Electronic Components Provenance Assurance Through AI and Big Data in Assembly Processes - Eyal Weiss

This paper presents a groundbreaking method for
real-time authentication of electronic components during the
assembly process. Leveraging images captured by pick-and-place
machines, our approach utilizes AI and big data algorithms to
discern subtle differences in component appearances, allowing
for the identification of specific manufacturing lines and
authentic manufacturers. By training our algorithm on a vast
dataset of over 4 billion components, we achieve exceptional
performance parameters in most cases exceeding 99.8%,
contingent upon the package case of the component. Real-life
examples illustrate instances of authenticity mismatches
encountered in production, highlighting the method’s
effectiveness in identifying errors in the supply chain and
detecting counterfeit components. Through cross-referencing
with top-side images and supply chain documentation, we
validate the results, establishing a new standard for authenticity
assurance without reliance on traditional methods. This
methodology represents a true AI revolution, promising
enhanced supply chain integrity, cost reduction, and mitigation
of risks associated with counterfeit components.
Index Terms— Authentication, Electronic Components,
Counterfeit Detection, Artificial Intelligence, Big Data Analytics,
Manufacturing, Supply Chain Management, Provenance,
Traceability, Pick-and-Place Machines, Machine Learning,
Image Analysis.
I. INTRODUCTION
HE rise of counterfeit electronic components poses
significant challenges to manufacturers, distributors,
and consumers alike. Counterfeit components not only
jeopardize the integrity and performance of end products but
also pose serious safety risks, particularly in critical
applications such as aerospace, automotive, and medical
devices. In-line production, characterized by the continuous
assembly and testing of electronic products within
manufacturing facilities, presents a unique set of challenges
for ensuring component authenticity. Traditional methods of
component authentication, such as manual inspection [1], [2],
[3], [4], [5], [6], [7], [8], [9], [10], [11], [12], are often
inadequate in detecting sophisticated counterfeit schemes
prevalent in today’s global supply chains. Consequently, there
is a pressing need for advanced and automated solutions
capable of identifying counterfeit components in real-time
during the production process. The significance of electronic
component authentication and counterfeit detection in in-line
production extends beyond ensuring product quality and
E. Weiss is with Cybord ltd, Tel-Aviv, 6744332 ISRAEL (e-mail: eyal.w@
cybord.ai).
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org
safety. It also encompasses broader concerns related to supply
chain integrity, intellectual property protection, and regulatory
compliance. Manufacturers face increasing pressure to
implement robust authentication measures to safeguard their
brand reputation and mitigate potential legal liabilities
associated with the use of counterfeit components.
A. Overview of Existing Challenges in Ensuring Component
Authenticity
Despite advancements in manufacturing and supply chain
management, ensuring the authenticity of electronic
components remains a formidable challenge for industry
stakeholders. The globalization of manufacturing has led to
increasingly complex and fragmented supply chains, spanning
multiple countries and involving numerous suppliers and
distributors. This complexity creates opportunities for
counterfeiters to introduce fake components into the supply
chain, often exploiting vulnerabilities in sourcing practices
and distribution networks. Counterfeiters continually evolve
their techniques to mimic genuine electronic components,
making detection increasingly difficult. These techniques may
include re-marking, re-packaging, and even manufacturing
counterfeit components that closely resemble authentic ones
[9], [10], [11], [12], [13]. As a result, traditional methods
based solely on examination or manual testing are no longer
sufficient to identify counterfeit components reliably and in
scale.
To the untrained eye, components from different
manufacturers may appear identical when housed in the same
package case. This similarity is expected, given that they serve
the same function and should occupy the same footprint on the
PCB. However, upon closer examination of images depicting
components from various manufacturers, subtle yet consistent
differences may be identified. Fig. 1, Fig. 2, and Fig. 3

Fig. 1. Example images of MLCC’s made by 6 different
manufacturers. The names of the manufacturers are presented
below each image

Fig. 2. Example images of Small Outline Transistor packages
(SOT-23-3) made by 6 different manufacturers. The names of
the manufacturers are presented below each image.

Fig. 3. Example images of IC’s in package case QFN-16 made
by 6 different manufacturers. The names of the manufacturers
are presented below each image.

