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

net/publication/351022607

CONTROLLING AND MONITORING SEMICONDUCTOR MANUFACTURING

PROCESS USING SMART TECHNOLOGIES

Conference Paper · April 2020

CITATION READS

1 859

1 author:

Rammohan Vadavalasa
Technical University of vienna
5 PUBLICATIONS 2 CITATIONS

SEE PROFILE

All content following this page was uploaded by Rammohan Vadavalasa on 21 April 2021.

The user has requested enhancement of the downloaded file.

 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com

CONTROLLING AND MONITORING SEMICONDUCTOR MANUFACTURING

PROCESS USING SMART TECHNOLOGIES
Rammohan Vadavalasa*1
*1
IT Professional

ABSTRACT
Manufacturing semiconductor is one of the most important and technologically advanced industrial sectors.
Controlling and monitoring the semiconductor manufacturing process drastically impact overall production
performance because it reduces the gap between the manufacturing process and the supply chain. Smart
technologies and methodologies have proliferated in the semiconductor manufacturing process in order to
improve productivity and decrease maintenance costs.
The immense, time-sensitive, and diversified nature of the data brings significant challenges to the real-time
collection, processing, and decision making.
Purpose of the paper:
The main goal of this work is to provide information about the usage of statistical modeling, IoT, sensors, big
data, and control systems used in semiconductor manufacturing with the intention of improving production
efficiency, control process defects, and reducing machine downtime. Finally, presented a framework of an
industrial data management system that is able to process huge industrial data, support monitoring, and control
in the manufacturing process.

I. INTRODUCTION
The semiconductor played a huge role in the history of mankind because of its effect on accelerating technology
and advancing the world's economy in the 20th century.Gian et al.(2012) in their work emphasize and addressed
issues and challenges in automatic Control and Machine Learning for Semiconductor Manufacturing: Review
and Challenges [[2]]. When the semiconductor was born in the Bell Labs in 1948, its effects are largely
unexpected. A megabit of memory cost at that time is $75 Million, but now it is worth around 15 cents.
Semiconductors together with VLSI and ULSI brought huge changes in the electronic industry so far.
Because of the increasing ability to store and process information, it has become an extremely important
operator of social, scientific, economic, and legal processes. Semiconductors play a major role in every
electronic device, for example, personal computers, mobile phones, refrigerators, and so on.
Most of the semiconductor manufacturing companies give high priority to improve quality and computation
capability in the manufacturing process. Semiconductor companies are investing a lot of resources because it
would be possible to improve manufacturing capabilities using statistical prediction and control systems to
decrease defective products and increase process stability. Transmitting real-time data across the industrial
network IoT creating manufacturing intelligence.
There are a lot of observable parameters such as historical time series, sensor data, a characteristic behavior of
the equipment, using all of these parameters possible to predict the future status of the equipment, and making
strategic decisions about maintenance and production.
Produced an immense amount of industrial data used for prognosis, prediction, real-time control, and monitoring
manufacturing activities. This data helps organizations for failure identification and prevention, predicting
machine health status, and making smarter business decisions.
The future of electronic manufacturing industry depends on smart manufacturing because it improves
production efficiency, safety and security using emerging technologies such as industry4.0, IoT, effective

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[147]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
sensors, and artificial intelligence enables real-time analytics for the predictive process; machine control and
visualization in the manufacturing process.

II. RELATED WORK

Smart manufacturing increases economy, safety, gives complete control on manufacturing process; supports
reducing global warming; suitable for rapid change in the complex business process; handling increasing factory
intricateness; intended to streamline reliability, productivity, capability, flexibility and cost requirements in the
manufacturing process; chain with global market demands using advanced technologies like industrial internet
of things, big data solutions, industry 4.0 and artificial intelligence Methodologies.
Many efforts have been made for the IoT data acquisition, preparation, processing, and storage in smart
manufacturing. Industrial internet of things connects billions of devices intended to revolutionize the
manufacturing industry, increase process efficiency, reliability, and safety. Smart manufacturing increases
product quality, time to market, increases customer satisfaction and enhances product market size.
In the worldwide manufacturing process, industrial data [[27]] plays a crucial role intended to generate huge
wealth and create new job opportunities in developing as well as developed countries. Industrial data are
complex and unstructured, managing and storing this data requires [[37]] high processing power. IoT, sensors,
Industry 4.0, and predictive models help to build an efficient production system in smart manufacturing.
Manufacturing plays an important role in developing human living standards, and it is a crucial part of economic
growth in every sector in this globalization era.
Emerging technologies are used by the manufacturing industry to magnify the economic competitiveness of
individual manufacturers and the sustainability of the entire industrial sector using smart technologies as shown
in figure 1.

Fig-1: Using smart technologies in semiconductor manufacturing

Emerging technologies bring significant improvements in the manufacturing process and influence reducing
energy consumption, improving production ability, diminishing social risks, and influencing decision making.
Big data technologies reduce the processing time, provides effective solutions from generating huge data
volumes in real-time. Technologies like Apache Kafka, apache storm and MongoDB add significant benefits to
big data processing and storage in semiconductor manufacturing.
IoT based sensors add significant advantages to collect data for big data processing.
Generated data from manufacturing industries were analyzed using machine learning (example fault detection,
quality prediction, defect classification, and run to run control) technologies to make suitable decisions in the
www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[148]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
company. The growth of big data in recent trends reduces the gaps in technology and contributes to new
technological trends for the future.
Higher volumes of data generation will impact positively on developing more robust real-time solutions in the
industry. In semiconductors, the connecting architecture enhances the smart industry, increases communication
between devices, and real-time monitoring and controlling.
This paper is organized as:
Section 3. Semiconductor device fabrication process.
Section 4. Using smart technologies in the semiconductor manufacturing process.
Section 5. Presented an “Architecture for monitoring and controlling the end to end semiconductor
manufacturing process in industries.“ that utilizes the above-mentioned technologies such as IoT based sensors,
big data, and machine learning models to control and monitor semiconductor manufacturing.
Section 6. Challenges and future work are presented.

