Journal of Crystal Growth 201/202 (1999) 40}44

Feedback control of substrate temperature during the growth of

near-lattice-matched InGaAs on InP using di!use re#ection
spectroscopy
S.R. Johnson*, E. Grassi, M. Beaudoin, M.D. Boonzaayer, K.S. Tsakalis, Y.-H. Zhang
Center for Solid State Electronics Research, Department of Electrical Engineering, Arizona State University, Tempe, AZ 85287-6206, USA

Abstract

Di!use re#ection spectroscopy (DRS) is used to control substrate temperature to within $23C of user speci"ed
setpoint during the growth of near-lattice-matched InGaAs on InP. The same growth under constant thermocouple
control would result in a 503C rise in real substrate temperature. Feedback control is achieved using a nested
proportional-integral-derivative (PID) control loop; the inner loop consists of a conventional Eurotherm-thermocouple
feedback loop that controls the substrate heater power; the outer loop updates the thermocouple setpoint based on the
di!erence between the user setpoint and the substrate (DRS) temperature using a PID control loop implemented in
the control software. Frequency loop shaping, based on a dynamical model of the system obtained from an identi"cation
experiment, is used to tune the outer PID loop. In addition, the thermal disturbances that occur during e!usion
cell shutter operations must be rejected. In the simplest case, a single correcting step in the Eurotherm (thermocouple)
setpoint is input when a shutter is toggled. Through disturbance identi"cation and model inversion a more sophisticated
disturbance rejection action from the controller can be obtained.  1999 Elsevier Science B.V. All rights reserved.

PACS: 68.55; 78.20.Ci; 81.15.Hi

Keywords: MBE; Closed-loop control; Semiconductors

1. Introduction rate reliable temperature measurement method is

Substrate temperature control is critical in deter-
mining the quality and composition of deposited
layers in molecular beam epitaxy (MBE). As a "rst
step in controlling substrate temperature, an accu-
* Corresponding author. Tel.: #1 602 965 2565; fax: #1 602
965 8118; e-mail: shane.johnson@asu.edu.
required. Recently, several optical methods that
infer substrate temperature from the bandgap of
the substrate material have been developed [1}5].
In these methods the temperature of the substrate is
given by the spectral position of the bandedge of
the substrate material. The bandedge of semicon-
ductors typically shifts toward lower energies and
broadens with temperature [6].