showcase examples of images from six different
manufacturers of MLCC, SOT-23-3, and QFN-16
components, respectively. These nuanced distinctions become
evident upon scrutiny.
Limited transparency and traceability within supply chains
hinder efforts to verify the authenticity of electronic
components[14], [15], [16], [17], [18]. In many cases,
manufacturers lack comprehensive visibility into the origins
and movements of components throughout the supply chain,
making it challenging to validate their authenticity accurately.
Furthermore, the absence of standardized traceability practices
across the industry exacerbates these challenges, resulting in
inefficiencies and inconsistencies in authentication processes.
Traditional authentication methods, such as destructive testing
or sample-based forensics, are often costly and timeconsuming, especially in production environments.
Manufacturers face pressure to balance the need for rigorous
authentication measures with the imperative to maintain
efficient production workflows and meet demanding
production schedules. As a result, there is a growing demand
for non-invasive and real-time authentication solutions
capable of integrating seamlessly into existing production
processes [19], [20], [21], [22]. Compliance with regulatory
requirements, such as RoHS (Restriction of Hazardous
Substances) and REACH (Registration, Evaluation,
Authorization, and Restriction of Chemicals), adds another
layer of complexity to electronic component authentication.
Non-compliance with these regulations not only exposes
manufacturers to legal liabilities but also undermines
consumer confidence in product safety and environmental
sustainability.
B. Introduction to the Proposed Method Leveraging AI and
Big Data Approaches
To address the challenges and advance the state-of-the-art
in electronic component authentication, this paper introduces a
novel method that harnesses the power of artificial intelligence
(AI) and big data analytics[19], [23]. Our approach is designed
to provide manufacturers with a robust and scalable solution
for detecting counterfeit components and ensuring the
integrity of in-line production processes. At the heart of our
proposed method is the utilization of existing cameras
installed in pick and place machines within production lines
[24]. These cameras capture bottom-side images of assembled
components on printed circuit boards (PCBs) as they are
mounted during the manufacturing process. Leveraging AI
algorithms trained on large datasets of authentic components,
our method intelligently analyzes these images to identify and
classify manufacturers with a high degree of accuracy[23].
Furthermore, our method incorporates big data techniques to
compare the manufacturer assignments obtained from image
analysis with documented traceability information stored in
centralized databases [14], [15], [25]. By cross-referencing the
identified manufacturers with expected suppliers and
component origins, our approach enables real-time detection
of mismatches that may indicate counterfeit components or
errors in the supply chain. The integration of AI and big data
approaches offers several key advantages over traditional
authentication methods. Firstly, it enables non-invasive and
real-time authentication of components during the production
process, minimizing disruptions to workflow efficiency.
Secondly, the method can adapt and evolve to detect emerging
counterfeit trends and sophisticated counterfeit techniques.
Moreover, the scalability and versatility of AI and big data
technologies make the method applicable across a wide range
of production environments and electronic component types.
C. Challenges in Handling a Vast Array of Component Types
One of the primary challenges in developing an effective
electronic component authentication system for in-line
production lies in the diversity and volume of components
present in the market. With hundreds of millions of
component types manufactured by thousands of different

companies across various locations worldwide, creating a
model capable of accurately identifying and classifying such a
vast array of classes presents a formidable obstacle. The
complexity of the challenge becomes evident when
considering the diverse range of factors that contribute to the
uniqueness of each component type. Variations in size, shape,
markings, and packaging, coupled with subtle differences in
manufacturing processes and materials, further exacerbate the
difficulty of developing a universal authentication model.
Moreover, the dynamic nature of the electronics industry,
characterized by rapid innovation and product evolution,
necessitates constant updates and adaptations to keep pace
with emerging component types and counterfeit techniques.
In addition to the breadth of component types, the demand
for high accuracy poses another significant challenge in this
endeavor. In in-line production environments, where
production processes operate at high speeds and efficiency is
paramount, even a small error in component authentication
can have substantial consequences. A false identification
could potentially disrupt production flow, leading to delays,
rework, and increased costs. Furthermore, the cost
implications of errors in authentication cannot be overstated.
In addition to direct financial losses resulting from production
downtime and component replacement, there are broader
ramifications for brand reputation, customer trust, and
regulatory compliance.
While advancements in AI and big data offer promising
solutions [21], [22], [26], overcoming the inherent
complexities of electronic component authentication in in-line
production will require ongoing research, experimentation,
and refinement.
II. LITERATURE REVIEW
Electronic component authentication and counterfeit
detection pose persistent challenges in manufacturing and
supply chain management[9], [10], [11], [13], [19], [20], [21].
Various methods and techniques have evolved over time to
tackle these issues, ranging from External Visual Inspection
(EVI), packaging configuration and dimensions, scanning
electron microscopy inspection, remarking and resurfacing,
resistance to solvents, scrape, dynasolve as per the growing
SAE AS6171/2 standard [27]. Visual inspection, historically a
primary method, involves examining physical characteristics
like markings and packaging, but it struggles with
sophisticated counterfeits. Destructive testing offers detailed
insights into material composition but is impractical for largescale production. Chemical analysis and electrical testing
provide non-destructive options, albeit with limitations [27].
RFID technology and blockchain offer potential solutions but
face implementation challenges. Collaborative platforms
facilitate information exchange to combat counterfeits.
AI has transformed electronic manufacturing by optimizing
processes and ensuring compliance. Predictive analytics
enable demand forecasting and inventory management, while
AI-powered quality control systems detect defects in realtime. Predictive maintenance systems prevent equipment
failures, minimizing downtime. In visual inspection for
defects, deep learning algorithms enhance accuracy by
recognizing patterns indicative of anomalies [23], [28], [29],
[30], [31], [32], [33], [34], [35]. These AI-based systems offer
scalability and consistency, transforming quality control in
manufacturing.
Our method for detecting counterfeits, defects, solderability
issues, and corrosion was published in [35], [36], [37]. By
applying this method to mass materials, we attained a unique
position to evaluate the state of components from a statistical
perspective. Consequently, we published work on the
occurrence of corrosion and solderability issues in passive
components [35], [37]. Additionally, we documented several
use cases of corrosion detected using this method,
subsequently verified by lab analysis [32], [33], [34].
Moreover, we devised a method to mitigate the risk of cracks
in MLCCs (Multilayer Ceramic Capacitors) by early detection
of evidence of corrosion in the soldering terminals, as we
demonstrated that corrosion serves as a precursor for crack
development [31].