III. SEMICONDUCTOR DEVICE FABRICATION PROCESS

The process used to manufacture semiconductor devices is called Semiconductor device fabrication as shown in
figure 2. Usually, the semiconductor manufacturing process takes 8 weeks of time. In the manufacturing process,
integrated circuits with components such as transistors are formed on the surface of a silicon crystal wafer.
In integrated circuit formation, a thin film layer will form the wiring, the photoresist is coated with this thin film;
as a mask for etching to process the thin film into the shape of the wiring and other components, this photoresist
is used to shape it. This is the first layer of the circuit formation, later transistors are going to form on the lowest
layer, and this process is repeated until many different layers of circuits are formed on top of one another.

Fig-2: Semiconductor device fabrication process

In this process of creation of integrated circuits, there are several examinations and measurements to check
whether or not the patterns are fabricated as designed or not.
If in case any defects are found in the fabrication process, the process will be interrupted to remove and correct
the errors in the fabrication operation. On a single wafer, more than one hundred semiconductors dies are
fabricated.
Semiconductor device manufacturing contains wafer processing operation, it is called the front-end process, and
the assembly process is called the backend process.
In a wafer processing operation, there contains several steps, Ultra large-scale integration (ULSI) follows strict
requirements in a controlled environment (example:- dust, vapors, and pollution-free). Initially, wafers are
cleaned, later dielectric and polysilicon are used for fabricating integrated circuits. The main deposition

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[149]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
techniques used here are chemical vapor deposition and physical vapor deposition. Next photo mask exposition
is used to create ULSI circuits on wafers, and the etching process will be conducted for the desired patterns.
This process repeats several times for multiple interconnected layer formations on the wafer surface. In order to
send the wafer for chip preparation, every single IC on the wafer is tested for functional defects.
In the assembly process, packaging provides an electrical connection in order to protect the chip from
environmental and mechanical stress.
In conclusion semiconductor device fabrication, the different clarification ways fall into 4 general classifications
such as deposition, removal, modeling and modification of electrical properties and it is achievable using IOT,
Big data, sensors, machine learning, these automatic control techniques bring huge changes in the production
quality and manufacturing process.

IV. USING SMART TECHNOLOGIES IN SEMICONDUCTOR MANUFACTURING PROCESS

4.1 Internet of things(IOT):
The internet of things or the internet of everything is a global application where devices are connected together
with sensors or actuators. IoT plays a significant role in semiconductor manufacturing because IoT enables
things to connect, process, and transmit the signals in the production process.
Integration of systems, linking of physical and cyber capabilities, gateways, Bigdata, helps improvement in
smart semiconductor manufacturing. IoT helps semiconductor companies not only manufacturing but also
includes high-level business transformation like extending sales processes, cloud delivery, and new forms of
customer support.
It is critical for the semiconductor fabrication process to maintain high self-confidence because it requires huge
amounts of capital investments, and it contains more than 150 long production cycle times and requires huge
raw materials, so it suffers from the critical process and testing times and hard to implement process control in
production. But the smart manufacturing process helps sensors to record yields and use predictive analytical
models for improving the production process.
Next-generation technologies like 5G (it transfers data 100 – 200 times faster than 4G LTE), industry 4.0, The
internet of things and artificial intelligence tend to automate complete data exchange in the manufacturing
process for fast, smart, accurate manufacturing, and these technologies blurring the lines between the physical
and digital world.
Reinsel et al.(2018) in their work emphasize and addressed issues and challenges in the digitalization of the
world from edge to core [[34]] The global data will increase 10X to 163 zettabytes or one trillion
gigabytes by 2025, and it adds a lot of values to the IoT and artificial intelligence infrastructure.
Now, 5G brings huge amounts of the industrial revolution, and this revolution requires a huge amount of
semiconductor chips to manufacture sensors, light-emitting diodes, flexible display technologies, and so on. For
example, automobile industries require 50% more semiconductors for autonomous vehicles compared to current
vehicles because they are intended to provide more automated, cleaner, connected, and secure services, and it
demands increasing production in the semiconductor industry.
In conclusion, IoT enabling real-time decision making not just industrial manufacturing but also data to
accelerate decision making in product marketing and management.
4.2 Sensors used in different stages of manufacturing:
In semiconductor manufacturing, statistical process control has been used since 1960 within the prescribed
specification limits. This statistical process control tracks certain unique and individual metrics, and it intimates
to the user when the limits exceed with a specified statistical significance. This approach is adding value to the
process, but it delays the problem recognization and rate of processing of wafers because it concentrates on only
some specific parameters.