0022-0248/99/$ } see front matter  1999 Elsevier Science B.V. All rights reserved.
PII: S 0 0 2 2 - 0 2 4 8 ( 9 8 ) 0 1 2 7 3 - 1
 S.R. Johnson et al. / Journal of Crystal Growth 201/202 (1999) 40}44
Currently, we use di!use re#ection spectroscopy
(DRS) [1,2] to measure the bandedge of InP and
GaAs substrates. In this method, the substrate is
illuminated by a broad spectrum W}halogen lamp
placed at an e!usion cell port; part of this light is
transmitted through the substrate, di!usely re#ec-
ted, transmitted back through the substrate, and
detected at the pyrometer port. Substrate temper-
ature is inferred from the wavelength of the onset of
substrate transparency to the di!usely re#ected
light.
Radiatively heated substrates are known to heat
up dramatically during the growth of absorbing
overlayers, such as near-lattice-matched InGaAs
on InP, when constant thermocouple control is
used [7,8]. In this case, the thermocouple is insensi-
tive to the substrate temperature increase caused by
the strong absorption of heater radiation in the
overlayer. Under these growth conditions, it is im-
perative that one has an alternative method of
controlling substrate temperature.
2. Nested proportional-integral-derivative loop
control
The nested loop approach shown in Fig. 1 is
a convenient way to add an accurate external tem-
perature sensor to the conventional Eurotherm-
thermocouple control loop found in most MBE
systems. In this case, the outer control loop updates
Fig. 1. Schematic diagram of the nested proportional-integral-
derivative control loop used to control substrate temperature.
The inner loop is the conventional Eurotherm-thermocouple
control loop. The outer loop updates the Eurotherm (ther-
mocouple) setpoint using the di!erence between the user set-
point and the substrate temperature measured by di!use
re#ectance spectroscopy (DRS).
41
the Eurotherm (thermocouple) setpoint based on
the di!erence between the user setpoint and the
substrate temperature measured by DRS, using
a proportional-integral-derivative (PID) algorithm
implemented in the control software. This method
has the advantage that the Eurotherm-thermo-
couple control system is left intact, so that the
heater power will continue to be controlled in the
event that the optical thermometer or the control
computer fails.
A near-lattice-matched, 400 nm thick, InGaAs
layer is grown on a radiatively heated 50 mm semi-
insulating InP substrate in a DCA MBE system,
using the DRS sensor and the nested PID control
loop shown in Fig. 1. The substrate (i.e., DRS)
temperature, thermocouple temperature, and
heater voltage recorded during the growth of this
layer are shown in Fig. 2. The DRS sensor is used
to control the inner loop (thermocouple) setpoint,
which in turn controls the substrate heater voltage
(or power). The supervisory computer samples the
DRS temperature every 5}30 s (30 s in Fig. 2); if the
DRS temperature has deviated more than $13C
from the substrate setpoint of 4353C, the outer PID
loop updates the Eurotherm (inner) setpoint.
Throughout most of the growth of the InGaAs
layer, the substrate temperature is about 13C above
its setpoint. Decreasing the update time would
Fig. 2. Feedback control of substrate temperature during the
growth of near-lattice-matched InGaAs on semi-insulating InP,
using di!use re#ectance spectroscopy (DRS). The substrate set-
point (4353C), the DRS temperature, the thermocouple temper-
ature and setpoint, and the heater voltage (at power supply) are
shown.
42 S.R. Johnson et al. / Journal of Crystal Growth 201/202 (1999) 40}44
bring the substrate temperature closer to the set-
point during the "rst part of the growth. As can be
seen in Fig. 2, each time the thermocouple setpoint
is updated, there is a sharp drop in the heater
voltage and the thermocouple temperature quickly
moves to its new setpoint. The thermal response of
the thermocouple}heater system is fast (4 s time
constant) compared to the thermal response of the
substrate (60 s time constant); this allows appropri-
ate control action.
The small-bandgap epilayer absorbs a larger
part of the heater radiation spectrum than the InP
substrate, which is transparent to much of the
blackbody spectrum at these temperatures. There-
fore, as the InGaAs layer thickens, the transmission
losses of heater radiation are reduced. In order to
maintain a constant substrate temperature, the
control system reduces the heater power by de-
creasing the Eurotherm (thermocouple) setpoint.
The small oscillations in the substrate temper-
ature at the start and stop points are caused by the
abrupt changes in the thermal load on the substrate
when the In and Ga cells are opened or closed. At
the start of growth the thermocouple setpoint is
reduced by 63C in an e!ort to correct for the ther-
mal disturbance caused by additional radiant heat
load of the In and Ga e!usion cells. This is the
simplest way of rejecting thermal disturbances. At
the end of the growth, no shutter disturbance rejec-
tion is used and the substrate momentarily drops
about 43C below the setpoint. A more sophisticated
approach to disturbance rejection is discussed in
the following section.
3. Outer loop tuning
In order to improve feedback control of substra-
te temperature, the outer PID is tuned using fre-
quency loop shaping (FLS) [9], which attempts to
approximate in a minimax sense the loop transfer
function of the system to a desired one in the
frequency domain. The ideal loop transfer function
is selected, based on stability and performance con-
siderations, and on the characteristics of the sys-
tem. FLS requires a model of the system (in the
frequency domain); this model can be obtained
using standard system identi"cation techniques.
As a "rst step, an input signal is designed that
covers the frequency range of interest. Next, the
temperature of the substrate is brought to and
allowed to stabilize at the desired operating point.
With the external (outer) control loop open, the
identi"cation signal is applied at the Eurotherm
(inner) setpoint (see Fig. 1), while the DRS temper-
ature is collected. Using the method described in
Ref. [10], a state space model is found [11], that
describes the dynamic behavior of the system from
the power supply input to the DRS temperature
measurement. (A state space model is a set of forced
linear ordinary di!erential equations.) The DRS
temperature, the temperature predicted by the
model, and the identi"cation signal are shown in
Fig. 3. A second-order state space model was used
to identify the system. The model was veri"ed using
a separate data set.
The desired time constant of the controlled sys-
tem is about 15 s. Therefore, we need to obtain
a model of the system around the frequency,
 rad/s. The input signal for the identi"cation ex-

periment is a random-switching binary signal with
levels of $103C around the desired operating
point. The switching time is uniformly distributed
over 1}160 s, with a mean of 80 s. Since the samp-
Fig. 3. Identi"cation experiment used to determine the state
space model that relates heater voltage to substrate temperature.
The identi"cation signal (Eurotherm thermocouple setpoint),
substrate temperature measured by di!use re#ectance spectro-
scopy (DRS), and substrate temperature predicted by the model,
are shown.
 S.R. Johnson et al. / Journal of Crystal Growth 201/202 (1999) 40}44
ling interval is 1 s, the maximum frequency spanned
is 1 Hz. This is adequate for our goals. The fre-
quency resolution is 1/N Hz, where N is the num-
ber of data points collected. Usually a resolution of
 of the closed loop bandwidth is enough for this

type of system where no sharp resonant peaks are
present.
The PID parameters obtained with FLS are very
di!erent from those obtained with the traditional
step response PID tuning method (Ziegler}Nichols
method 1) [11]. The Ziegler}Nichols rule gives
[K , ¹ , ¹ ]"[8.4, 12 s, 3.0 s], while FLS gives
 