To address the imperative need for real-time reaction and to
prevent the mounting of defective components, it became
crucial to shift the processing to the edge near the mounting
machines. The round trip of the image to the cloud would
exceed the available time frame between pickup and
placement, which is measured in milliseconds. Therefore, in
[38] a method for performing real-time processing at the edge
in a manner fast enough to react instantaneously and prevent
the mounting of defective components within milliseconds is
presented.
III. METHODOLOGY
The process initiates with acquiring images of assembled
components on printed circuit boards (PCBs) using cameras
installed in pick and place machines within the production
line[24]. These cameras capture bottom-side images of
components during the mounting process, serving as input
data for subsequent analysis.
Prior to analysis, the acquired images undergo
preprocessing to enhance quality and remove noise.
Preprocessing techniques may include noise reduction,
contrast enhancement, and image normalization to ensure
consistency and accuracy in subsequent processing steps.
Once preprocessed, the images are analyzed to extract
relevant features and characteristics indicative of component
authenticity. The features may identify unique markings,
indentations, metallization shape, leads shape, ball array,
surface roughness, graininess, or other distinguishing
attributes specific to each manufacturer. The features
extraction is done by a convolutional neural network, trained
on large datasets of components from many different sources.
The AI model analyzes the extracted features to classify
components into different categories based on their
manufacturer specific fingerprint.

To address the challenge posed by the vast number of
component types, the method utilizes a patented [23]approach that analyzes smaller groups with more prior knowledge
specific to the components’ manufacturing and composition.
We use the fact that manufacturers typically use the same
packaging machine and processes to package many
components of the same package case. For example, the
standard SOT-23 package case may host thousands of
different functionalities and variations, yet they are all
packaged by the same machine regardless of the functionality
and the die inside it. Therefore, finding the fingerprint of the
machine that packaged it imprinted is key to further reducing
the number of classes without affecting accuracy. By
classifying the package case used, we narrow down the
variants, thus enhancing performance and accuracy.
We employ an unsupervised approach to cluster the visual
features of components into groups of the same package case.
These homogeneous visual feature groups are then correlated
with the identified package case and all its synonyms. This
process allows us to establish a common ground for package
case names and enables accurate classification of components
based on their visual features.
AI algorithms for manufacturer assignment and mismatch
detection are optimized for efficiency and accuracy. Cluster
analysis techniques verify consistency in labeling, essential for
training accurate models. Real-time decision-making
facilitates proactive measures, such as blocking counterfeit
reels during production

Subsequently, the feature map is flattened and fed into a
fully connected neural network layer, which further processes
the extracted features to make predictions regarding the
component’s manufacturer. The neural network parameters are
optimized during training using a labeled dataset of authentic
components, enabling the model to learn discriminative
representations for different manufacturers.
Integration of big data techniques enhances traceability
information comparison. Verification of authenticity through
top marking inspection allows differentiation between errors
and potential counterfeits. Discrepancies prompt further
investigation to confirm component authenticity and trace
origins.
A. Algorithm Description
The algorithm flowchart is presented in Fig. 4. The
algorithm’s core functionality revolves around the
classification of manufacturers using deep learning
methodologies within the realm of computer vision.
Insufficient data can severely impair the model’s ability to
generalize effectively, potentially leading to overfitting and
suboptimal performance on unseen data. Therefore, ensuring
an ample volume of high-quality training data is paramount to
the success of the method.
To mitigate the risk of overfitting and enhance the model’s
capacity to discern subtle patterns and variations in images,
we employ a rigorous approach to dataset selection.
Specifically, we only initiate the training process for
manufacturers for which we have amassed a sufficient volume
of data. This entails ensuring that the number of images within
the training set surpasses a predefined threshold, thereby

Fig. 4. Image processing involves two parallel blocks: one for
the original image and another for its edge representation. The
final output combines the results from both blocks, each
designed similarly but with slight variations in filter usage.