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[150]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
In semiconductor manufacturing since 1990, the problems associated with decreasing feature size and increasing
wafer size have added problems in the manufacturing process, so in order to stay competitive in the
manufacturing field, both these constraints need to be removed. In manufacturing processing anomalies need to
be detected in shorter timeframes. Advanced process control is the current paradigm that strives to solve these
problems in the manufacturing process.
It contains several methodologies to solve these challenges in the process. Compared to statistical process
control, advanced process control is a closed-loop, more relevant, and contains a more stable and flexible
architecture for the manufacturing process. This advanced process control system uses methodologies, sensors
to process data for better performance. The national technology roadmap for semiconductors (NTRS) in 1997
enabled this mode of operation using sensors to process wafers for more stability, reliability, and reproducibility
in production.
Doering et al. in their work emphasize and addressed issues and challenges in their handbook of
semiconductor manufacturing technology[[9]]. For semiconductor manufacturing, the process requires a
different type of sensors and these sensors should be reliable and ease of integration into the processing.
Sensors for keeping track of both the process state of a specific tool and the wafer state of the processed wafer.
Here wafer state processing control is the main parameter to monitor and control the effective manufacturing
processes. For controlling and measuring wafer temperature contains different processing techniques such as
chemical vapor deposition (CVD), rapid thermal processing (RTP), epitaxial (EPI), physical vapor deposition
(PVD) are used for annealing and film growth.
For film thickness, gas flow, and pressure control different types of sensors are used intended for controlling and
monitoring the manufacturing process. Different type of sensors required for the processing in semiconductor
manufacturing, such sensors[[9]] are:
Scatterometer: It is used[[1]] in metrology for the roughness of polished and lapped surface detection. It
provides a quick and non-contact alternative to traditional methods for topography computations.
Fourier transform infrared (FTIR) spectrometer: It is being used for chemical kinetic relationships involved
in the semiconductor manufacturing process. Mostly used to obtain an infrared spectrum of absorption or
emission of a gas, liquid, or solid, and it is able to collect high spectral resolution data above a wide spectral
range.
RF sensor: Radiofrequency sensor is used to measure radio frequency signals in an electronic circuit.
Residual gas analyzer (RGA): To check the leakage and test contamination or moisture. In the semiconductor
manufacturing process, it is used for establishing a correlation between the wafer's quality and measured
contamination in the process chamber.
Noncontact IR thermometers: It is used to measure the temperature under certain circumstances where
thermocouple not able to use because thermocouple is not able to produce accurate data.
Interferometer: Measuring certain characteristics formed by waves (example light, radio, or sound waves) or
of materials that reflect, refract, or transmit waves and it can also be used to make precise measurements of
distance. Mostly used in the run to run control process in semiconductor manufacturing.
Image sensor: Detecting pattern defects and minute foreign particles generated in the process because with the
image sensor high sensitivity and the high-resolution image is possible to detect.
Leak detectors: Capturing atmospheric pressure that is sensitive to sound in the semiconductor manufacturing,
it is used to check the tightness of vacuum chambers for wafer production.
Spectrometer: Measuring the emission spectrum from a plasma chamber and for diagnosis of the
manufacturing process and analysis of materials in optical emission spectrometer (OES) and endpoint detection
(EPD). Er production.

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[151]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
Ellipsometry: Used for monitoring and measurement of changes in the thickness of thin films and optical
properties in the manufacturing process.
Quartz Crystal Microbalance (QCM): QCM is an extremely sensitive mass balance that is able to measure
nanogram to microgram level transformations in mass per unit area.
Humidity sensor: Used to monitor and measure moisture and air temperature in semiconductor manufacturing
because the relative humidity is an important factor for the production process.
Pyrometry: It is a remote sensing thermometer used to measure the temperature of distant objects in
manufacturing.
Spectrophotometers: Used to measure the transmission of light through a liquid sample and used to analyze
and measure light intensity between the sample and a reference sample in the semiconductor manufacturing
process.
Diffuse reflectance spectroscopy: Measuring the flux per wavelength of light reflected in a scattered manner
from a sample in the process.
In semiconductor manufacturing, when device features become smaller, require tighter control of thermal
conditions, they are necessary for a successful and risk-free device fabrication process.
In conclusion, the Indutry4.0 concept is unchanged for a long time. The above sensors are just part of smart
technologies what some sensors can provide to the smart manufacturing and different type of factories requires
different types of sensors in order to monitor and control. In the future more sophisticated smart sensor
technologies will come to the market intended to support the smart revolution and wireless communication
intended to provide robust connections to those sensors.
In conclusion in this sub section
4.3 Big Data Processing
Big data (a large amount of data collected from different sources), data mining (discovers a hidden pattern in
large datasets) and data analytics (is a superset of Data Mining that elicits transforming, cleaning, extracting,
modeling and visualization of data) together all of them are finer levels of control and diagnostics in the
fabrication process.
Methodologies like predictive maintenance, virtual metrology, fault detection, and classification, run-to-run
control enable impact on the complete production process. As per International Technology Roadmap for
Semiconductors (ITRS) big data contains different dimensions as shown in figure 3 such as volume, veracity
(data quality) velocity, value (analytics), and variety. Data analytics support real-time decision making and
predictive analytics in the manufacturing process.

Fig-3: Different dimensions of big data

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[152]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
Different technologies support data processing such as Apache Kafka, apache storm, which has big data
processing capabilities, and no SQL databases such as MongoDB for data storage in big data.
In conclusion, big data processing techniques help for better decision making aiming to extract valuable insights
from production data in industry 4.0.
4.4 Machine learning Methodologies :
4.4.1 VIRTUAL METROLOGY
A virtual metrology system contains methods to predict the properties of a wafer based on machine parameters
and sensor data produced by the machines in the production. Mostly it deals with production quality control
during nonsampling periods. Machine learning models like Neural networks, classification, and regression tasks
are used to perform such operations. These models are able to predict production accuracy from time to time,
but when new products are added for the fabrication, the behavior of tools changes over certain maintenance
cycles, therefore the system needs to be updated constantly.
The semiconductor manufacturing process is a challenging and risky task due to:
 High dimensionality: Regression problems contain hundreds of input variables, and it adds more challenges
[[14]] in order to solve them, and it is computationally expensive.
 Extracted data from different sources required to be processed in order to reduce noise data and prepare it
for predictive model building.
 Using dimensionality reduction techniques (example correlation analysis (CA) and principal component
analysis (PCA)) in order to reduce the dimensionality of the dataset and variable section techniques
(example stepwise selection (SS)(with a minimum number of predictive variables)) is used to prepare data
for predictive modeling.
 The semiconductor manufacturing process requires sequential processing operations and all these
operations impact wafer quality measurement, but virtual metrology is able to account for a single process
at a time.
There are different modeling approaches in the semiconductor manufacturing process such as regression
(Ordinary Least Square, Partial Least Squares)(all Virtual Metrology problems can be tackled with regression
approaches), classification (K nearest neighbors) and Artificial neural networks.
Based on [[2]] neural networks guarantee better performance in modeling semiconductor manufacturing
processes compared to other modeling approaches because neural networks provide a wide degree of accuracy
and approximation.
The neural network adopted a multi-layer perception (MLP). MLP contains at least 3 layers of nodes, and each
node is a neuron. These neurons are highly interconnected processing elements, and each node in the neuron
uses a nonlinear activation function.
The central idea of MLP is to extract linear combinations of inputs and then model the target as a nonlinear
function. MLP uses a backpropagation for model training. A single layer network in a multi-Layer network is
called perception, as shown in figure4.

