Improving
spin-on-glass processes by using pump and filtration technology
Abhay
Ramrao Deshmukh, National Semiconductor
The
planarization of interlevel dielectric (ILD) is critical to the success
of multilevel metallization processes. In multilevel metallization,
patterned conductors and dielectric layers at one level can create
severe topography problems for the layers at the next level. Topography
problems can produce opens in the interconnects or voids in the dielectric
layers because of step coverage issues associated with deposited dielectrics.
Topography-related lithography issues can also cause shorts between
adjacent lines. The purpose of planarizing multilevel metallization
layers is to eliminate or smooth out the topographic undulations caused
by conductors, dielectrics, contacts, and vias. Planarization ensures
that metal, at any level, is deposited on a flat surface, preventing
step coverage issues.
ILD
layers are primarily planarized with spin-on glass (SOG) to avoid
processing problems such as poor step coverage of deposited metal
or void formation in deposited dielectric films. However, the use
of SOG is not without drawbacks. Gel or SiO2
particles can easily form in the SOG solution. Even in sealed bottles,
SOG has a finite shelf life because of the danger of particle formation.
Extended sit times, even at room temperatures, can lead to high particle
levels in SOG materials. Although SOG is shipped from manufacturer
sites with very low particle counts, shipping, material storage and
handling, and manufacturing processes can affect the material. Particulation
becomes even more pronounced after an SOG bottle is opened, since
solvent evaporation enhances particle formation. A more troublesome
source of particle generation is that excess SOG spun onto the walls
of the spinner bowl dries quickly and forms particles.
Ever-shrinking
device geometries mandate increasingly stringent defect-reduction
goals to decrease the sizes and numbers of particles, pinholes, and
microbubbles created during the manufacturing process. Consequently,
improvements in the quality of semiconductor process materials, such
as SOG, are necessary. This article discusses tests involving the
use of a pump and point-of-use filtration system to filter out the
large particles generated during SOG processing. Designed experiments
were carried out to understand pump characteristics, process parameters,
and process outputs. Data from those tests were compared with data
collected from a pressurized-bottle system, the old method used for
dispensing SOG.
SOG
Pump and Microfiltration Technology
Tests
to optimize SOG dispense used a Model 450 system from Integrated Designs
(Carrollton, TX), which dispenses chemicals in very accurate and repeatable
amounts using patented pressure-on-demand technology. The unit's all-Teflon
flow path is compatible with a variety of chemicals used in the semiconductor
industry. Comprehensive software drives the system's sophisticated
electronics.
A
user-friendly system, the pump provides the operator with complete
control over the dispense function. It can be integrated into most
commercially available process equipment. Additionally, it can dispense
fluids at varying rates within a single dispense cycle. The unit operates
with or without filtration in the chemical reservoir, depending on
the chemicals used and process demands.
The
Dispense Function. Under operating conditions, the pump maintains
vacuum, atmosphere, and pressure conditions in a programmed idle state
until the dispense unit receives a trigger signal from the process
equipment requesting that the dispense function begin.
At
that point, the unit activates a pulse-width-modulated (PWM) pressure-regulation
mechanism, which applies pressure to the system reservoir. PWM pressure
regulation consists of a pressure-application solenoid valve (N2
on) and a pressure-release or exhaust solenoid valve (N2
exh). When a dispense command is given and the reservoir is at atmospheric
pressure or vacuum, the N2 on valve opens to
bring the reservoir to the programmed dispense pressure. An output
pressure transducer monitors the reservoir, and as the reservoir approaches
the programmed pressure, the N2 on valve and
the N2 exh valves begin to open and close alternately
to maintain a stable dispense pressure.
As
the unit is being pressurized, an autovent function is enabled, allowing
the unit to purge any air trapped in the filter and return it to the
source. Changing the state of the autovent three-way valve and opening
the source valve for a brief period causes the autovent to function.
The pressure in the reservoir pushes air in the filter to the autovent
valve through an internal vent tube. The autovent function terminates
when either the vent time or the dispense delay time expires.
Once
the unit has reached the programmed dispense pressure and the dispense
delay timer has expired, the output valve opens and dispense occurs.
By manipulating time and pressure, the operator can achieve accurate
dispense volumes. The pump dispenses chemicals for the programmed
dispense time unless the trigger signal is disengaged for any reason,
causing the dispense function to terminate.
The
pressure-on-demand pump uses helium pressure to push chemicals through
the system and onto the wafer. The helium blanket is crucial in SOG
coating processes to prevent crystallization.
During
dispense, the unit's output pressure transducer monitors the programmed
dispense pressure. Feedback from the transducer is used to adjust
the pressure and regulate the chemical output. The difference between
the applied pressure and the desired output pressure is the differential
pressure. Differential pressure is monitored during each dispense
cycle and is used by the unit to monitor chemical filter loading.
