The Semiconductor Industry: A Journey from Basics to AI Dominance

The term "semiconductor" derives from the material's unique electrical conductivity property, which falls between that of a conductor, like copper, and an insulator, like glass. The "semi" in "semiconductor" reflects this halfway or partial conductivity. What makes semiconductors especially useful is their ability to have their electrical properties altered and controlled through the introduction of impurities (a process known as doping) or by the application of electric fields or light. This controllable conductivity allows semiconductors to be the foundation of modern electronics.

In essence, semiconductors are the magic beans of our electronic world, allowing us to control and manipulate electricity in incredibly sophisticated ways to power all the gadgets and technologies we use every day.

In the semiconductor industry, key players and market segments are defined by their roles in the value chain, technological expertise, and the end markets they serve. The industry encompasses a wide range of activities from material supply and equipment manufacturing to the design, fabrication, assembly, and testing of semiconductor devices. This comprehensive ecosystem is pivotal in driving innovations across numerous sectors, including computing, telecommunications, consumer electronics, automotive, and industrial applications.

Eco System of the Semiconductor Industry

The ecosystem of this industry is complex and presently largely concentrated in certain parts of geolocation which are creating back-and-forth surges in several global redistributions of design, manufacture, and developing activities. This paper will largely avoid geo-tension issues and focus on a value chain-based ecosystem.

1.???? Integrated Device Manufacturers (IDMs)

These companies control the entire production process, from design to manufacturing to sales. Examples include Intel, which dominates in microprocessors for computers, and Samsung Electronics, which leads in memory chips and also produces a wide range of semiconductor products.

2.???? Foundries

Foundries are companies that manufacture chips designed by their customers, providing fabrication services. Taiwan Semiconductor Manufacturing Company (TSMC) is the largest and most prominent foundry, manufacturing chips for companies that do not have their fabrication facilities, such as Apple, NVIDIA, and AMD.

3.???? Fabless Companies

Fabless semiconductor companies focus on the design and sale of hardware devices and semiconductor chips while outsourcing the fabrication of these chips to foundries. Notable fabless companies include Qualcomm, Broadcom, and NVIDIA, which specialize in designing a range of semiconductor products from processors to networking chips.

4.???? Equipment Suppliers

These companies provide the machinery and tools needed to produce semiconductor devices. ASML, a Dutch company, is a leading provider of photolithography equipment essential for chip fabrication. Other significant players include Applied Materials and Lam Research, which supply equipment for various fabrication processes.

5.???? Material Suppliers

This segment includes companies that supply the raw materials and chemicals used in semiconductor manufacturing, such as silicon wafers, gases, and photolithography chemicals. Companies like Shin-Etsu Chemical and SUMCO are major suppliers of silicon wafers, which are the substrates for chip fabrication.

The Evolution of Nodes:

What is nodes

(Semiconductor Engineering, 2024) Nodes in semiconductor manufacturing indicate the features that node production lines can create on an integrated circuit, such as interconnect pitch, transistor density, transistor type, and other new technology.

In another language, In the world of electronics, think of a "node" as a measure of how small the features on a computer chip can be. The smaller the features, the more we can fit on a chip, making it faster and more efficient. Imagine a tiny city: the smaller the buildings, the more buildings you can fit into the city, and the faster you can move around it. So, when we talk about advancing to smaller nodes in chip manufacturing, it's like making our city more compact and efficient, allowing it to perform tasks faster and use less energy.

The Nodes and Semiconductor

The semiconductor industry is segmented based on node size, with divisions like 180nm, 130nm, 90nm, and down to 5nm, representing different technological advancements and capabilities within the industry. Nodes in semiconductor manufacturing refer to the process technology's feature size, affecting the chip's performance, power consumption, and area. The industry is rushing towards smaller nodes as chips are demanding to be smaller. ?Thus there are two types of nodes such as Mature Nodes and Advanced Nodes. Mature nodes are used in almost every electronic device where manufacturing leadtime product life cycle may vary on different types of chip according to the appliance such as cars, electric razors, or GPUs. Advanced nodes are smaller and are being used for AI and military applications. ?

To learn more about nodes please refer to: https://semiengineering.com/knowledge_centers/manufacturing/process/nodes/

Getting back to the segmentation based on size and efficiency can be described as below.

