESCRIPTION actuator attached with STM tip to move it in appropriate
direction according to the applied voltage (v 2 ) in order to
A. Working Principle
keep the distance (d ) constant. An amplifier (Gain = 4) of
Scanning tunneling microscope (STM) works by scan- bandwidth 100 k H z is used before piezoelectric actuator
ning a very sharp metal wire tip over a surface. By bringing as the output of controller (v 1 ) can be between ±10 V .
the tip very close to the conducting sample surface (less The piezoelectric actuators are now widely used for high
than 1×10−9 m), and by applying an electrical voltage v b positioning accuracy at nanometer and sub-nanometer
(v b = 0.1 V ) to the tip or sample, a small tunnel current resolution with high bandwidths [13], [14]. One of the
(i t ) is produced between STM tip and sample surface. This advantages of using piezoelectric actuators is that under
tunnel current depends exponentially on the distance (d ) certain experimental conditions their dynamics can be
between STM tip and sample surface with following non- well approximated by linear models [15], that’s why a
linear relation : p second order linear model can be used for piezoelectric
i t = i o · e −α Φ · ∆d (1) actuator :
γ0
G a (s) = µ ¶ (3)
where Φ is the work function, ∆d is the variations in the 1
³
1
´
s 2+ s + 1
distance (d ), α is a constant term (α = 1.025 eV −1 Å−1 ) ω2 0
Qω0
and i o is the initial tunnel current when ∆d = 0. Control-
ling this tunnel current (i t ) by keeping the distance (d ) where γ0 is the sensitivity (85 × 10−10 mV −1 ), ω0 is the
constant (∆d = 0) in the presence of external disturbances resonance frequency (28 k H z) and Q is the quality fac-
(sensor noise (n), surface variations (z S ) etc.) is the main tor (4.5) of piezoelectric actuator model. The nonlinear
objective of the feedback control system of STM. A com- phenomenon like hysteresis is not expected for the pie-
plete overview of the closed-loop control scheme which zoelectric actuator as the amplitude of input voltage (v 2 )
will be here considered is presented in Fig. 1. is very small for the vertical movement (z-direction) of
A feedback loop constantly monitors the tunneling STM tip. The output of the piezoelectric actuator (z) is
current (i t ). The current amplifier (pre-amplifier or I- used to find out the distance (d ) between STM tip and
V converter) converts the small tunneling current into sample surface (z s ) from d = z 0 −z −z s (Fig. 1) where z 0 is
a voltage (v 3 ) with a gain (R) of 1 × 109 Ω (bandwidth the initial position of STM tip when no voltage is applied
15 k H z). This pre-amplifier is usually the most important to the piezo.
source of noise (n) in the loop. A logarithmic amplifier
B. Tunneling Characteristics of STM with Experimental
(bandwidth 60 k H z) is used in the feedback loop to
Setup
deal with exponential nonlinearity and to make the entire
electronic response linear (approximately) with respect to Some experimental tests of STM are performed in
the distance (d ). The output (v y ) is given by the following Institut Néel at room temperature in order to observe the
nonlinear relation : tunneling characteristics with the change in the distance
µ ¶ between STM tip and sample surface. These results helped
|v 3 |
v y = K L · l og 10 (2) us to observe the tunneling phenomenon and to identify
EL
its characteristics. Different tests are performed by moving
where K L is the conversion factor (2.5 V ) and E L is the the STM tip vertically upward (7×10−10 m away from the
sensitivity of log amplifier (0.001 V ). The piezoelectric surface) and vertically downward (7×10−10 m towards the

6556
surface) in order to observe the variations in the tunneling
current with the distance. Test results are plotted in Fig.
2. The variations in the different tunnel current profiles
are because of the different initial conditions and also
the environmental condition which includes temperature
variations, vibrations, noise etc. We tried to simulate the
tunneling characteristics and identified the value of the
work function Φ. We observed that with work function Φ =
0.5 eV , our simulation approximately follows the tunnel
current profile obtained from the experiment (Fig. 2). For
simulations and synthesis of control, we have taken the
experimentally identified value of work function and large
variation (20% from the nominal value) of Φ is considered
for uncertainty model of the system. The parameter values
of piezoelectric actuator and amplifiers in the closed loop,
Fig. 2. Experimental and simulated response of tunnel current
used in this article, are taken from the experimental setup.