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[153]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com

Fig-4: Perception in neural network.

FERREIRA et al. in their work emphasize and addressed issues and challenges in virtual Metrology Models for
predicting physical measurement in semiconductor manufacturing [[16]]. Feedforward multilayer networkswith
sigmoid nonlinearities are called multilayer perception. The multilayer perception is based on input vector X
and parameter vector W (weight).
Y = X.W + b ___(1)
There is a bias added to the system in equation (1) b. The activation function most often used are liner functions
such as hyperbolic tangent function and sigmoid function. In multilayer perception using sigmoid activation
function in a single layer, neurons result in a linear output of neurons. There is a nonlinear function for given
inputs and parameters of the output of the network.

____(2)

In equation (2) X is the input of (n+1) vectors; W is the vector for (n + 1)Nc +(Nc + 1) parameters. Here hidden
neurons are from 1 to Nc and output neuron is Nc + 1. The parameter Wij conveys information from input
neuron j to i. X1, X2, X3…. X22, X23 re the input variables for the process parameters from the fault detection
and classification system (FDC). Later here the target is the thickness of the layer after Chemical vapor
deposition (CVD).
For the variable selection, process subsets selection regression technique is widely used to interpret and select
variables and stagewise selection is used for prediction accuracy in algorithm procedure. This approach allows
for continuous, smooth estimation during the assembly line process.
4.4.2 PREDICTIVE MAINTENANCE
In semiconductor manufacturing, efficient process maintenance and complete control[[2], [17]] of the process is
essential to decrease the costs connected with defective wafers and appliance inactivity. Maintenance strategies
are divided into different categories based on complexity and efficiency.

• Run to failure maintenance is performed when repairs are completed, otherwise, restoration actions are
performed after the occurrence of a failure. With increasing age, some of the equipment fails from time to
time and other manufacturing defects due to incorrect installations. In the run to failure, the conditional
probability of the failure remains low with increasing time.

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[154]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
• In preventive maintenance or scheduled maintenance strategy, it is a very challenging task in the
semiconductor manufacturing process due to its complexity of highly integrated fabrication tools and
systems, maintenance is carried out periodically on a planned schedule with the aim of anticipating the
process failures, but these failures are usually avoided.
• In condition-based maintenance, if there is any degradation in the processor in the equipment, it takes
place based on one or more conditions, and this approach is based on a continuous monitoring of the
machine's health and status.
In predictive maintenance, when facilitating the implementation of planning and scheduling schemes, some
prediction tools are used for taking required actions.

Fig-5: Predictive maintenance in semiconductor manufacturing

Using sophisticated maintenance tools such as sensor data, IoT, and machine learning tools with predictive
maintenance as shown in figure 5 can help to increase the percentage of non-defective products, increase in
system uptime, be able to increase production capability and ability to provide strategic decisions about
maintenance and scheduling. All of the predictive maintenance problems are different from each other, and each
problem should be analyzed, studied individually, and evaluated as customized solutions. There are a lot of
modeling techniques that are suitable for sophisticated maintenance.
Wavelet-Based Machine Learning Algorithm is used to identify important features for the detection of process
faults. Kalman Predictor and Particle Filter techniques are used for filtering and prediction purposes in
semiconductor manufacturing.
Predictive maintenance is dealing with maintenance issues and increasing the need to minimize downtime and
associated costs. In this scenario, K-Nearest Neighbours is the simplest approach to classify because it requires

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[155]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
computations of the distance between samples. In K-NN each point of the input space is labeled according to the
labels of the closest neighboring samples, and here use euclidean distance conditions.
In order to tuning k-NN required suitable K value for the number of samples in the considering neighborhood.
Choosing a suitable k value is a challenging task. It is possible to choose using cross-validation. The larger K
value reduces the effect of noise on the classification, but decision boundaries between classes are not clear.
There are other modeling techniques such as Ridge Regression and Elastic Nets, classification methods
( Support Vector Machine), and neural networks that are widely adopted in predictive maintenance.
4.4.3 FAULT DETECTION AND CLASSIFICATION
In the semiconductor manufacturing[[2], [22]] process, in order to reduce machine unavailability and maintain
process stability, it requires fast, accurate, and sensitive fault detection and faults classification system. Fault
detection and classification aim to identify the root cause of the abnormal behavior, but, compared to predictive
maintenance, it does not predict the probability of the future behavior of the production process.
Classification techniques such as k-Nearest Neighbours [[23]] (Principal Component Analysis) and support
vector machine is used for fault detection in manufacturing.
Using a support vector machine algorithm for wafer classification is one of the suitable processes. Classifying
low yield and high yield classes using a multidimensional hyperplane. This model is adaptively updated in a
manufacturing environment using g sing a non-linear support vector machine intended to classify wafers into
two categories with a predefined yield threshold that distinguishes between good and bad wafers. Here, the
maximum margin hyperplane (MMH) is the largest separation at the edges with good and bad clusters
depending on the kernel function and using predefined parameters in wafer measurement during inline
processing.
Baly et al. (2012) in their work emphasize and addressed issues and challenges in wafer Classification Using
Support Vector Machines [[20]]. Initially, a linear SVM model is described to help make a decision boundary
using linearly separable hyperplane called maximum marginal hyperplane (MMH).
Maximum Marginal Hyperplane (MMH) is