[K , ¹ , ¹ ]"[2.1, 34 s, 0.7 s]. The Ziegler}Nichols
 
method approximates the system by a "rst-order
model with delay, and is only applicable for &s'
shaped step responses (see Ref. [11], p. 670). As can
be seen in Fig. 3, the lag in the step response (6 s) is
very short compared with the rise time (42 s), which
makes the Ziegler}Nichols rules inappropriate
for the present system. Using the Ziegler}
Nichols rule in this case, results in a closed loop
system that is unstable.
Another important control objective is good
temperature regulation during external distur-
bances, such as thermal shutter transients. Since it
is known precisely when each e!usion cell will be
opened or closed a control algorithm that takes
preventive action can be used. Identi"cation experi-
ments similar to the one described above can be
used to model the disturbance signal associated
with shutter operations. Based on the inverse
model, corrective action in the form of a time-de-
pendent signal added at the Eurotherm (inner) set-
point, can be used to reject the disturbance. This
time-dependent signal takes into account the dy-
namics of the system, which are not included in
a simple step adjustment used in Fig. 2. Inclusion of
the system dynamics will be most visible in cases
where the substrate response lags the shutter dis-
turbance.
Optical thin "lm interference is another source of
disturbance present when growing periodic struc-
tures, such as, quarter wave mirror stacks [12].
This disturbance cannot be modeled with regular
state space models. However, since its e!ect is
a low-frequency disturbance (time constant on the
order of minutes) compared to the desired band-
width of the controller, the outer PID should reject
43
most of it.
During the growth of thick absorbing overlayers,
the absorption}emission properties of the sub-
strate-epilayer system change considerably. This
means that in general, the optimal PID parameters
would change accordingly and several PIDs would
have to be scheduled. Gain scheduling was not used
in the growth shown in Fig. 2; in this case a single
set of PID parameters was su$cient to control
temperature. Finally, the DRS signal is reduced
when an absorbing layer is present, as can be seen
by an increase in the noise of the DRS data in
Fig. 2. Since the data acquisition frequency of the
DRS data is much larger than that of the control
system, much of the noise in the DRS data is
removed at the DRS sensor input, using a low-pass
"lter.
4. Conclusions
A nested proportional-integral-derivative con-
trol loop is used to achieve feedback control of
substrate temperature during the growth of near-
lattice-matched InGaAs on InP to within $23C of
the user speci"ed setpoint. The outer control loop
controls the inner loop (thermocouple) setpoint us-
ing di!use re#ection spectroscopy as an external
temperature sensor. The inner loop is the standard
Eurotherm-thermocouple control loop commonly
found on most MBE systems. A system identi"ca-
tion experiment and frequency loop shaping are
used to optimally tune the PID parameters of the
outer control loop. Thermal shutter transients
are rejected at the input of the Eurotherm (inner)
control loop.
Acknowledgements
This work is part of a DARPA-funded program
entitled `Integrated MultiSensor Control of Mo-
lecular Beam Epitaxya under Grant No.
MDA972-95-1-0016, monitored by Lt. Col. Gernot
S. Pomrenke, and is part of a collaborative e!ort
involving scientists from Arizona State University,
the University of Colorado, HRL laboratories, and
the J.A. Woollam Co.
44 S.R. Johnson et al. / Journal of Crystal Growth 201/202 (1999) 40}44
References
[1] S.R. Johnson, C. Lavoie, T. Tiedje, J.A. Mackenzie, J. Vac.
Sci. Technol. B 11 (1993) 1007.
[2] S.R. Johnson, C. Lavoie, M.K. Nissen, T. Tiedje, US Pat-
ent No. 5,388,909, 1995.
[3] J.A. Roth, T.J. DeLyon, M.E. Adel, Mater. Res. Soc. Symp.
Proc. 324 (1994) 353.
[4] E.S. Hellman, J.S. Harris, J. Crystal Growth 81 (1987) 38.
[5] D.M. Kirillov, R.A. Powell, US Patent No. 5,118,200,
1992.
[6] S.R. Johnson, T. Tiedje, J. Appl. Phys. 78 (1995) 5609.
[7] B.V. Shanabrook, J.R. Waterman, J.L. Davis, R.J. Wagner,
Appl. Phys. Lett. 61 (1992) 2338.
[8] J.A. Roth, Mater. Res. Soc. Symp. Proc. 502 (1998) 97.
[9] E. Grassi, K.S. Tsakalis, Proc. 35th CDC, Kobe, Japan,
December 1996, pp. 4776}4781.
[10] K. Tsakalis, K. Stoddard, Proc. 6th IEEE Int. ETFA
Conf., Los Angeles, CA, September 1997, pp. 514}519,
[11] K. Ogata, Modern Control Engineering, 3rd ed., Prentice-
Hall, NJ, 1997.
[12] S.R. Johnson, C.-H. Kuo, M.D. Boonzaayer, W. Braun,
U. Koelle, Y.-H. Zhang, J.A. Roth, J. Vac. Sci. Technol.
B 16 (1998) 1502.