1.???? Memory Chips

(DRAM, NAND): Often utilize advanced nodes (e.g., 7nm, 5nm) for higher density and efficiency, crucial for storage solutions.

This segment includes Dynamic Random Access Memory (DRAM) and NAND flash memory chips, essential for data storage and memory in computers, smartphones, and other electronic devices. Samsung Electronics, SK Hynix, and Micron Technology are leaders in this segment.

2.???? Microprocessors and Microcontrollers

These are the "brains" of computers, servers, and embedded systems. Microprocessors might use leading-edge nodes (e.g., 5nm, 7nm) for high performance in computing tasks. Microcontrollers, serving varied applications, might use a broader range of nodes, balancing cost and performance. Intel and AMD are major players in microprocessors, while companies like Microchip Technology and NXP Semiconductors lead in microcontrollers for automotive and industrial applications.

3.???? Analog Semiconductors

Analog chips are used for converting real-world signals like sound and light into digital signals that electronic devices can process. Typically fabricated on larger node sizes (e.g., 65nm, 180nm) due to their less intensive computational requirements and the analog characteristics they must preserve. Texas Instruments and Analog Devices are prominent in this segment.

4.???? Connectivity Chips

These semiconductors enable wireless communication and include Wi-Fi, Bluetooth, and cellular modem chips. Qualcomm is a leading company in this segment, especially for smartphone and telecommunications equipment. (Wi-Fi, Bluetooth): Can vary widely but often use intermediate to advanced nodes (e.g., 14nm, 28nm) to optimize power efficiency and speed, essential for communication tasks.

5.???? Automotive and Industrial Semiconductors

This segment has seen significant growth due to the increasing electrification and automation of vehicles and industrial systems. These sectors often prioritize reliability and robustness over cutting-edge performance, leading to the use of larger node sizes (e.g., 40nm, 65nm, or even larger) for many applications. However, as the demand for more advanced features and better power efficiency in vehicles and industrial equipment grows, there is a trend towards using more advanced nodes (e.g., 28nm and smaller) for certain high-performance or power-sensitive components. Companies like Infineon Technologies and STMicroelectronics are key players here.

What is AI CHIP

An AI chip is designed specifically to efficiently process AI tasks, such as neural network computations and machine learning algorithms, distinguishing it from general-purpose chips. It features specialized hardware accelerators for tasks like deep learning inference and training, offering higher performance and energy efficiency for AI applications than standard CPUs or GPUs. This specialization allows AI chips to handle the massive parallelism and data-intensive nature of AI workloads effectively.

AI Chips are golder than Gold:

According to a research article?(Future, 2024), The Gold Mining industry is projected to grow from USD 208.2 Billion in 2023 to USD 274.2 Billion by 2032, exhibiting a compound annual growth rate (CAGR) of 3.50%during the forecast period (2023 - 2032). However, according to (Deloitte, 2024) they predict that the market for specialized chips optimized for generative AI will be over US$50 billion in 2024. From close to zero in 2022, that sum is expected to make up two-thirds of all AI chip sales in the year. Deloitte predicts that total AI chip sales in 2024 will be 11% of the predicted global chip market of US$576 billion. On the other hand, some fear a generative AI chip bubble: Sales will be massive in 2023 and 2024, but actual enterprise generative AI use cases could fail to materialize, and in 2025, AI chip demand could collapse, similar to what we’ve seen in crypto mining chips in 2018 and 2021. The Compound Annual Growth Rate (CAGR) for the AI chip market from 2024 to 2027, under the aggressive forecast scenario ($400 billion by 2027), would be approximately 84.82%. Under the conservative forecast scenario ($110 billion by 2027), the CAGR would be approximately 20.19%. Whatever is true it’s sure AI chip is a major disruption in the semiconductor value chain.

The Artificial Intelligence (AI) chip market is a rapidly evolving segment of the semiconductor industry, driven by the increasing demand for high-performance computing in data centers, autonomous vehicles, consumer electronics, and other AI applications. AI chips are specialized silicon chips designed to efficiently process AI tasks, such as neural network inference and machine learning algorithms, offering superior performance and energy efficiency compared to general-purpose processors.