Fig. 5. Algorithm for manufacturer identification. The
combination of logit scores and manufacturer classification
allows the classifiers to assess the similarity of unknown
manufacturers to known ones based on their respective scores.

ensuring the model’s robustness and efficacy.
Central to the authentication process is the combination of
logit scores and manufacturer classification. Logit scores play
a role in discerning unknown manufacturers, serving as
indicators of confidence levels associated with each
manufacturer. By analyzing the logits, classifiers can assess
the similarity of unknown manufacturers to known ones based
on their respective scores (see Fig.5).
One of the fundamental components of our algorithm is the
feature model, which diverges from conventional
classification models through its parallel architecture (Fig. 4).
Upon receiving an input image, the feature model processes it
through two parallel blocks: one dedicated to the original
image and another to its edge representation. The final output
is a fusion of the outputs from these two blocks, leveraging the
strengths of both approaches to enhance classification
accuracy.
Our base model lays the foundation for this feature model,
with a design that emphasizes parallel processing and efficient
information fusion. While both blocks share a similar design,
slight variations exist in the configuration of convolutional
layers and the number of filters utilized, allowing for
flexibility and adaptability across different datasets and
scenarios.
A. Visualization of Dominant Features with Heat Maps
Understanding where a neural network focuses its attention
can provide valuable insights into its decision-making process.
One method for visualizing this focus is Gradient-weighted
Class Activation Mapping, or Grad-CAM. Grad-CAM
operates by leveraging the gradients of a target concept, which
flow into the final convolutional layer of the network. By
analyzing these gradients, Grad-CAM generates a coarse
localization map, referred to as a heat map. This heat map
highlights the important regions in the input image that
contribute to the network’s prediction of the target concept.
However, in our specific network architecture, the final
convolutional layer is positioned distantly from the output
layer. Moreover, this layer is associated with parallel branches
of the network, further complicating the application of GradCAM. As a result, utilizing Grad-CAM for meaningful
localization in our network proves challenging and may not
yield direct results but point on the leading features. Fig. 7
presents some examples of heat maps of the components that
are also presented in table I.

In the presented examples, distinctive visual features play a
role in component classification. For the resistors and MLCCs
case, the texture of the leads emerges as a dominant feature. In
the tantalum capacitors case, it’s the rounded edges of the lead
plates that stand out. Conversely, in QFN components,
attention is drawn to the shape of the heat plate, while in SOD323 packages, the defining feature lies in the trapezoidal shape
of the edges. It’s important to note that while these features are
significant for the highlighted examples, they may not
universally apply across all classifications.

Fig. 6. Feature heat maps of different cases highlighting the
dominant features that resulted in the classification.

IV. EXPERIMENTAL SETUP
Our experiments are conducted in a controlled
manufacturing environment equipped with modern production
lines and inspection systems. These production lines feature
state-of-the-art pick and place machines from industry-leading
manufacturers, including ASMPT and Fuji, recognized for
their precision and reliability in electronic component
assembly.
Operating under real-world manufacturing scenarios, the
production lines facilitate continuous assembly and testing
processes. Multiple production lines, each equipped with pick
and place machines, are utilized to generate diverse datasets
reflecting variations in manufacturing conditions and
component types.
To evaluate the performance of electronic component
authentication algorithm, we employ two primary methods:
Classical machine learning techniques are utilized to
construct confusion matrices based on test datasets. We select
representative package cases and evaluate metrics such as
accuracy, precision, and recall assessing the overall
effectiveness of our authentication method.
In addition, we compare top-view images captured by
automated optical inspection (AOI) machines further along the
line with bottom-view predictions generated by the
authentication algorithm. This visual comparison enables us to identify fits and mismatches between expected and observed
manufacturers, providing insights into the accuracy and
reliability of our authentication method.
Our experimental dataset comprises data collected from
over 20 different production lines utilizing ASMPT and Fuji
pick and place machines. This extensive dataset includes over
10 million components, representing various component types,
manufacturers, and production conditions, ensuring robust
evaluation and validation of our authentication method under
diverse real-world scenarios.
The dataset employed for training and validation comprises
over 4 billion labeled images of components captured from the
bottom side by pick and place machines and over 100 million
components imaged from the top side using automated optical
inspection (AOI) systems.
The dataset of over 4 billion components reflects the natural
production mix, with about 80% passive and 20% active
components, meaning not all entries are used simultaneously.
Our method focuses on package cases, which limits the
number of manufacturers per class, facilitating efficient
training. The dataset is updated gradually as new sources are
scanned, simplifying the process. Training is conducted using
cloud-based GPU resources to ensure adequate computational
capacity and flexibility.
Data acquisition was facilitated through a custom
application programming interface (API) designed to interface
with pick and place machines and AOI systems
seamlessly[24]. This API enabled the extraction of images
without disrupting production operations, ensuring continuous
data collection in real-time.
A rigorous verification process, as outlined in our
methodology, was implemented to establish the authenticity
and traceability of components within the dataset. This process
involved cross-referencing traceability information obtained
from production logs with manufacturers assigned by our
authentication algorithm, ensuring dataset accuracy and
reliability.
The system also conducts quality checks on all components,
identifying issues such as bad pick-ups, empty nozzles, and
visual defects [38], [39]. The training models utilize images
that have passed these quality tests; components disqualified
for quality reasons are excluded from the dataset. Minor
defects are retained to accurately reflect the inspection reality
and enhance the dataset’s robustness.
The non-intrusive nature of the data collection process
minimized disruption to production operations, with the API
interfacing with machines in full producing mode. Continuous
data acquisition was achieved while maintaining optimal
manufacturing efficiency.
IV. RESULTS
A. AI Model Performance Parameters
In this section, we present the results of our experiments,
focusing on the machine learning training and testing
parameters for sample passive components and integrated
circuits (ICs). The objective is to demonstrate the
effectiveness of the proposed method in authenticating
electronic components in in-line production lines.
For the ML training and testing phase, we selected sample
passive components and ICs to represent a diverse range of
component types commonly encountered in electronic
manufacturing. It is worth noting that the number of features
varies depending on the size of the component, with smaller
components typically exhibiting fewer distinctive features.
Superficially, it may appear that all multilayer ceramic
capacitors (MLCCs) are indistinguishable to the naked eye
(see Fig. 1). However, close analysis reveals subtle differences
in surface texture, roughness, roundness, and other visual
attributes that allow for differentiation. Fig. 7 illustrates these
differences, highlighting the distinct characteristics that enable
ML algorithms to classify MLCCs accurately.
Similarly, the analysis of integrated circuits demonstrates
that components housed in the same packages but
manufactured by different companies exhibit discernible
differences. Fig. 2 and Fig. 3 showcases examples of IC
packages from various manufacturers, showcasing the
variations in surface features and markings. These differences,
although subtle, are crucial for accurate authentication and
counterfeit detection.
To assess the accuracy of the authentication method, we
employ quantitative evaluation metrics such as accuracy,
precision, recall, and F1 score. These metrics provide insights
into the performance of our ML models in correctly
classifying components across different package cases and
manufacturers.

