Autonomous on-wafer sensors for process modeling, diagnosis, and control

Abstract

This paper explores the feasibility of constructing an autonomous
sensor array on a standard silicon wafer. Such a sensor-wafer would
include integrated electronics, power, and communications, and would be
capable of being placed into a standard production process step, or
short sequence of steps. During the processing of the sensor-wafer,
various process parameters would be measured and recorded. There are
several uses for such a sensor wafer, including equipment
characterization and design, process calibration, and equipment
qualification and diagnosis. In this paper, various sensor
architectures, power supplies, communications methods, and isolation
techniques are discussed, and particular choices are made. Several
proof-of-concept designs that measure film-thickness and temperature are
discussed, and test results are reviewed for each design

Full text

Page 1

IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 14, NO. 3, AUGUST 2001255
Autonomous On-Wafer Sensors for Process
Modeling, Diagnosis, and Control
Mason Freed, Student Member, IEEE, Michiel Krüger, Student Member, IEEE, Costas J. Spanos, Fellow, IEEE, and
Kameshwar Poolla, Fellow, IEEE
Abstract—This paper explores the feasibility of constructing
an autonomous sensor array on a standard silicon wafer. Such
a sensor-wafer would include integrated electronics, power, and
communications, and would be capable of being placed into a
standard production process step, or short sequence of steps.
During the processing of the sensor-wafer, various process param-
eters would be measured and recorded. There are several uses for
such a sensor wafer, including equipment characterization and
design, process calibration, and equipment qualification and diag-
nosis. In this paper, various sensor architectures, power supplies,
communications methods, and isolation techniques are discussed,
and particular choices are made. Several proof-of-concept designs
that measure film-thickness and temperature are discussed, and
test results are reviewed for each design.
Index Terms—Control, calibration, in situ, monitoring, real-
time, sensor.
I. INTRODUCTION
O
metrology to in-line metrology. Wafer measurement equipment
has been moved, where possible, from stand-alone measure-
ment stations to integrated measurement systems on or near
the processing equipment. The benefits of this shift have been
significant. Among the advantages of in line metrology are
improved process monitoring, reduced product variance, and
higher throughput. By placing the sensors on the equipment,
every wafer may be examined, as opposed to just a fraction,
as is the case with standalone metrology stations. Because
more data is available, process fluctuations and trends can be
better monitored and recorded. Also, because the data is taken
more frequently, adjustments to the process can be made more
frequently, so product variance can be reduced. Finally, by
measuring all of the wafers in-line and allowing them to con-
tinue instead of removing selected wafers for metrology, more
VER the past few years, the semiconductor processing
industry has undergone a paradigm shift from ex situ
Manuscript received April 16, 2000, 2001; revised November 24, 2000.
This work was supported by the University of California SMART under Grant