C. Control Design Model

The complete simulation model (Fig. 1) needs to be
transformed into an appropriate linear design model (Fig.
3) which is required for the linear controller design for
feedback control system of STM.
A first order linear approximation approach is used to
linearize both non-linearities Eq. (1) and Eq. (2) indepen-
dently around their equilibrium points (d 0 and v 30 ). The
corresponding linearized equations are respectively :

i t = c 1 + c 3 − c 2 · ∆d (4)
Fig. 3. Design model for closed-loop system of STM
v y = c4 − c6 + c5 · v 3 (5)

where c 1 , c 2 , . . . , c 6 are constants which depend on the

III. NOMINAL AND UNCERTAINTY MODEL
parameters of Eq. 1 and Eq. 2 and defined by the following
relation : The nominal model of the plant G 0 (s) is obtained
p
−α Φ · d 0 considering the values of physical parameters of STM
c1 = io · e
p taken from the experimental setup. Then, its worst case
c2 = α Φ · c1
behavior is analyzed by taking into account the variations
c3 = d0 · c2
of the model parameters inside a given range of values.
v 30
µ ¶
c4 = K L · l og 10 (6) The nominal values of STM model, together with their
EL
KL percentage variation are shown in Table I. In order to
c5 = check the robust stability (RS) and robust performance
v 30 · l n (10)
KL (RP) conditions, a set of plants Π may be represented
c6 =
l n (10) by multiplicative uncertainty with a scalar weight W I (s)
using :
Now, the feedback dynamics (Fig. 1) where tunnel current
(i t ) is converted into a voltage (v y ) can be represented by G ∆ (s) = G 0 (s)(1 + W I (s)∆I (s)) (8)
a 2nd order linear model H (s).
c · ω1 · ω2 TABLE I
H (s) = (7)
s 2 + (ω1 + ω2 ) s + ω1 · ω2 VALUES OF THE STM MODEL PARAMETERS TOGETHER WITH THEIR %
VARIATION
where c is a constant term depending on parameters of
Eq. 4 and Eq. 5 and ω1 , ω2 are two bandwidths of the pre-
Parameter Value % variation
amplifier and logarithmic amplifier. After linearization, the ω0 [k H z] 28 ± 1.4 5%
equivalent linear control design model is given in Fig. 3 γ0 [mV −1 ] 85 × 10−10 ± 4.25 × 10−10 5%
where G(s) represents the 3r d order linear model including Q 4.5 ± 0.225 5%
Φ [eV ] 0.5 ± 0.1 20 %
piezo pre-amplifier and piezoelectric actuator model and
R [Ω] 1 × 109 ± 0.05 × 109 5%
G n is a constant term which represents the noise (n) ω1 [k H z] 15 ± 0.75 5%
transfer.

6557
 TABLE II
where |∆( j ω)| ≤ 1, ∀ω represents the normalized real per-
PARAMETERS FOR THE RATIONAL WEIGHT W I (s) ( THE VALUES OF THE
turbations and G ∆ ∈ Π. In case of multiplicative uncer-
FREQUENCIES ARE EXPRESSED IN RAD / SEC )
tainty model, the relative error function can be computed
as :
¯ G ∆ ( j ω) −G 0 ( j ω) ¯
¯ ¯
A M ζ1 ζ2 ωn ωB
l I (ω) = max ¯¯ ¯ (9) 0.21 0.165 0.7 0.33 1.9 × 105 2.4 × 105
G ∆ ∈Π G 0 ( j ω) ¯

and with rational weight W I ( j ω) is chosen as follows :

| W I ( j ω) | ≥ l I (ω), ∀ω (10) surface. It can also be formulated as a disturbance re-