WT ・ X + b = 0 …. (1)

Where W = w1, w2, w3 …..wm and b represents the desired coefficients. Here X is
{X1, X2, X3 ….. Xm} and m is the number of wafer and equipment measurement during the manufacturing
process.
MMH used to measure leads to a user-specified yield threshold. Wafer measurement right of the MMH
represents below the user-specified yield threshold, and the left side represents above the user-specified
threshold value.
In order to provide the best nonlinear separating curve, the training data is transformed from the original
coordinate to higher coordinate space, this can be achieved using Kernal function such that computes the
similarity between the training samples in the transformed space using the original set of attributes.
The kernel function becomes such as

______(2)

In equation (2) K(Xi, Z) is the applied function for the kernel, and here Z is the new wafer to be classified. Xi is
the wafer measurement for the training set. There is a Gaussian RBF kernel for classifying good and bad clusters
in the algorithm in semiconductor manufacturing suitable kernel.

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[156]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
Gaussian RBF kernel is

______(3)

In equation (3) X represents a training wafer, Z represents a new wafer and σ is the kernel width parameter.
This method is used to be guided by the ability to perform better prediction and classification accuracy in the
semiconductor manufacturing process.
4.4.4 RUN-TO-RUN CONTROL
Using information acquired during the manufacturing process, ameliorating process productivity by fine-tune
process[[2]] inputs run by run are called run to run control in semiconductor manufacturing, and it is a standard
approach in process control.
Run to run control as shown in figure 6 enhances process and devices performance, boosting process throughput,
and limiting of defective wafers.
WANG et al.(2013) in their work emphasize and addressed issues and challenges in a batch-based run-to-run
process control scheme for semiconductor manufacturing [[25]]. Exponentially Weighted Moving Average
(EWMA) algorithmis used for a run to run control procedure in manufacturing.

Fig-6: Run to Run control system

Improving quality in the semiconductor manufacturing process run to run control has been widely used. In the
EMWA controller, using both input and recipe information characterizing the output of the process is as follows:

yi j = α + aTxi j + bTui + di ____(1)

where yi j here i = 1,2,3….. n and j = 1,2,3 ……M, it is jth product in batch i after processing; xi j; α(alpha) is
the intercept and ui is the recipe for batch i (here i = 1,2,3,4 ….) and here everything processed sequentially at a
time interval t= i; process distribution(di) is the white noise in the process. In the run to run control
coordination between upstream and downstream stations should take into effect based on incoming information
in the process. In order to achieve a higher level of process quality and capacity, the process recipes should be
optimized,
i.e. The mean square error (MSE) of the process output should be minimized

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[157]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com

_______(2)

Here the main objective of Run to Run control is to minimize mean square error.
Here the structure of the batch EMWA controller is proposed
A modified K-means clustering algorithm
As equation 2 there are two variables such as within batch variation and batch to batch variation. A batch
allocation algorithm is to reduce the number of batch variations.
Using process model equation 1 substitution in equation 2 (MSE) transforming within batch variations to
process the output to the within batch variation in the process input.
The minimization of the batch variation can be expressed as::

__________(3)

Here k means the algorithm only provides local minimum. Repeating an algorithm with different initial
assignments may provide a better solution in the process.
Different control and monitoring strategies and approaches are able to implement in the production process
based on physical measurements and probabilistic distributions in semiconductor manufacturing.
In conclusion, the use of machine learning in industry 4.0 fundamentally changing the manufacturing strategies
from design to product. If hardware needed for data collection and processing gets cheaper; then this
transformation not only helps big companies but also possible to transform the small and medium-sized
organizations.
Earlier most of the siloed data from the manufacturing industry go to waste because of a lack of platforms that
can truly leverage these diverse data and extract valuable insights to improve, productivity, and quality in the
manufacturing.
Smart technologies such as big data, IoT, sensors, and machine learning in manufacturing helps to achieve the
collaborative, automate, and distributed workflow deliberate to improve capacity and standard in industry4.0.