We construct confusion matrices based on test datasets to
visualize the classification results and evaluate the
performance of our authentication method (example for
MLCC 1210 is presented at Fig. 7. By analyzing the confusion
matrices, we can identify any misclassifications or false
positives/negatives and refine our ML models accordingly.
The performance parameter for this case exceeds 99.3%.
The summary of the performance parameters for different
component types is presented in Table I.
Furthermore, we compare the performance of our proposed
method with traditional authentication methods such as visual

TABLE I
ALGORITHM PERFORMANCE PARAMETERS FOR 3 PASSIVE AND 3
IC COMPONENTS OF DIFFERENT PACKAGE CASES

Fig. 7. Confusion matrix for a small batch of MLCC 1210
classification results.

inspection, destructive testing, and manual verification. This
comparison highlights the advantages of our AI-driven
approach in terms of accuracy, efficiency, and scalability. The
confidence level of a lab operating SAE AS 6171A/B is
approximately 90% [27]. The performance of the presented
algorithms exceeds 99%.
B. Real-world Case Studies
In this section, we present visual examples derived from
actual production line data. Our method has been rigorously
applied to real-world scenarios, and the discrepancies between
the expected and analyzed data serve as clear indicators of
mismatched cases. In one example, the system detected the
mounting of an obsolete component, the MAX3232ID, which
is a 5V RS-232 transceiver IC from Texas Instruments. This
component, no longer in production, was mistakenly mounted
instead of the required ADM3202ARNZ, a 3V RS-232
transceiver IC from Analog Devices. The detection occurred
through analysis of the bottom view image (depicted on the
left side of Fig. 8), which identified the component as
originating from Texas Instruments. This determination was
subsequently confirmed through visual inspection of the top
view image, which displayed the TI marking (depicted on the
right side of Fig. 8).
The algorithm identified Texas Instruments as the
manufacturer during bottom view analysis, contrary to the
documented manufacturer indicated through the traceability
system, which was Analog Devices. This discrepancy doesn’t
necessarily imply fraudulent activity. It could stem from errors
in traceability logging or supply chain issues. These instances
highlight the vital role of automated inspection systems in
identifying discrepancies and maintaining component integrity
and provenance.
In another scenario, the production system documented the
component as QS3VH245QG, identified as Renesas
Electronics Digital Bus Switch ICs. However, upon analyzing

Fig. 8. Detecting Obsolete Component Substitution. The
figure shows detection of an obsolete MAX3232ID from
Texas Instruments instead of the required ADM3202ARNZ
from Analog Devices. The left side displays the bottom view
analysis, identifying the component as Texas Instruments,
confirmed by the TI marking in the top view (right side).