SM97-01, by the National Science Foundation under Grant ECS-96-28420, by
the Department of Defense Graduate Research Fellowship, and by the Intel
Foundation Graduate Research Fellowship.
M. Freed was with the Department of Electrical Engineering and Computer
Science, University of California at Berkeley, Berkeley, CA 94720-1772 USA.
He is now with OnWafer Technologies, Inc., Pleasant Hill, CA 94523 USA.
M.KrügerandK.PoollaarewiththeDepartmentofMechanicalEngineering,
University of California at Berkeley, Berkeley, CA 94720-0001 USA (e-mail:
kruger@jagger.me.berkeley.edu; poolla@jagger.me.berkeley.edu).
C. J. Spanos is with the Department of Electrical Engineering and Computer
Science, University of California at Berkeley, Berkeley, CA 94720-1772 USA
(e-mail: spanos@eecs.berkeley.edu).
Publisher Item Identifier S 0894-6507(01)06551-4.
production wafers can reach process completion, improving
throughput.
While the benefits of in-line metrology are numerous, the
money and time spent to integrate metrology stations onto
equipmentisnot insignificant.Inaddition, equipmentengineers
are reluctant to modify existing equipment designs to allow
the addition of sensors and associated hardware, because such
changes could affect process stability and this work is expen-
sive. Also, if the metrology portion of the equipment goes down
during production, the equipment must also be taken down to
allow repairs to be performed, reducing the throughput of the
machine. For these reasons, the next paradigm shift might be
from sensors on the equipment to sensors on the wafer.
Such on-wafer sensors could provide the same information
about the wafer and process state as is currently available
through equipment-based in situ sensors, but without the
added cost and complexity of integrating the sensor onto the
equipment. Such a wireless sensor-wafer could be loaded into
a boat along with product wafers and sent into the processing
chamber. Then, as the sensor-wafer is processed, it would either
transmit out (via RF, IR, or other wireless method) or record
in on-board memory the measurement data. In this fashion the
same process information is gleaned, but with minimal invasion
into the process chamber.
In this paper, the feasibility of such an on-wafer, spatially-
resolved sensor is investigated. A background is then given for
the basic features and options that such a sensor could have.
Finally, the design, fabrication and testing of several sensor test
designs are discussed.
A. On-Wafer Sensor Applications
Spatiallyresolvedinsitusensor-waferswouldhavemanyim-
portant applications in the semiconductor processing industry.
These include:
1) Process Characterization/Design: One of the more
obvious applications is spatial uniformity characterization
during design and development of process equipment. The
current equipment development procedure involves processing
a set of dummy wafers, measuring the finished wafer parame-
ters, adjusting the machine parameters (chamber geometries,
etc.), and then repeating the process. Because this process is
typically repeated many times, and each iteration can take a
number of days, the entire process can take several months.
The reason that such a lengthy procedure is required is that
with post-process measurement, only an integrated effect is
measured. With a real-time measurement scheme, far fewer
0894–6507/01$10.00 © 2001 IEEE