Relative errors l I ( j ω) together with the rational weight jection problem. The variations in the sample surface (z S )
W I ( j ω) for 729 possible combinations of the STM physical and also noise (n) are considered as external disturbances
parameters are plotted in Fig. 4. The weighting function where the first one can be considered as a slow varying
W I ( j ω) is chosen equal to a third order transfer function disturbance and the latter one can be considered as a
in a way that it includes the set of all possible plant fast varying disturbance. These disturbances are rejected
models. by moving the STM tip in appropriate direction so that the
distance (d ) should always remain constant at its desired
((1/ωB )s + A)(s 2 + 2ζ1 ωn · s + ω2n )
W I (s) = (11) value (0.8 × 10−9 m).
((1/(ωB M ))s + 1)(s 2 + 2ζ2 ωn · s + ω2n ) The main objective of the control system is to achieve
The values considered for all parameters of W I (s) are given better performance of STM in terms of high positioning
in Table II. accuracy ±8×10−12 m in the presence of good robustness
Using the small gain theorem, the condition for Robust margin (kSk∞ ≤ 6 d B and kT k∞ ≤ 3.5 d B and stability
Stability (RS) is given by [16] : margins (gain margin > 6 d B and phase margin > 30o ).
Such positioning accuracy is required with high closed-
1
RS ⇐⇒| T | < , ∀ω (12) loop bandwidth in order to get fast scan speed as slower
| WI |
scan speed can introduce certain considerable points,
Considering the performance specifications in terms of like drift. The desired performance is required with the
the sensitivity function, the condition for Robust Perfor- maximum continuous surface variations of frequency 1 ×
mance (RP) is obtained as [16] : 103 r ad /sec having amplitude 4×10−10 m in thep presence
RP ⇐⇒ | W1 S | + | W I T | < 1, ∀ω (13) of sensor (pre-amplifier) noise (n) of 45 mV / H z.

where S and T are sensitivity function and complementary V. H∞ CONTROL DESIGN APPLIED TO STM
sensitivity function respectively. The desired performances are imposed on the closed-
loop sensitivity functions using appropriate weighting
IV. CONTROL PROBLEM AND DESIRED functions and then the mixed-sensitivity H∞ control de-
PERFORMANCE sign methodology is adopted to fulfill the requirements.
For a feedback control of the STM, the control problem The functions W1 , W2 and W3 weight the controlled
can be formulated as a tracking problem where the STM outputs y 1 , y 2 and y 3 respectively (Fig. 3) and should be
tip tracks the unknown sample surface (z S ) by keeping chosen according to the performance specifications.
the distance (d ) constant between STM tip and sample The generalized plant P (Fig. 5) (i.e. the interconnection
of the plant and the weighting functions) is given by :

y1 W1 W1 H −W1 HG n −W1 HG VRE F

    
 y2   0 W2 
0 0   ZS 
 
 =
 y3   0 −W3 0 W3 G   n 
VE I H −HG n −HG V1
| {z }
P

Thus, the H∞ control problem is to find a stabilizing

controller K (s) which minimizes γ [16] such that :
° °
° W1 S W1 H S −W1 HG n S °
° <γ
° °
° W2 K S W2 H K S W2 HG n K S ° (14)
°
° W GK S −W3 S −W3G n T °
3 ∞
We have chosen the weighting functions as follows :

Fig. 4. Relative plant errors l I ( j ω) and rational weight W I ( j ω) for 729 (1/M s ) s + ωs
possible combinations of the STM physical parameters W1 (s) = (15)
s + ωs · ²s