Key Players in the AI Chip Market:

1.???? NVIDIA

A leading player in the AI chip market, NVIDIA's Graphics Processing Units (GPUs) are widely used for deep learning and AI applications. The company's CUDA platform has become a standard for AI research and development. NVIDIA's AI chips are primarily produced in East Asia, leveraging the manufacturing capabilities of TSMC and Samsung.

2.???? Intel

Intel has made significant investments in AI through its acquisition of Nervana Systems and Habana Labs, focusing on developing AI processors for data centers and neural network processing. Intel manufactures its chips in its own facilities located in the United States, Ireland, Israel, and recently announced expansions in Europe.

3.???? AMD

AMD, traditionally known for its CPUs and GPUs, has entered the AI market through its Radeon Instinct GPUs and collaborations with AI companies. AMD relies on TSMC and other foundries for chip production.

4.???? Google (Alphabet Inc.)

Google has developed its own AI processors, the Tensor Processing Units (TPUs), designed to accelerate machine learning workloads for its data centers. TPUs are custom-designed and produced by Google's manufacturing partners in East Asia.

5.???? Apple

With the introduction of the Neural Engine in its A-series and M-series chips, Apple has positioned itself as a key player in AI chips for consumer devices. Apple's chips are produced by TSMC in Taiwan.

6.???? Huawei (HiSilicon)

Despite facing significant geopolitical challenges, Huawei's HiSilicon division has developed the Kirin series and Ascend AI chips for smartphones and data centers, respectively. Production has been heavily impacted by US sanctions, affecting their manufacturing capabilities primarily located in China.

AI Chips and Their Dominance

AI chips vary widely in architecture, including GPUs, Field-Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), and others tailored for specific AI tasks. Production of these chips is highly concentrated in East Asia, with TSMC (Taiwan) and Samsung (South Korea) being the leading foundries for many companies due to their advanced manufacturing technologies.

AI chips have geo-political regulations. US-China arguments over AI dominance topic is endless and deep. The US issued a ban banning American companies from exporting chips with a score of more than 80 points by this Nvidia was banned from exporting its H-100 chip to China. However, they worked on other innovations of other chips such as H800. Fast forward to the present, Nvidia holds more than 80% of the share of AI chips in China’s $7 Billion market. The five forces of business boosed value Nvdia as one of the historical growths of a company in recorded human history. Some even argue that an 80% present market share isn’t challenged seriously. So the current potential of AI chips is a work-in-progress table of calculation conjuring.? ????

So what is GPU, CPU, or DPU?

Before further deep into the research let’s make an idea of the thing itself. How do we imagine when we say GPU or a Chip?

A CPU (Central Processing Unit) is the primary component of a computer that performs most of the processing inside a PC. Its role is to execute program instructions and manage data flow in computers. Intel and AMD are leading manufacturers.

A GPU (Graphics Processing Unit) is specialized hardware designed for rendering images and videos but now also accelerates complex computations. NVIDIA and AMD are key players in this space.

A simple GPU

An Advanced GPU

An Advanced GPU with AI Chip in middle

A DPU (Data Processing Unit), often referred to as a "SmartNIC," is designed to offload and accelerate networking, security, and storage tasks from CPUs. NVIDIA (through its acquisition of Mellanox) and Marvell are notable manufacturers.

In recent years, Intel has ventured into the discrete GPU market with its Iris Xe and Intel Arc series. The Iris Xe graphics cards are aimed at mainstream users and professional applications, offering competitive graphics performance and efficiency. The Intel Arc series represents Intel's entry into the high-performance gaming and content creation market, aiming to compete with established players like NVIDIA and AMD.

NVIDIA A Case Study:

So why so Juiced up is NVIDIA Value Chain in the AI GPU offering:

NVIDIA's dominance in the AI and HPC (High-performance Computing)sectors is largely due to its GPUs' parallel processing capabilities. Unlike traditional CPUs that process tasks sequentially, GPUs are composed of thousands of smaller, efficient cores designed to handle multiple tasks simultaneously. This architecture is particularly well-suited for the matrix and vector computations that are fundamental to machine learning and deep learning algorithms.

To understand more about parallel processing capabilities please visit here: https://www.turing.com/kb/understanding-nvidia-cuda

CUDA Platform: Central to NVIDIA's success is its CUDA (Compute Unified Device Architecture) parallel computing platform and programming model. CUDA allows developers to harness the power of NVIDIA's GPUs for general-purpose processing (GPGPU). It provides a comprehensive development environment for writing software that can scale across hundreds or thousands of GPU cores.