Page 2

256IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 14, NO. 3, AUGUST 2001
iterations could be performed because much more information
is conveyed during each iteration.
2) Process Calibration: A similar application for in situ
sensors is process parameter adjustment. Traditionally, dummy
wafers are put into the production flow every so often, and
measurements from these processed wafers are used to monitor
and adjust the process. Typically, about 18% of the production
capacity is wasted on process characterization [1]. An alter-
native to this procedure that is offered by in situ sensors is
real-time control. By placing an autonomous sensor into the
process chamber and closing a real-time control-loop around
the tool, process parameters can be optimized automatically, in
one step. By using this method, far fewer test wafers would be
needed, thereby improving the throughput of the equipment.
3) Equipment Diagnosis and Re-Qualification: Another
use for in situ sensors is for equipment diagnosis and re-qual-
ification. If a tool in a fabrication facility is malfunctioning,
the repair staff needs a quick method of determining the source
of the problem, without lengthy disassembly procedures.
Typically, the diagnosis of an equipment problem takes several
hours. Sometimes, this involves venting and disassembling the
process chamber to place wired sensors on the wafer-chuck. By
instead using an in situ sensor, possible sources of the problem
can be eliminated or confirmed more rapidly than by using
this disassembly procedure. In a typical fabrication facility
machine downtime accounts for 10% of the total equipment
time [2]. By reducing this downtime, substantial improvements
in throughput can be realized.
In addition, once a machine has been repaired, it needs to be
re-qualified for return to production flow. This is typically done
by running a minimally processed wafer through the chamber,
and by measuring the results. The measurement process is time-
consuming, because the metrology equipment in a fabrication
facility is highly utilized. By using an in situ sensor, the nec-
essary process parameters can be sensed and displayed imme-
diately, eliminating the lengthy metrology delays, and speeding
the equipment’s return to service.
4) Difficult Measurements: Some measurands are difficult
or destructive to measure using current techniques. A few ex-
amples of this type of variable are etched sidewall profile and
lithographic latent image. Because of the nature of these sensor
wafers, they offer the opportunity to measure some of these
quantities in situ. This would greatly increase the ability of the
engineer to evaluate the process.
B. Organization
This paper is organized as follows. First, a general discussion
of sensor arrays is undertaken: general requirements, associated
difficulties, and possible solutions are discussed. Next, several
prototype sensor-wafer designs are presented, and test results
from each are reported. Finally, a summary of the ideas pre-
sented and the results obtained is provided.
II. DESIGN ISSUES
In developingan autonomous sensor wafer, severalimportant
issues must be considered. These issues can be grouped into
three main categories: power, communications, and isolation.
A. Power
An in situ sensor wafer must contain some type of wireless,
regulated power source to provide power for the electronics and
sensors. There are several constraints on such a power source.
First, to avoid problems with wafer-handling robotics, the pro-
trusionofthepowersourceabovethewafersurfacemustbelim-
itedtoabout3mm.Anotherrequirementforthepowersourceis
that it does not take up excessive area on the wafer. The smaller
thepowersource,themorespaceisavailableforsensors.Lastly,
the power source must be capable of supplying roughly 5-V
output,withaminimumof1-mAcurrent,foratleast5min.This
is approximately the amount of power required to keep elec-
tronics and sensors running for the duration of a typical process
(including loading and unloading).
Given these constraints, several power-supply opportunities
exist. The most appealing candidate is battery-power. A range
of thin, high-energy batteries exist, ranging from commercial
1-mm-thick watch batteries capable of 25 milliamp hours
(mAh), to research-grade 10- m-thick thin-film batteries
capable of 100 microamp hours ( Ah) [3]–[10]. In between
these ranges lie a number of candidates for on-board power.
Anotherpossibilityforapowersourceisthephotovoltaiccell.
In several processes, plasma is used. Because the plasma emits
intense visible light at a range of wavelengths, this light could
be used as a power source. Using photovoltaic cells with a 15%
efficiency [11], a cell area of 1 cm , and 1 mW/cm of available
broadbandlightpowerinsideatypicalplasmachamber[12],the
available electrical power is:
%
mW
cm
cmW(1)
In nonplasma processes, or where this power level is too low,
an external laser could be focused on the photovoltaic cell to
provide power from outside the chamber. Using this method, as
much as 5 mW could be generated using a similar area [13].
Other, more exotic power sources exist, such as using a large
capacitor to store energy for the sensor. In such a scenario, the
capacitorwouldbechargedupbyanexternalpowersourceprior
to being placed into the processing equipment. Then, while the
sensor is being processed, the capacitor would be discharged to
provide power for the sensor. The main problem with this idea
is that, for the typical energy levels needed by the sensor, the
capacitor would have to be much too large. For example, using
typical capacitor materials and dimensions, to maintain a cur-
rent of 1 mA at a voltage above 3.5 V for 5 minutes, the capac-
itor would have to be 6.7 m in area [13], which is clearly much
larger than a wafer. Even using novelcapacitive structures to re-
duce the area, very large areas would still be required. One final
power source option is the direct utilization of the RF energy
present in plasma discharge processes. By inductively coupling
totheRFbiasespresentacrosstheplasmasheath,largeamounts
of power can be extracted. The most difficult aspect of such a
power source is the coupling and regulation of RF energy.
B. Communication
For an in situ sensor wafer to be useful, the data it measures
must be communicated to the outside world. Therefore, several