V. ARCHITECTURE FOR MONITORING AND THE CONTROLLING END TO END

SEMICONDUCTOR MANUFACTURING PROCESS IN INDUSTRIES:
Semiconductor manufacturing includes different stages such as wafer manufacturing, chip manufacturing, and
product manufacturing. Wafer manufacturing includes building electronic circuits on a wafer, chip
manufacturing contains probing and testing, in the product manufacturing stage, integrated circuit assembly and
testing involved.
Semiconductor manufacturing involves a lot of challenges, and it is a very complicated production system with
huge capital investment. It Requires advanced technologies intended to control quality, inconsistency, and
reliability. Here, automation helps to reduce investments, produce the best products, and bring the complete
process into a sequential order intended to produce faster uniform, and stable output.
Manual, semi-automated, and fully automated manufacturing process supports the semiconductor production
process. The manual mode of operation required no computer assistance in the fabrication process, and it is
www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[158]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
implemented at the beginning of semiconductor manufacturing. Semi-automated manufacturing, most of the
fabrication processing tools are automated and controlled by computers but handling materials still is handled by
fabrication operators. In the automation manufacturing process complete fabrication life cycle controlled by
computers without human intervention.
The automation process in lower-level architecture includes embedded controllers and sensors with the aim of
real-time control, monitoring, and analysis of the fabrication process. At a higher level, a combination of
metrology operations and material movements is sequenced, handled, and executed. The automation process
involves coordination, communication, control, allocation, mediation among the flow of data, and system
dynamics in a timely manner.
Smart manufacturing and industry 4.0 are required because computer integrated systems are monolithic, loosely
coupled, and difficult to support current needs. Using smart manufacturing is possible to develop a framework
intended to support distributed systems, data integrity, and able to control critical issues in the manufacturing
process.
Using six Sigma capabilities, we are able to predict if something fails in the semiconductor manufacturing
process with smart infrastructure because it provides closed-loop feedback. Smart manufacturing required
stable security for communication between manufacturing facilities because companies mostly build with new
and old infrastructure together.
Saqlain et al. (2019) in their work emphasize and addressed issues and challenges in framework of an IoT-
based Industrial Data Management for Smart Manufacturing [[11]]. This architecture for a
semiconductor industrial data management system supports huge amounts of industrial data, controlling and
monitoring the industrial manufacturing process. Sensing, communication, and actuation technologies have
enabled smart industries to acquire, store, transmit, and process data intended to control factory behavior. Data
generated from various factory devices in a distributed industrial environment is collected and converted into
useful information in order to predict, control, prognosis, and improve productivity in the production lines.
There are different control systems available for collecting a massive amount of industrial data such as
distributed control systems, supervisory control, data acquisition system, and programmable logic controllers.
The storage of data is based on using JSON and XML files. Initially, industrial data from sensors is converted
into structured data, and this structured data is stored in cloud platforms for permanent storage. There, using
different data mining and machine learning algorithms can extract useful information.
5.1 Industrial data management system for monitoring and controlling:
The industrial data management system framework supports real-time data collection, transformation,
processing, and storage. Data collected from sensors from the manufacturing process is real-time and
unstructured data.
The management system supports data processing using Apache Kafka as a message queue; real-time data
processing using apache Strom, storing sensor data using MongoDB from the production activities in the
manufacturing process. The proposed model uses different machine learning models as explained in section 4.4
"virtual metrology", "predictive maintenance", "fault detection and classification" and "run to run control".
These models analyze sensor data and provide desired information to the user.
This framework supports improving decision making, and it will avert unexpected losses caused by irregularities
during the manufacturing process.Machine learning models utilized to detect certain patterns in various
situations such as quality control, defect control, system monitoring, and visual inspection because machine
learning shows extraordinary results in the manufacturing industry.
5.2 System design and implementation:
The real-time monitoring system helps management for better control and monitoring assembly line process in
the semiconductor manufacturing industry. It uses sensors, big data, IoT, and machine learning technologies.

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[159]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
Complete machine learning model data is presented to managers in real-time using a web-based monitoring
system intended to update production status from time to time.
Apache Kafka is for message queue in this framework because it supports low latency, high throughput, fault
tolerance, and able to publish streams of data. Apache Strom is for data processing in this system for fault
tolerance, horizontal adaptability, and possible to process large volumes of data. IoT based sensors are attached
to the desk of a work station panel along the assembly line process. Here, IoT based sensors sense
environmental conditions and machine status every 5 seconds and send data to a cloud server. Sensor data
consists of the ID number of the IoT device and other details like recorded and processing time, sensor data, and
other related information.
Initially, sensor data is transmitted wirelessly using a python based program to publish in the Kafka server. The
JSON format is used to send data to the apache Kafka server. Later streams of published data from Kafka are
processed by Strom in parallel and in real-time. Data classification is implemented inside Strom, and these
classification results are stored in MongoDB. These stored results in MongoDB can be represented in a web-
based monitoring system for real-time monitoring.

Fig-7: System architecture for real-time monitoring and controlling semiconductor manufacturing process
This monitoring and control system is intended to process the semiconductor manufacturing process. It is
possible to demonstrate using raspberry pi[[35]] single mainboard and sense-hat[[36]] board. Machine learning
methods play a crucial role in a monitoring system in different conditions. Using this system, as shown in
figure 7 Possible to predict whether the process is functioning normal or abnormal based on the events detected
during the manufacturing.
Using user interface data visualization present to the user for real-time monitoring.
Users can monitor the complete manufacturing process in a timely manner and be able to receive a warning if in
case any abnormal events are detected during real-time production.
This real-time monitoring system is possible to access from remote locations via a web browser on a
tablet/computer/mobile phone. This proposed system is possible to adopt in any manufacturing industry using
IoT and sensor technologies.

VI. CHALLENGES AND FUTURE WORK

In new manufacturing companies, this approach is easily adaptable and infrastructure supports for smart
manufacturing, but in already established industries, the infrastructure already exists, and making changes in the
equipment and adopting the manufacturing process with smart technologies is a challenging, potentially

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[160]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
expensive, and riskier task. So research is required in this direction to take steps such as what technologies
required to make it work already exist in the infrastructure.
Because of the high dimensionality of data, it is hard and time-consuming in order to process and identify
suitable elements in the data and research required in this direction.
Require more research work on network delays, effective CPU, and memory usage.
In the future, it is required to study the effective security of IoT devices implementation and data collection
procedures.
Furthermore, a variety of abnormal behaviors and instability in the system should be carefully identified. This
monitoring and controlling data management system should be implemented in real-time using mentioned smart
technologies.