Fig. 9. Left: Bottom view image analysis detected the
manufacturer as Texas Instruments, while the traceability
system documented the component as QS3VH245QG from
Renesas Electronics. Right: Top view image obtained through
AOI confirmed the detected component as
SN74CBTLV3245ADQBR from Texas Instruments, an
interchangeable 5-Channel, 3.3-V Bus Switch with
Automotive Grade 1 specifications.
the bottom view of the component (depicted on the left side of
Fig. 9), the algorithm detected the manufacturer as Texas
Instruments, a determination later confirmed by the top view
image obtained through AOI (shown on the right side of Fig.
9). Specifically, the detected part was
SN74CBTLV3245ADQBR from Texas Instruments, classified
as a 5-Channel, 3.3-V Bus Switch with Automotive Grade 1
specifications.
Despite both components being considered interchangeable
according to the AVL, the traceability information was
incorrect. Consequently, in the event of a malfunction, the
traceability system would erroneously advise recalling the
wrong parts. While cases like these could be resolved through
top-view automated marking verification (as presented in
[25]), many components lack such markings. Therefore, the
provenance of the component can be verified solely by the
algorithm and confirmed through scrutiny of the supply chain
documents.
An instance of this analysis is evident in the following case
study. Initially documented as Panasonic, a passive component
was identified as Vishay solely through bottom view analysis.
For visual comparison, the image can be compared with
authenticated images of Vishay (depicted in Fig. 10, center)
and Panasonic (shown in Fig. 10, right). Each recognized
component is meticulously recorded with the corresponding
PCB serial number on which it was mounted, alongside its
specific reference designation, facilitating reworking if necessary.
In the following example (Fig. 11, top right), a component
was reported as Bourns; however, upon inspection, the
component image did not match the model of any Bourns
manufacturer in the AVL or any known manufacturer (top
left). The PCB serials and reference designations of the
mounted components are presented at the bottom of Fig. 11.
V. DISCUSSION
Interpreting the results of the study in the context of
electronic component authentication and counterfeit detection
sheds light on the effectiveness and implications of our
proposed methodology. The high accuracy rate achieved by
our authentication method, averaging over 99.8% with no
signs of overfitting, underscores its potential to address the
pressing challenges posed by counterfeit electronic
components in in-line production. is applicable only where
such images are available. Furthermore, the observed
differences in visual features among seemingly identical
components, such as MLCCs and ICs, highlight the nuanced
characteristics that serve as distinguishing markers for
authentication. These subtle variations, captured by the neural
network model, enable reliable classification and detection of
counterfeit components, enhancing the integrity and safety of
electronic products.

Fig. 10. Comparison of bottom view images of a passive
component initially documented as Panasonic (center) and
subsequently identified as Vishay (top right). The verified
image of Vishay (top left) provides a basis for comparison,
highlighting the subtle differences that led to the accurate
identification of the component.

Fig. 11. Comparison of reported Bourns component image
(top right) with manufacturer model (top left) and presentation
of mounted component PCB serials and reference designations
(bottom).
The implications of the study extend beyond the realm of
electronic component authentication, with far-reaching
benefits for manufacturing processes and supply chain
management. By integrating AI-driven authentication systems
into in-line production lines, manufacturers stand to gain
significant improvements in quality control, operational
efficiency, and supply chain resilience. The real-time detection
of counterfeit components enhances product integrity and
safety, mitigating the risk of faulty products reaching
consumers. The methodology relies on image data and is
applicable only where such images are available. While
machines without vision systems cannot directly utilize it, the
presented approach can also be applied to top-side images
from Automated Optical Inspection (AOI) machines.
The methodology is adaptable to various assembly lines and
can be effectively applied wherever images are captured,
including from pick-and-place machines or Automated Optical
Inspection (AOI) systems. Furthermore, it can be utilized for
authenticating components on any board, given a good quality
image of the board.
To ensure reliable feature extraction, input images must
meet specific quality standards, including resolution and
contrast. The algorithm preprocesses images, and if poor
quality is detected, it alerts the machine to adjust lighting or
resolution. This is vital, as low-quality images can lead to
improper component placement. The minimum resolution is
25 µm/pixel, and there are no significant limitations on
component size.
By establishing standardized authentication practices and
data-sharing protocols, manufacturers, distributors, and regulatory agencies can collectively combat counterfeit threats
and uphold industry standards. This collaborative approach
strengthens supply chain integrity, enhances consumer
confidence, and fosters a culture of trust and accountability.
VI. CONCLUSION
The article introduces a pioneering method for real-time
authentication of electronic components during the assembly
process. Leveraging images captured by pick-and-place
machines for alignment, the approach harnesses AI and big
data algorithms to discern subtle differences in component
appearances, indicative of the machine that packed them. By
training the algorithm on a vast dataset of over 4 billion
components, we establish a fingerprint for each of the
components production line’s specific features, extending to
authenticate the manufacturer. Through extensive testing on a
large dataset, our Convolutional Neural Network achieves
performance parameters averaging over 99.8%, contingent
upon the package case of the component.
Real-life examples are presented to explain instances of
authentic mismatches encountered in production. These
discrepancies can be attributed to supply chain errors,
inaccuracies in traceability processes, and, in some cases,
fraudulent counterfeit components. To validate the results, we
cross-reference them with top-side images containing
component traceability data and supply chain documentation.
This methodology marks a paradigm shift, as it enables the
authentication of all components without relying on
cumbersome, costly, and often unreliable supply chain
documents or sample testing. By ensuring the provenance of
all components used in production, the method establishes a
new standard for authenticity assurance. Moreover, the same
algorithmic concept is extended to authentication using topview images obtained from AOI machines, offering a
comprehensive solution for component authentication. 