Page 3

FREED et al.: AUTONOMOUS ON-WAFER SENSORS FOR PROCESS MODELING, DIAGNOSIS, AND CONTROL 257
restrictions exist for the sensor’s communication system. First,
typical semiconductor process variations, such as those occur-
ring in plasma etch, CVD, and lithography, occur across length
scales of several centimeters and time scales of several seconds.
Therefore, the system must be capable of handling measure-
ments from about 100 sensors (to get the desired spatial reso-
lution of 1 sensor per
cm ), each operating at a minimum
frequency of about 1 Hz. Second, the communications system
must allow measurements with reasonably high dynamic range
to be transmitted, at least 8 bits per measurement. Therefore,
the overall communications bandwidth must be at least (100
sensors)*(1 Hz)*(8 bits)
the communications system is that it uses very little power. Be-
cause the power source is only capable of delivering a limited
amount of power and energy, the communications system must
useonlyafraction oftheseamounts.Forthesamereasonsasfor
the powersupply, thecommunications system mustbe wireless,
and must fit within the same size constraints. Lastly, the com-
munication system must not, as much as possible, depend on
the particular geometry of the process-chamber. For example,
if optical communications is used, the light-source must not be
directed only toward the view port in a particular type of equip-
ment, because then this sensor would be useless in other cham-
bers in which the viewport is situated differently with respect
to the wafer chuck. Alternatively, a directional communications
system should be easily reconfigurable to work with multiple
chamber geometries.
For a multisensor wafer the communications can be either
modular or central. For modular communication, each sensor
(or possibly each group of sensors) would have its own commu-
nication system, so that separate sensors or groups of sensors
transmit their data in parallel. With central communications, all
of the sensors on a wafer are connected to a central commu-
nications system, which communicates the data for the entire
wafer. In this way, there would be a single transponder on each
wafer that multiplexes the measurement values, and transmits
them serially. At the receiving end, the receiver demultiplexes
the information to yield sensor information from all points on
the wafer. Typically, the type of communication system (LED,
RF, etc.) will determine whether modular or central communi-
cationswillbeused.Thiswillbediscussedinmoredetailbelow.
Perhaps the easiest communications option is optical trans-
mission. With this method, a light-emitting diode (LED) on
the wafer transmits the sensor data to a photosensitive receiver
outside the chamber using some type of frequency-modulation
scheme. Plenty of low-profile, low-power commercial compo-
nentscurrentlyexistthatcouldprovidethistypeoffunctionality.
This method has the disadvantage that a line-of-sight path to the
wafer is required. In many cases, such a view port exists, but in
several types of processing equipment this is not possible. This
technique can be used with either modular or central commu-
nications, but because LED power requirements are relatively
high, and modular communications would require many LEDs
onthewafer,thismethodislessadvantageous.Therefore,foran
optical scheme, centralized communications would most likely
be employed.
Using radio-frequency (RF) communication is also an op-
tion. By placing an RF oscillator and modulation circuitry on
bps. Another requirement for
Fig. 1. MEMS grating light modulator diagram.
the wafer, sensor data can be transmitted out of the chamber to
external receiver circuitry. Commercial components also exist
that provide RF communication capability, although they use
more energy than their optical counterparts. This has the advan-
tage that it does not depend on chamber geometry (placement
of viewports, etc.) as much as optical transmission. However,
for plasma processes the plasma itself forms a conductive cloud
aroundthewafer,impedingRFtransmission.Anotherdownside
to using RF is the complexity of RF transmission circuits and
their high power requirements. Because of these requirements,
centralized communications techniques would be used for RF.
A third possible communication system is the so-called
grating light modulator (GLM) [14], [15]. This microelec-
tromechanical system (MEMS) consists of a microfabricated
set of refractive beams, placed as shown in Fig. 1. In the
unactuated state, shown on the top of Fig. 1, light that strikes
the surface will be completely reflected, because the surface is
flat. However, when the modulator is actuated, shown on the
bottom of Fig. 1, every other beam is electrostatically pulled
down to the substrate, creating a diffraction grating. In this
way, information can be modulated onto an incoming beam of
light. An SEM photograph of this structure is shown in Fig. 2.
The major advantage of this method of communication is that it
requires very little power. Since the only power required is that
needed to electrostatically move the bottom plate, which typi-
cally weighs about
kg, extremely low power levels
are needed. Because GLMs require such low power and can
be interrogated by a directed laser beam, modular communi-
cations techniques would probably be used. One retroreflector
could be placed next to each sensor, and the scanning laser
would read the data from the sensor nearest the laser spot.
However, this method requires a directed line-of-sight view of
the wafer so that a laser can be focused on the retroreflector.
As with optical communications, this makes the design very
equipment-dependent.
C. Isolation
Because most of the processing techniques used in the semi-
conductor manufacturing industry place the wafers in “harsh”