VII. CONCLUSION
This paper presented a complete conceptual approach to smart technologies Usage in the semiconductor
manufacturing process is intended to control and monitor production operations.
A complete overview of semiconductor manufacturing using different advanced technologies has been presented.
Smart manufacturing in the industrial sector handled a variety of functions.
In this study presented different smart technologies such as IoT based sensors, Big data processing, predictive
machine learning models for semiconductor manufacturing.
Finally, I proposed a data monitoring and controlling system for effective and efficient processing and analyzing
data in the manufacturing process.
This system controls the assembly line process and is able to identify faults in the process, machine status, and
able to track the complete production process using smart technologies and able to prevent faults and control
unexpected losses in the production. This system makes use of technologies such as Apache Kafka, apache
Strom and MongoDB for effective data processing and storage.
This system is more scalable and able to process large amounts of data in a continuous manner effectively
compared to traditional models.
This system is able to identify in a timely manner whether the functioning is normal or abnormal in every aspect
of the production process with high accuracy. Therefore, smart technologies provide effective solutions to the
semiconductor manufacturing process, but there is a possibility of making scalable systems for processing
complex manufacturing processes.
The presented approach addresses a complete overview of the controlling and monitoring process. Furthermore,
the approach will prove useful in overcoming the subjective determination of smart technologies usage in
semiconductor manufacturing.

VIII. REFERENCES
[1] B. V. Dasarathy. (1991). Nearest Neighbor (NN) Norms: NN Pattern Classification Techniques. Los
Alamitos, in CA: IEEE Comput. Soc. Press.
[2] Gian Antonio Susto., Simone Pampuri., Andrea Schirru., Giuseppe De Nicolao., Sean McLoone.,
Alessandro Beghi. (2012). Automatic Control and Machine Learning for Semiconductor
Manufacturing: Review and Challenges.Technical University of Denmark, Kgs. Lyngby, Denmark,
The 10th European Workshop on Advanced Control and Diagnosis.
[3] Moyne James., Iskandar Jimmy. (2017). Big Data Analytics for Smart Manufacturing: Case Studies
in Semiconductor Manufacturing. Processes(MDPI).

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[161]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
[4] Jang Seok-Woo., Kim Gye-Young. (2017). Amonitoring method of semiconductor manufacturing
processes using Internet of Things– based big data analysis. International Journal of Distributed Sensor
Networks, Vol. 13(7).
[5] Stanisavljevic Darko., Spitzer Michael. A Review of Related Work on Machine Learning in
Semiconductor Manufacturing and Assembly Lines.
[6] Gardner Mike., Bieker Jack. (2000). Data Mining Solves Tough Semiconductor Manufacturing
Problems. Motorola Laboratories. Boston, MA ,USA.
[7] Wang Junliang & Zhang Jie. (2016). Big data analytics for forecasting cycle time in semiconductor
wafer fabrication system, International Journal of Production
[8] Research,54:23,DOI:10.1080/00207543.2016.1174789, (p.p.7231- 7244).
[9] Doering Robert., Yoshio Nishi. Handbook of semiconductor manufacturing technology. CRC
press(Taylor and Francis Group)
[10] KHAKIFIROOZ MARZIEH 1., FATHI MAHDI AND KAN WU. (2019). Development of Smart
Semiconductor Manufacturing: Operations Research and Data Science Perspectives. Date of
publication.
[11] Saqlain Muhammad., Piao Minghao., Shim Youngbok and Lee Jong Yun. (2019) . Framework of an
IoT-based Industrial Data Management for Smart Manufacturing. Journal of sensor and Actuator
networks.
[12] Syafrudin Muhammad , Alfian Ganjar , FitriyaniD Norma Latif and Jongtae Rhee.(2018).
Performance Analysis of IoT-Based Sensor, Big DataProcessing, and Machine Learning Model for
Real-Time Monitoring System in Automotive Manufacturing. Sensors(MDPI).
[13] Tilouche Shaima, Bassetto Samuel and Partovi Nia Vahid. Classification Algorithms for Virtual
Metrology. GERAD Research Center.
[14] Koitzsch Matthias., Merhof Jochen., Michl Markus and Lebrecht Holger. (2011). IMPLEMENTING
VIRTUAL METROLOGY INTO SEMICONDUCTOR PRODUCTION PROCESSES – AN
INVESTMENT ASSESSMENT. Proceedings of the 2011 Winter Simulation Conference S. Jain, R.R.
Creasey, J. Himmelspach, K.P. White, and M. Fu, eds.
[15] Yang Ryan and Chen Rongshun. (2010). Real-Time Plasma Process Condition Sensing and
Abnormal Process Detection. Open accesss sensors. Sensors 2010. ISSN 1424-8220.
[16] FERREIRA .A, ROUSSY., A, CONDE .L.“ Virtual Metrology Models for predicting physical
measurement in semiconductor manufacturing“.
[17] Susto, G. A., Schirru, A., Pampuri, S., McLoone, S., & Beghi, A. (2015). Machine Learning for
Predictive Maintenance: A Multiple Classifiers Approach. IEEE Transactions on Industrial
Informatics, 11(3), (p.p.812-820). https://doi.org/10.1109/TII.2014.2349359
[18] Scheibelhofer Peter., Gleispach Dietmar., Hayderer G¨unter and Stadlober Ernst. (2012). A
Methodology for Predictive Maintenance in Semiconductor Manufacturing. AUSTRIAN
JOURNAL OF STATISTICS Volume 41, Number 3, (p.p.161–173).
[19] Suykens .J.A.K. and Vandewalle . J. (1999). Least Squares Support Vector Machine Classifiers.
1999 Kluwer Academic Publishers. Printed in the Netherlands.
[20] Baly Ramy and Hajj Hazem. (2012). Wafer Classification Using Support Vector Machines. IEEE
TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 25, NO. 3.