References

[1] X. Meng, B. Thörnberg, and L. Olsson, “Strategic
proactive obsolescence management model,” IEEE
Trans Compon Packaging Manuf Technol, vol. 4, no.
6, pp. 1099–1108, 2014.
[2] J. Harrison, N. Asadizanjani, and M. Tehranipoor,
“On malicious implants in PCBs throughout the
supply chain,” Integration, vol. 79, pp. 12–22, 2021.
[3] N. Asadizanjani, C. Xi, and M. M. Tehranipoor,
Materials for Electronics Security and Assurance.
Elsevier, 2024.
[4] M. M. Tehranipoor, U. Guin, and D. Forte,
“Counterfeit integrated circuits,” in Counterfeit
Integrated Circuits, Springer, 2015, pp. 15–36.
[5] M. M. Tehranipoor, U. Guin, and D. Forte, “Electrical
Tests for Counterfeit Detection,” in Counterfeit
Integrated Circuits, Springer, 2015, pp. 95–107.
[6] M. M. Tehranipoor, U. Guin, and D. Forte, “Physical
tests for counterfeit detection,” in Counterfeit
Integrated Circuits, Springer, 2015, pp. 75–93.
[7] S. P. Gayialis, E. P. Kechagias, G. A. Papadopoulos,
and D. Masouras, “A Review and Classification
Framework of Traceability Approaches for Identifying
Product Supply Chain Counterfeiting,” Jun. 01, 2022,
MDPI. doi: 10.3390/su14116666.
[8] L. W. Kessler and T. Sharpe, “FAKED PARTS
DETECTION The counterfeiting of electronics
components and related parts is widespread. But
newly developed methods promise now to help
identify counterfeit plastic-encapsulated components
using detection methods that cannot be tricked. PLUS:
,” Printed Circuit Design & Fab, vol. 27, no. 6, p. 64,
2010.
[9] a S. Report, “counterfeit parts : increasing Awareness
and Developing countermeasures Counterfeit Parts :
Increasing Awareness and,” T, no. March, 2011.
[10] U. Guin and M. Tehranipoor, “On selection of
counterfeit IC detection methods,” in IEEE north
atlantic test workshop (NATW), 2013.
[11] Semiconductor Industry Association, “Winning the
battle against counterfeit semiconductor products,” no.
August, 2013, [Online]. Available:
https://www.semiconductors.org/clientuploads/AntiCounterfeiting/SIA Anti-Counterfeiting
Whitepaper.pdf
[12] E. Components, “Counterfeit Examples Electronic
Components: Counterfeit Components Avoidance
Program CCAP-101 Certified * Appendix A-6, Rev.
E,” 2013.
[13] C. E. Component, “CCAP-101 COUNTERFEIT
COMPONENTS AVOIDANCE PROGRAM , CCAP101,” pp. 1–21, 2013.
[14] G. M. Razak, L. C. Hendry, and M. Stevenson,
“Supply chain traceability: a review of the benefits
and its relationship with supply chain resilience,”
Production Planning and Control, vol. 34, no. 11, pp.
1114–1134, 2023, doi:
10.1080/09537287.2021.1983661.
[15] P. Rosales, “Traceability in the Electronics
Manufacturing Industry,” 2021.
[16] R. Schuitemaker and X. Xu, “Product traceability in
manufacturing: A technical review,” in Procedia
CIRP, Elsevier B.V., 2020, pp. 700–705. doi:
10.1016/j.procir.2020.04.078.
[17] P. Gartner, M. Benfer, A. Kuhnle, and G. Lanza,
“Potentials of Traceability Systems – A Cross-Industry
Perspective,” in Procedia CIRP, Elsevier B.V., 2021,
pp. 987–992. doi: 10.1016/j.procir.2021.11.166.
[18] A. Corallo, M. E. Latino, M. Menegoli, and P.
Pontrandolfo, “A systematic literature review to
explore traceability and lifecycle relationship,” Aug.
02, 2020, Taylor and Francis Ltd. doi:
10.1080/00207543.2020.1771455.
[19] E. Weiss, “Counterfeit mitigation by in-line deep
visula inspection,” SMTA. [Online]. Available:
http://iconnect007.uberflip.com/i/1440051-smt007-
jan2022/87? [20] and G. G. E. Weiss Dr, E. Zeev, “System and method
for detection of counterfeit and cyber electronic
components,” WO2020202154A1, 2019
[21] R. Moudgil, D. Ganta, L. Nazhandali, M. Hsiao, C.
Wang, and S. Hall, “A novel statistical and circuitbased technique for counterfeit detection in existing
ICs,” in Proceedings of the 23rd ACM international
conference on Great lakes symposium on VLSI, 2013,
pp. 1–6.
[22] C. Xi, N. Jessurun, J. True, A. A. Khan, M. M.
Tehranipoor, and N. Asadizanjani, “Machine learning
assisted counterfeit ic detection through nondestructive infrared (ir) spectroscopy material
characterization,” in 2022 IEEE 72nd Electronic
Components and Technology Conference (ECTC),