For more info visit here: https://developer.nvidia.com/cuda-toolkit

AI and Deep Learning: NVIDIA's GPUs are particularly effective for AI because they can accelerate the training and inference phases of deep learning models. The parallel processing capabilities allow for the handling of large datasets and complex neural networks more efficiently than CPUs. This is critical for reducing the time it takes to train models, which can involve weeks or months of computation.

For more info visit here:? https://developer.nvidia.com/deep-learning

Tensor Cores: Introduced with the Volta architecture and expanded in subsequent generations, Tensor Cores are specialized hardware units within NVIDIA's GPUs designed to accelerate the performance of tensor operations, which are at the heart of deep learning computations. This allows for significantly faster training and inference of AI models.

For more info visit here:?? https://www.nvidia.com/en-us/data-center/tensor-cores/

The continuous innovation in hardware, coupled with the comprehensive CUDA development ecosystem mentioned above, enables developers and researchers to push the boundaries of what's possible in AI and computing.

NVIDIA's GPU Products and Applications:

The type and application of GPU vary depending on usance and industry.

AI Server Chip Series: ?

These server Chips include Google TPU (Tensor Processing Unit), Intel Nervana NNP-T (Neural Network Processor for Training), Graphcore IPU (Intelligence Processing Unit)

According to (Investopedia, 2024) top revenue earner is Nvidia's AI server chips, such as the H100—a fourth-generation GPU that?can train AI language models four times faster than previous chips.?The so-called Magnificent 7,?member houses its AI chip sales in its Data Center business, which reported impressive year-over-year revenue growth of 409% in the period.

Other data center GPUs include AMD Radeon Instinct MI50, NVIDIA A100 T-Core GPU, and NVIDIA Tesla V100.

To read about more on server chips please visit this link:

https://www.investopedia.com/nvidia-stock-jumps-as-surging-ai-chip-demand-sends-earnings-soaring-8598414

*Please note the product name mentions Tesla has nothing to do with Elon Musk’s EV company but might have shared a similar pursuit to appreciate the historical figure of Nicolas Tesla.

GeForce Series:

Aimed at gamers and consumer-level graphics, the GeForce series is perhaps the most well-known of NVIDIA's offerings and one of my favorites.

For more info visit here: https://www.nvidia.com/en-us/geforce/

Quadro and NVIDIA RTX Series:

Targeted at professionals in design, animation, and other visual industries, these GPUs offer precise and high-quality rendering capabilities.

For more info visit here: https://www.nvidia.com/en-us/design-visualization/rtx/

NVIDIA cloud and data centers:

These GPUs are specifically designed for data centers, HPC, and AI workloads. They offer massively parallel processing power necessary for training and running neural networks, scientific computations, and analytics. The A100 GPU, for example, is a cornerstone of NVIDIA's AI and HPC lineup, offering unprecedented acceleration for AI research and large-scale cloud computing environments.

For more info visit here: https://www.nvidia.com/en-us/data-center/

Jetson for Edge AI: NVIDIA's Jetson platform is tailored for edge computing and AI applications. These small, power-efficient modules are designed for use in robotics, embedded computing, and IoT devices, bringing AI capabilities to the edge of the network.

For more information visit here: https://developer.nvidia.com/embedded-computing

Key Players and GPU and data centers:

NVIDIA is the leading manufacturer of GPUs, known for its CUDA parallel computing platform that allows developers to leverage GPUs for general computing tasks. NVIDIA's GPUs, particularly the Tesla and A-Series for data centers, are designed with a focus on parallel processing, offering thousands of cores for simultaneous computation. NVIDIA continues to innovate, with its architecture allowing for significant advancements in AI, gaming, and professional visualization.

AMD competes with NVIDIA in the gaming and professional GPU markets and is making strides in the data center segment. AMD's Radeon Instinct GPUs are designed for high-performance computing and machine learning, offering an alternative to NVIDIA's products. AMD utilizes its RDNA and CDNA architectures to optimize for gaming and compute workloads, respectively, emphasizing parallel processing capabilities.