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[162]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
[21] Kerdprasop Kittisak and Kerdprasop Nittaya. (2011). A Data Mining Approach to Automate Fault
Detection Model Development in the Semiconductor Manufacturing Process. INTERNATIONAL
JOURNAL OF MECHANICS. Issue 4, Volume 5.
[22] Zhou Zhe., Wen Chenglin and Yang Chunjie. (2014). Fault Detection Using Random Projections
and k-Nearest Neighbor Rule for Semiconductor Manufacturing Processes. IEEE
TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING.
[23] Zhou Zhe., Li Zuxin., Cai Zhiduan and Wang Peiliang. (2019). Fault Identification Using Fast k-
Nearest Neighbor Reconstruction.Processes (MDPI).
[24] Goodlin .Brian E.; Duane S. Boning.; Herbert H. Sawin. (2002). SIMULTANEOUS FAULT
DETECTION AND CLASSIFICATION FOR SEMICONDUCTOR MANUFACTURING TOOLS.
201st Meeting of the Electrochemical Society, International Symposium on Plasma Processing XIV,
Abs. 413, Philadelphia, PA.
[25] WANG KAIBO and HAN KAI. (2013). A batch-based run-to-run process control scheme for
semiconductor manufacturing. IIE Transactions 45, (p.p.658–669)
[26] Rossa Gordon .J., Adamsa Niall .M., Tasoulisa Dimitris .K., David .J. Handa. Exponentially Weighted
Moving Average Charts for Detecting Concept Drift.
[27] Moyne, J.; Iskandar, J.; Armacost. (2017), M. Big Data Analytics Applied to Semiconductor
Manufacturing.In Proceedings of the AIChE 3rd Annual Big Data Analytics Conference, San Antonio,
TX, USA.
[28] Alfian .G., Syafrudin .M., Ijaz .M.F., Syaekhoni .M.A., Fitriyani .N.L., Rhee .J. (2018). A Personalized
Healthcare Monitoring System for Diabetic Patients by Utilizing BLE-Based Sensors and Real-Time
Data Processing. Sensors 2018, 18, 2183.
[29] Heng A., Tan .A., Mathew J., Montgomery N., Banjevic D., and Jardine .A. (2009). Intelligent
condition-based prediction of machinery reliability. Mechanical Systems and Signal Processing,
23:(p.p.1600–1614).
[30] Sabbaghi M., Behdad S., and Zhuang .J. ( 2016). Managing consumer behavior toward on-time return
of the waste electrical and electronic equipment:A game theoretic approach,'' Int. J. Prod.
Econ., vol. 182, (p.p. 545- 563).
[31] Kuo .W. and Kim .T. (1999). “An overview of manufacturing yield and reliability modeling for
semiconductor products.” Proceedings of IEEE, vol. 87,no. 8, (p.p. 1329–1344).
[32] Lin T. H., Cheng .F. T., Wu .W. M., Kao .C. .A, Ye .A. J. and Chang .F. C. (2009). “Nn-based key-
variable selection method for enhancing virtual metrology accuracy,” Semiconductor Manufacturing,
IEEE Transactionson, vol. 22, no. 1, (p.p. 204–211).
[33] Fordyce K., Milne R. J., Wang C.-T., and Zisgen .H. (2015). ``Modeling andintegration of planning,
scheduling, and equipment conguration in semiconductor manufacturing: Part I. Review of successes
and opportunities,''Int. J. Ind. Eng., Theory, Appl. Pract., vol. 22, no. 5, (p.p. 575-600)
[34] Reinsel David., Gantz John and Rydning John. (2018). The digitalization of the world from edge to
core (Data age 2025). IDC analyse the future.
[35] Rasperry pi 4. location: https://www.raspberrypi.org/
[36] Sense Hat. Location: https://www.raspberrypi.org/products/sense-hat/
[37] Godoy Calderón. , Pérez González. (2018). Integration of Sensor and Actuator Networks and the
SCADA System to Promote the Migration of the Legacy Flexible Manufacturing System towards the
Industry 4.0 Concept. J. Sens. Actuator Netw, 7, 23.

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[163]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
Volume:02/Issue:07/July-2020 www.irjmets.com
[38] Hasserjian .K. (2016). Emerging Trends in IC Manufacturing Analytics and Decision Making.
Keynote. In Proceedings of the Advanced Process Control Conference XXVII, Phoenix, AZ, USA,
(p.p.17–20).
[39] Fang .C. ; Liu .X. ; Pardalos .P. M. ; and Pei .J. (2016), ``Optimization for a three-stage production
system in the Internet of Things: Procurement, production and product recovery, and acquisition,'' Int. J.
Adv. Manuf. Technol., vol. 83, nos. 58, (p.p. 689-710).
[40] Chang .P. C. ; and Liao T. W. (2006). ``Combining SOM and fuzzy rule +base for how time prediction
in semiconductor manufacturing factory,'' Appl. Soft Comput., vol. 6, no. 2, (p.p. 198-206).
[41] Lohokare, J. ; Dani, R. ; Rajurkar, A. ; Apte .A. ( 2017). An IoT ecosystem for the implementation
of scalable wireless home automation systems at smart city level. In Proceedings of the IEEE
Region 10 Conference, Penang, Malaysia, ; (p.p. 1503–1508).
[42] Ho .T.; Wang .W. ; Villareal .G. ; Hall .C. ; McCormick .E. (2016). Dynamic Full Trace FDC. In
Proceedings of the Advanced Process Control Conference XXVII, Phoenix, AZ, USA, (p.p.17–20).

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[164]
View publication stats