Page 4

258IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 14, NO. 3, AUGUST 2001
Fig. 2.MEMS grating light modulator SEM photograph.
environments, any sensor that will be processed by the equip-
mentneedstohavesometypeofisolationfromtheenvironment.
Themainconditionsthatwouldbe detrimentaltoasensorwafer
arehightemperatureandelectricalnoise.Alsoofimportanceare
chemical attack and physical damage (such as etch damage).
In rapid thermal processing (RTP), for example, the temper-
ature typically exceeds 1000 C. Any electronics on the wafer
that are not isolated will stop functioning above about 150 C,
and will physically melt above 660 C (the aluminum inter-
connect melting temperature). Therefore, any sensor wafer that
might operate inthisenvironment must be thermallyshielded so
that the electronics remain at a lower temperature.
In plasma environments, the plasma is typically created by
coupling radio-frequency (RF) power into a gas. Because of
this high-powerRF energy, surface currents are generated in ex-
posed, unshielded conductors on the wafer surface. Therefore,
if electronics are functioning on the wafer (as in the case of a
sensor-wafer), then these generated currents might interfere. So
a sensor wafer taking measurements inside a plasma chamber
must be electrically isolated from the plasma environment to
function properly.
One possible option to isolate the electronics and sensors
from electrical noise is to add a metal layer over all of the elec-
tronics (but isolated by an oxide), and have a contact from this
overcoat to the substrate. In this way, the top metal layer would
shield the electronics below by providing a ground path. The
main method for offering shielding from chemical and physical
attack is through the use of some type of overcoat. By covering
the electronics, and possibly the sensors, with an oxide layer,
for example, most chemical and physical processes would
only attack the oxide and not the underlying electronics. When
used in conjunction with the metal layer described above, this
method provides isolation from electronic, chemical, and phys-
ical problems. To isolate electronics from high temperatures,
more complex measures need to be taken. One method for
providing this isolation is to create a microfabricated “vacuum
chamber” around the electronics, so that thermal conduction
and convection are virtually eliminated [17]. Then, by adding
appropriate coatings to the walls of the chamber, radiation
heat transfer can be minimized. The process to achieve an
evacuated chamber around electronics is somewhat complex,
but Klaassen [18] has successfully fabricated thermally isolated
RMS power sensors for RF applications.
III. EXPERIMENTAL RESULTS
We have designed and fabricated several sensor wafers to as-
sess the viability of our ideas. To better organize this research,
the project was divided into two halves. One half investigates
the power, communication, and isolation issues, and the other
half researches novel sensor structures. In this way, the project
complexity can be reduced to a manageable level. For example,
in researching sensors, wired power and communications are
utilized to eliminate those concerns.
A. Power, Communications, and Isolation
One of the intended uses for this type of sensor is the devel-
opment of next generation processing equipment, which typi-
cally uses larger wafers than the currently available technology.
Therefore,ifthesensorwafersaretobefullyfabricatedusingan
available process, large sensor wafers cannot be made because
the processes needed to do so are still in development. If, in-
stead, only one or two low-resolution metal lines are patterned
onto the wafer (the “base wafer”), and these lines serve to inter-
connect electronic modules that are mounted to the surface of
the wafer, then the wafer will be much less expensive, and more
feasible to construct. This is the technique we have chosen to
use for our power, communications, and isolation test bed (the
“component-based” wafer).
Our first component-based wafers have been temperature-
sensor wafers, for use on both bakeplates and in plasma-etch
equipment. In the post-exposure bake (PEB) process for deep-
ultraviolet (DUV) lithography, the thermal transients that the
waferexperienceswhenitisdroppedontoandremovedfromthe
bakeplate are critical to the accurate reproduction of features on
the wafer [19]. Currently, there is no accurate method for mea-
suring these transients, since the sensors must be wireless to ac-
curately reflect the transients experienced by the wafer during
robotic handling. In the plasma-etching process, thermal gradi-
entsacrossthewafer arecriticalto theetchuniformity [20].The
chemicalnatureoftheetch-processmakesitextremelysensitive
to temperature as modeled by Arhennius relationships. There-
fore, accurate spatially and temporally resolved thermal mea-
surements are essential to characterizing and controlling spatial
etch nonuniformity.
1) Temperature Sensor—Design I: For this design, we de-
cided to employ a battery-based power source, optically based
communications, and no isolation. This design was targeted for
use on a bakeplate up to 150 C. For this reason, no physical
or electrical isolation is necessary, and only thermal protection
needs to be considered.
The first prototype wafer we built involves a simple resis-
tance-to-frequency conversion circuit. A 555 timer chip is used
to generate a pulse train input to a visible LED, and the fre-
quency of the pulse train is determined by the resistance of
a thermistor mounted on the wafer. In this way, the external
flash-frequency can be used to deduce the wafer temperature.