6558
 Fig. 5. Generalized design model for closed-loop system of STM Fig. 7. Closed-loop sensitivity function (KS) with H∞ control

s + (ωu /M u )
W2 (s) = (16)
²u · s + ωu
s + (ωt /M t )
W3 (s) = (17)
²t · s + ωt
The values considered for all parameters of above three
weighting functions in order to achieve desired perfor-
mance are given in Table III. After computation, the

TABLE III
PARAMETERS FOR PERFORMANCE WEIGHTING FUNCTIONS W1 (s), W2 (s)
AND W3 (s) ( THE VALUES OF THE FREQUENCIES ARE EXPRESSED IN RAD / SEC )
Fig. 8. Closed-loop sensitivity function (T) with H∞ control

Ms ²s ωs Mu ²u ωu Mt ²t ωt
2 0.0089 1 × 105 3.2 1 1 × 107 1.5 1 1 × 107
the requirements as the obtained sensitivity functions
fairly match the desired loop shaping (Fig. 6-8).
From robustness point of view, we obtained good mo-
minimal cost achieved for STM feedback control system dulus margin as kSk∞ = 1.55 d B and kT k∞ = 0.08 d B and
was γ = 1.6 which means that the obtained sensitivity good stability margins (gain margin = 10.1 d B and phase
functions match nearly the desired loop shaping. The ob- margin = 66.1o ). The obtained closed-loop bandwidth
tained sensitivity functions with the desired loop shaping is 2.5 × 105 r ad /sec which ensures the required good
in terms of weighting filters are shown in Fig. 6-8. performance with fast variations (1 × 103 r ad /sec) in the
sample surface (z S ). Similarly, all other constraints in
A. Control Loop Performance Analysis terms of better noise (n) rejection and to avoid actuator
The weighting functions (W1 , W2 and W3 ) were de- saturations are fully met with proposed control technique.
signed considering the requirement of high positioning
accuracy ±8 × 10−12 m with high bandwidth and good
VI. SIMULATION RESULTS
robustness. The proposed control technique achieves all
After the control design and closed-loop analysis, we
can now validate its performance with a simulation mo-
del, having actual non-linearities, sensor noise (n) and
physical limitations in closed-loop, aiming at representing
a real system as close as possible.
Fig. 9 shows the simulation result with the proposed
H∞ control technique in presence of surface variations
z S with a frequency of 1 × 103 r ad /sec and an amplitude
of 4 × 10−10 m, and in the presence p of sensor noise (n)
(in the pre-amplifier) of 45 mV / H z. The dotted lines
represent the positioning accuracy (acceptable bounds) of
±8 × 10−12 m. It can be observed that the movement of
the STM tip remains within the desired limits.
Finally, we have verified the robust stability (RS) and
Fig. 6. Closed-loop sensitivity function (S) with H∞ control robust performance (RP) conditions as mentioned in Eq.