A user-programmable differential limit function sends the user a warning
when the filter must be replaced. Constant monitoring of the output
pressure transducer during dispense allows the pump to compensate
for filter loading, ensuring that dispense remains constant throughout
the life of the filter.
End
of Dispense and the Fill Cycle. At the end of the programmed dispense
time, the unit terminates the dispense function by turning off the
dispense solenoid, closing the output valve. If the unit detects a
low liquid level during the dispense operation, it ignores the signal
until dispense is complete.
At
the end of the dispense cycle, the unit either returns to its programmed
idle state or begins the fill cycle. The fill cycle is initiated when
an ultrasonic liquid-level sensor registers that the unit's chemical
level has dropped below a certain point. The unit then shifts to the
programmed vacuum state and opens the source valve to allow the chemistry
to flow into the reservoir. The fill cycle continues until the sensor
detects that the unit is full (at which time the unit returns to the
programmed idle state) or until the process equipment sends out another
dispense command.
Pump
Features. For the purposes of the tests described in this article,
the pump was analyzed from a day-to-day production point of view.
As with the implementation of any technology, a primary concern was
the unit's production friendliness. From that vantage point, its operating
ease as well as pump interlocks and alarms were evaluated.
|
Alarm
or Warning
|
Condition
|
| Chemical
source empty |
Source
cannot draw any more SOG |
| High
differential pressure |
Filter
clogged |
| Low
dispense pressure |
Low
helium pressure |
| Overfill |
Sensor
was disconnectedno signal |
|
| Table I: Dispense pump alarms
and warnings. |
First,
the pump is easy to install and its recipes can be modified to accommodate
the addition of a 1.5-second pump-delay feature to optimize dispense
accuracy. Second, it is equipped with alarms and warning lights to
indicate potential or actual problems. In contrast to the old pressurized-bottle
system, which is only capable of sending alarm signals when the chemical
source is empty, the pump offers a variety of alarms and warnings,
as listed in Table I.
 |
| Figure 1: Schematic diagram of
the dispense pump showing all elements of the system, from chemical
input to output. |
Another
feature of the system is that it offers true point-of-use filtration,
which is achieved by an in-line filter. The schematic diagram of the
system presented in Figure 1 shows all the elements of the pump, from
chemical input to output.
Table
II demonstrates the superiority of the pump and filtration technology
over the old pressurized-bottle technology. The most important features
of the pump are its versatility and ability to control the dispense
process through Chemnet software, which allows the user to customize
a wide range of dispense functions.
Testing
the Pump
SOG
Material. SOG represents a family of materials that consist of
a silicon-oxygen backbone to which different side groups are attached.
Inorganic side groups are called silicates while organic ones are
called siloxanes.
The
experiments described here used Accuglass 211 spin-on glass manufactured
by Honeywell Electronic Materials (Sunnyvale, CA). Part of the Accuglass
T-11 series of SOG materials, Accuglass 211 polymer combines organic
groups (11% CH3 by weight) on an inorganic polymer
backbone (Si-O-Si). It has better dielectric properties, better crack
resistance, and lower shrinkage on cure than silicate-type SOGs. Used
primarily for metal partial-etch-back applications, the material has
an organic content of 10%, a shelf life of 8 months at 4°C, a
dielectric constant of 3.8 at 1 MHz, and gap fill capability of >0.3
µm. Accuglass 211's high thermal stability makes it compatible
with aluminum and tungsten plug processing. It also provides good
adhesion for top and bottom dielectric layers.
Experimental
Details. Designed experiments were carried out to understand the
effects of microfiltration on particle defect levels. Pump parameters
such as dispense pressure, time, and speed were optimized to match
the output of the process of record. Particle counts, SOG thickness,
and film uniformity were the primary output parameters.
Two
coats of SOG materials were applied to wafers using an 817 track from
Semix (Fremont, CA) and then baked at different temperatures for 30
seconds per temperature. A Surfscan 6200 0.30-µm wafer-inspection
tool from KLA-Tencor (San Jose) was used to detect and identify defects.
Film thickness was measured with a Prometrix analysis tool, also from
KLA-Tencor, using 49 measurement points to determine across-the-wafer
thickness variations. Wafers were cross-sectioned to compare the degree
of planarization.
That
procedure was repeated until 20 bottles of SOG material had been consumed.
Particle variation results were compared on a wafer-to-wafer, tool-to-tool,
and SOG batch-to-batch basis. Data were collected when the bottle
was full, half-empty, and empty.
Test
Results
To
measure the effectiveness of the microfiltration technology, particle
performance, the degree of planarization, and SOG thickness and uniformity
were analyzed.
 |
| Figure 2: Data showing a 50%
reduction in average particle counts and substantially improved
wafer-to-wafer variation after the introduction of the pump and
filtration system. |
Particle
Performance. The primary objective of the project discussed here
was to improve particle performance, one of the principal difficulties
in SOG coating technology. As shown in Figure 2, average particle
counts decreased by 50% after the introduction of the pump and filtration
system, and wafer-to-wafer variation was reduced substantially, indicating
a stable and sustainable process. A similar reduction in particle-count
levels was observed across different tools and SOG batches.
|
Pressurized-Bottle
Technology
|
Pump
and Filter Technology
|
|
Uses pressurized helium bubbles to dispense SOG.