Page 5

FREED et al.: AUTONOMOUS ON-WAFER SENSORS FOR PROCESS MODELING, DIAGNOSIS, AND CONTROL259
Fig. 3.First temperature-sensor wafer design.
An alkaline watch-cell battery provides the power source for
this wafer. A photo of this design is shown in Fig. 3.
The temperature data from this particular sensor wafer was
not very accurate (within 5 C). It was adequate, however, to
demonstrate the viability of battery-based power sources and
optical communications. Due to the amount of visible light in-
terference, it was decided that modulated infrared (Ir) commu-
nications were necessary for subsequent designs.
2) Temperature Sensor—Design II: The next design moved
to a microprocessor-based scheme. Instead of directly modu-
lating the information to the LED, a microprocessor with an in-
tegrated 8-bit analog-to-digital (A/D) converter is used to ma-
nipulate the data. It then uses a standard IrDA communications
protocol to transmit four bytes of temperature data (one byte
per sensor) and 23 bytes of IrDA overhead at 9600 baud to an
external detector. It makes two such transmissions per second.
The IrDA protocol is compatible with the PalmPilot1, which al-
lowed us to write an application to directly display the data in
real time. Again, thermistors are used as temperature sensors,
and the temperature readings from this sensor wafer are more
accurate (within 3 C). All of the components are mounted to
thewafer using silverpaint,which is a liquid paintthat becomes
electrically conductive when it dries. The silver paint was used
because ordinary solder does not adhere to the aluminum inter-
connections on the wafer. This design is shown in Fig. 4.
BecauseoftheimpedancerequirementsoftheA/Dconverter,
this design required an external amplifier to match the ther-
mistor impedance to the A/D converter input impedance. Also,
because of the thermal characteristics of the thermistors used,
the usable temperature range of the sensor system was limited.
For these reasons, we wanted to use a different type of temper-
ature sensor in subsequent designs. For convenience, we also
wanted to use a rechargable battery.
3) Temperature Sensor—Design III: The last temperature-
sensorwaferwedesignedimprovedonourseconddesigninsev-
eral respects. This wafer uses a voltage regulator chip to main-
tain constant supply voltage, and two lithium rechargable bat-
teries to extend its working life. The wafer uses temperature
1PalmPilot is a trademark of the Palm, Inc.
Fig. 4.Second temperature-sensor wafer design.
Fig. 5.Third temperature-sensor wafer design.
sensor modules, which are surface-mounted chips that output
a voltage proportional to temperature. In this way, the analog
amplifier can be removed, and the sensor outputs can be fed di-
rectlyintothe8-bitA/Dconverter.Onthiswafer,themetallayer
consists of an aluminum-nickel sandwich. Because solder ad-
heres to nickel, the electrical connections were made by placing
solder paste on the pins and dropping the wafer onto a heated
bakeplate for 1 minute. In this way, all connections were made
in parallel, and they are much more mechanically and electri-
cally reliable than the silver paint connections used previously.
This design is shown in Fig. 5.
This wafer is able to last about 11 h on a full battery charge,
and is able to give accurate temperature readings (within 1 C)
up to 120
C. The temperature-limiting components of the
wafer are the batteries, and by removing these and using
external power, temperatures up to 150 C can be measured.
A sample of this data for a multiple temperature cycle exper-
iment is shown in Fig. 6. It is evident from this figure that
the sensor tracks the temperature recorded by the bakeplate
instrumentation well, is able to function up to at least 120
C, and survives multiple temperature cycles with no adverse
effects. To examine the transient response of the sensor wafer, it