6559
 Fig. 11. System robust performance tested with H∞ control technique
Fig. 9. Simulation result with H∞ control having surface variations z S
of frequency of 1×103 r ad /sec, an amplitude
p of 4×10−10 m and in the
presence of sensor noise (n) of 45 mV / H z
R EFERENCES
[1] G. Binning and H. Rohrer, Scanning Tunneling Microscopy, IBM J.
Res. Develop., vol. 30, pages 355-369, 1986
[2] C.J. Chen, Introduction to scanning tunneling microscopy, 2nd
edition, Oxford science publications, 2008
[3] E. Anguiano, A.I. Oliva and M. Aguilar, Optimal conditions for ima-
ging in scanning tunneling microscopy : Theory, Rev. Sci. Instrum.,
Vol. 69 (2), pages 3867-3874, November 1998
[4] A.I. Oliva, E. Anguiano, N. Denisenko, M. Aguilar and J.L. Pena,
Analysis of scanning tunneling microscopy feedback system, Rev. Sci.
Instrum., Vol. 66 (5), pages 3196-3203, May 1995
[5] N. Bonnail, D. Tonneau, F. Jandard, G.A. Capolino and H. Dal-
laporta, Variable structure control of a piezoelectric actuator for
a scanning tunneling microscope, IEEE transaction on Industrial
Electronics, Vol. 51 (2), pages 354-363, April 2004
[6] I. Ahmad, A. Voda and G. Besançon, Controller design for a closed-
Fig. 10. System robust stability tested with H∞ control technique loop scanning tunneling microscope, 4th IEEE Conference on Auto-
mation Science and Engineering, Washington DC, USA, pages 971-
976, August 2008
[7] I.D. Landau and A.Karimi, Robust digital control using pole pla-
12 and Eq. 13. Simulation result shows that robust stability cement with sensitivity function shaping method, Int. J. Robust
(RS) condition is satisfied (Fig. 10) i.e. the closed-loop Nonlinear Control, vol. 8, pages 191-210, 1998
[8] H. Prochazka and I.D. Landau, Pole placement with sensitivity
system remains stable for all perturbed plants around
function shaping using 2nd order digital notch filters, Automatica
the nominal model up to the chosen worst-case model 39 (2003), pages 1103-1107, February 2003
uncertainty. Fig. 11 shows that robust performance (RP) [9] S. Salapaka, A. Sebastian, J.P. Cleveland and M.V. Salapaka, High
achieved according to the condition given in Eq. 13. bandwidth nano-positioner : A robust control approach, Rev. Sci.
Instrum., Vol. 73 (9), pages 3232-3241, September 2002
VII. CONCLUSIONS [10] G. Schitter, K.J. Astrom, B.E. DeMartini, P.J. Thurner, K.L. Turner and
P.K. Hansma, Design and Modeling of a High-Speed AFM-Scanner,
This article has dealt with H∞ control design for STM, IEEE transaction on Control Systems Technology, Vol. 15 (5), pages
under high performance specifications and real parame- 906-915, September 2007
tric uncertainties. Tunneling characteristics of STM are [11] D.Y. Abramovitch, S.B. Andersson, L.Y. Pao and G. Schitter,A Tutorial
on the Mechanisms, Dynamics, and Control of Atomic Force Micro-
analyzed experimentally and an important parameter Φ scopes, American Control Conference, NY, USA, pages 3488-3502,
is identified. Performance in terms of high positioning July 2007
accuracy (±8 × 10−12 m) with high bandwidth and good [12] I. Ahmad, A. Voda and G. Besançon, H∞ controller design for high
performance scanning tunneling microscope, 48th IEEE Conference
robustness is achieved. Simulation results show that RS on Decision and Control, Shanghai, China, December 2009
and RP conditions are fulfilled under parametric variations [13] M.E. Taylo, Dynamics of piezoelectric tube scanners for scanning
of the system model. probe microscopy, Rev. Sci. Instrum., Vol. 64 (1), pages 154-158,
1993
Real-time experimental validation of the H∞ controller
[14] G. Schitter, and A. Stemmer, Identification and open-loop tracking
is currently in progress. control of a piezoelectric tube scanner for high-speed scanning probe
microscopy, IEEE transaction on Control System Technology, vol 12
Acknowledgement : The authors would like to sincerely (3), pages 449-454, 2004
[15] B. Bhikkaji, M. Ratnam, A.J. Fleming and S.O.R. Moheimani, High-
thank Prof. Hervé Courtois for providing the access to an
performance control of piezoelectric tube scanners, IEEE transaction
experimental STM setup of the Institut Néel in Grenoble. on Control Systems Technology, Vol. 15 (5), pages 853-866, Septem-
Also we would like to thank M. Thomas Quaglio from ber 2007
Institut Néel for helping in all experimental tests and his [16] Skogestad, S and Postlethwaite, Multivariable feedback control :
analysis and design, John Wiley and Sons, 1996
useful discussions.

6560