Has no in-line filtration.
Causes low SOG pressure problems because of pressure
fluctuations.
Has high maintenance cost.
|
Uses helium pressure on demand to dispense SOG accurately.
Has 0.04-µm in-line filtration.
Offers differential pressure monitoringto signal
need for filter replacement.
Has no moving mechanical parts andrequires low
maintenance.
Has optical level sensors and liquid reservoir
for better control.
SOG is kept under helium blanket, limiting the
chances of crystallization.
|
|
| Table II: Comparison between
the pressurized-bottle system and the pump/filter technology. |
Degree
of Planarization. While considering the effects of microfiltration
on SOG coating processes, the researchers evaluated its effects on
the degree of planarization, since planarization is an important SOG
process output characteristic. Although it was known that the pump
and filter would not have any direct impact on the degree of planarization,
that characteristic was investigated for the sake of experimental
thoroughness.
 |
| Figure 3: Gap filling and global
planarization. The left side of the chart shows metal lines with
steep wiring spaces and small gaps that should be filled, while
the right side shows metal lines with gradually sloping transitions. |
The
principal objective of the planarization process is to render the
surface topography of the dielectric as smooth as possible. In other
words, the surface should be free of small gaps, and transitions should
be gradual and sloping rather than abrupt and vertical. Figure 3 explains
the concepts of gap filling and global planarization. The left side
of the figure shows metal lines with very steep wiring spaces and
small gaps that should be filled, while the right side shows metal1,
metal2, and metal3 lines with gradually sloping transitions. Proper
planarization must accomplish both objectives or it must be supplemented
with other methods. SOG generally produces short-range planarizations,
which are often adequate for double-level metal devices.
 |
| Figure 4: Comparison between
the gap fill and global planarization performance of (a) the old
pressurized-bottle technology, and (b) the pump and filter system. |
In
addition to evaluating the pump and filter system's effects on the
degree of planarization, the researchers also compared the gap fill
and global planarization performance of the old pressurized-bottle
technology with the new pump technology. The images in Figures 4a
and 4b illustrate that there is no performance difference between
the two technologies. The images are examples of gap filling in a
double-level metallization structure, where Metal1 and Metal2 layers
overlap a D2 base, D2 cap, and SOG pocket. That device region was
selected for cross-sectioning because it is critical and because planarization
problems, especially via poisoning from excess SOG, or metal bridging,
are easy to identify on it.
 |
| Figure 5: Data showing that SOG
thickness increased, and wafer-to-wafer variation decreased by
25% after the pump and filter system was installed. |
SOG
Thickness and Uniformity Results. The last output parameter to
be investigated was as-deposited SOG thickness and uniformity, as
measured by the Prometrix analysis tool. For the sake of experimental
accuracy, data were collected from several tools and SOG batches over
an extended period of time. As shown in Figure 5, a drastic improvement
in thickness variation was observed after the pump was installed.
Because the pump can dispense SOG materials consistently and accurately,
wafer-to-wafer thickness variation was reduced by 25%. In addition,
as-deposited SOG thickness nonuniformity was reduced by 40%. In further
tests, similar results were obtained from wafers processed using recipes
involving different SOG thicknesses.
Conclusion
The
SOG coating process has high housekeeping needs because of its propensity
to form particles. The process requires that fab personnel perform
frequent chamber cleans, cup changes, exhaust-line cleans, and general
maintenance on coating tools.
Using
a pump and microfilter technology, the investigators studied the effects
of true point-of-use filtration during SOG processing on highly critical
output parameters such as particles, film thickness, and film-thickness
uniformity. Process parameters such as dispense pressure, time, and
speed were optimized to reduce defect levels. The beneficial pump
and filter system reduced average particle defects by 50%, as-deposited
SOG thickness nonuniformity by 40%, and wafer-to-wafer thickness variation
by 25%.
The
technology resulted in consistent, reliable process outputs for all
measurable process output parameters. Point-of-use filtration improved
tool cleanliness and reduced the frequency of chamber and exhaust
cleans, increasing tool uptime.
Acknowledgments
The
author wishes to thank Ronald LaPosa and Florizel Dharmaratnam of
National Semiconductor (Arlington, TX) and Brian Kidd of Integrated
Designs (Carrollton, TX) for their invaluable assistance with this
project.
Abhay
Ramrao Deshmukh is a process engineer at National Semiconductor
in Arlington, TX, where he specializes in interconnect technology.
He has worked at the company for three years in the areas of PVD systems
and SOG planarization. He received an MS in materials sciences from
the South Dakota School of Mines and Technology in Rapid City. (Deshmukh
can be reached at 817/557-7602 or abhay.r.deshmukh@nsc.com.)