IEEE, 2022, pp. 2249–2255.
[23] E. Weiss and Z. Efrat, “SYSTEM AND METHOD
FOR ASSESSING QUALITY OF ELECTRONIC
COMPONENTS,” 2023
[24] “ASM Software Tool to Enable Cybord Integration to
Test Component Analytics and Counterfeit Detection
in Production,” Whats New in Electronics, 2021.
[Online]. Available: https://www.wnie.online/asmsoftware-tool-to-enable-cybord-integration-to-testcomponent-analytics-and-counterfeit-detection-inproduction/
[25] E. Weiss and Z. Efrat, “High Resolution Traceability,”
2023
[26] K. Huang, J. M. Carulli, and Y. Makris, “Parametric
counterfeit IC detection via support vector machines,”
in 2012 IEEE International Symposium on Defect and
Fault Tolerance in VLSI and Nanotechnology Systems
(DFT), IEEE, 2012, pp. 7–12.
[27] M. H. Azarian, “An Overview of Risk-Based EEE
Counterfeit Part Detection Based on SAE AS6171,” in
ISTFA 2018: Proceedings from the 44th International
Symposium for Testing and Failure Analysis, ASM
International, 2018.
[28] Z. Jokovic, G. Jankovic, S. Jankovic, A. Supurovic,
and V. Majstorović, “Quality 4.0 in Digital
Manufacturing–Example of Good Practice,” Quality
Innovation Prosperity, vol. 27, no. 2, pp. 177–207,
2023.
[29] K. S. S. Liyakat and K. K. S. Liyakat, “ML in the
Electronics Manufacturing Industry,” Journal of
Switching Hub, vol. 8, no. 3, pp. 9–13, 2023.
[30] M. Javaid, A. Haleem, R. P. Singh, and R. Suman,
“Significance of Quality 4.0 towards comprehensive
enhancement in manufacturing sector,” Sensors
International, vol. 2, p. 100109, 2021.
[31] E. Weiss, “Detecting Corrosion to Prevent Cracks in
MLCCs with AI,” Journal of Failure Analysis and
Prevention, 2023, doi: 10.1007/s11668-023-01824-3.
[32] E. Weiss, “Revealing Hidden Defects in Electronic
Components with an AI-Based Inspection Method: A
Corrosion Case Study,” IEEE Trans Compon
Packaging Manuf Technol, 2023, doi:
10.1109/TCPMT.2023.3293005.
[33] E. Weiss, “Preventing Corrosion-related Failures in
Electronic Assembly: A Multi-case Study Analysis,”
IEEE Trans Compon Packaging Manuf Technol, May
2023, doi: 10.1109/TCPMT.2023.3285776.
[34] E. Weiss, “AI Detection of Body Defects and
Corrosion on Leads in Electronic Components, and a
study of their Occurrence,” in 2022 IEEE
International Symposium on the Physical and Failure
Analysis of Integrated Circuits (IPFA), IEEE, 2022,
pp. 1–6.
[35] E. Weiss, “Electronic component solderability
assessment algorithm by deep external visual
inspection,” in 2020 IEEE Physical Assurance and
Inspection of Electronics (PAINE), IEEE, 2020, pp. 1–
6.
[36] E. Weiss, “Electronic Component Analytics and
Traceability.” [Online]. Available: www.cybord.ai
[37] E. Weiss and Z. Efrat, “System and method for
nondestructive assessing of solderability of electronic
components,” P-603537-PC, 2021
[38] E. Weiss, S. Caplan, K. Horn, and M. Sharabi, “RealTime Defect Detection in Electronic Components
during Assembly through Deep Learning,” Electronics
(Switzerland), vol. 13, no. 8, Apr. 2024, doi:
10.3390/electronics13081551.
[39] E. Weiss, “Advancements in Electronic Component
Assembly: Real-Time AI-Driven Inspection
Techniques,” Electronics (Basel), vol. 13, no. 18, p.
3707, Sep. 2024, doi: 10.3390/electronics13183707.

Eyal Weiss

Eyal Weiss is a multidisciplinary technology
expert, with a background in machine
learning, plasma physics, optical assemblies,
laser technology, and electromagnetics. He is
a recipient of multiple awards, including the
prestigious “Israel security prize” twice. He
has 17 years of experience leading research
at Soreq Research Center and has built
manufacturing production lines utilizing
advanced technology and AI. He holds a B.Sc in Mechanical
Engineering, M.Sc. in Plasma Physics from the Technion-Israel
Institute of Technology, and a Ph.D. in Electronic and Computer
Engineering. He is the founder and CTO of Cybord, which
develops electronic component qualification and authentication
technologies. Dr. Weiss is a member of the Israel Innovation
Authority, EIC, SAE, and IPC committees and has published over
28 peer-review articles, 6 patents, and a book.