For more info please visit here: https://www.amd.com/en/graphics/workstations

Intel has traditionally been known for its CPUs but has entered the GPU market with its Xe architecture, targeting a range of applications from integrated graphics to data center AI accelerations. Intel's GPUs, including the Ponte Vecchio for high-performance computing and AI, aim to offer competitive parallel processing capabilities. Intel's strategy involves leveraging its manufacturing capabilities and extensive software ecosystem to gain a foothold in the GPU market.

So what does Parallel Processing refer to in the world of AI chips?

Parallel processing refers to the ability of a computing system to perform multiple computations simultaneously. From the most prominent of the market choice, NVIDIA's GPUs are designed with thousands of cores capable of handling numerous tasks at once, unlike traditional CPUs that have a limited number of cores optimized for sequential serial processing. This architecture is particularly beneficial for AI and machine learning workloads, which involve processing vast amounts of data and performing complex mathematical computations, such as matrix multiplications and tensor operations, which are inherent in training neural networks.

Each parallel processing includes 2 phases, Training and Inference. It’s like learning and putting into use what you have learned respectively. This process initiates a learning experience pushing the boundaries of the unknown with known rationals. AI processing is inference-heavy while training is a one-time thing. ??

Training: The training phase of deep learning involves adjusting the weights of a neural network based on input data so that it can accurately predict or classify new, unseen data. This process is computationally intensive and data-heavy, requiring the repeated processing of millions or even billions of data points through multiple layers of the network. NVIDIA's GPUs accelerate this process by distributing the computations across their thousands of cores, significantly reducing the time required to train models. Technologies like NVIDIA's Tensor Cores, specialized hardware designed to accelerate tensor operations in deep learning, further enhance this capability. The training phase of deep learning, where NVIDIA's GPUs and Tensor Cores play a pivotal role, can be exemplified by the below training work applications.

Image Recognition and Classification: Training convolutional neural networks (CNNs) to recognize and classify images from datasets like ImageNet. GPUs accelerate the processing of vast numbers of images, enabling the network to learn distinguishing features of different categories (e.g., dogs vs. cats, or types of cars).

Natural Language Processing (NLP): Developing models like BERT (Bidirectional Encoder Representations from Transformers) for tasks such as language translation, sentiment analysis, and question-answering. The training involves processing vast corpora of text data, where GPUs shorten the time needed to train by parallelizing the computation.

Autonomous Vehicles: Training deep neural networks to navigate and make decisions in real-world environments. This involves processing massive datasets of sensor and camera data. NVIDIA's GPUs enable the rapid iteration of models, which is critical for developing systems that can accurately perceive and react to diverse driving conditions.

Healthcare and Medical Imaging: Training models to detect and diagnose diseases from medical images, such as X-rays or MRIs. The use of GPUs allows for the fast processing of large medical datasets, improving the ability of models to identify patterns indicative of specific health conditions.

Financial Services: Using deep learning for algorithmic trading, fraud detection, and risk management. Training models on historical financial data to predict market trends or detect anomalous transactions can be expedited with GPU acceleration.

Voice Recognition and Generation: Training deep learning models for voice assistants and speech-to-text services. This involves processing large datasets of spoken language to accurately understand and generate human-like responses. GPUs reduce the training time, enabling more sophisticated and responsive voice recognition systems.

Generative Models: Training generative adversarial networks (GANs) or variational autoencoders (VAEs) for tasks like creating realistic images, videos, or text.

Inference: Once a model is trained, it needs to make predictions or decisions based on new input data, a phase known as inference. While less computationally intensive than training, inference needs to be highly efficient and fast, especially in applications requiring real-time processing, such as autonomous vehicles or interactive AI services. NVIDIA's GPUs optimize inference workloads through parallel processing, ensuring that AI models can deliver prompt and accurate responses. It is the most costly and many innovation are driving to make this process much more economical.

In other words, Inference, in the context of artificial intelligence (AI), is like having a smart assistant that can make sense of things it has learned before. Imagine you've taught this assistant to recognize different types of fruits by showing it many pictures. Now, when you show it a new picture of a fruit, it uses what it learned to identify the fruit in that picture. This process of using learned information to make decisions or identifications on new data is called "inference." It's like using knowledge from past experiences to understand or predict something new.

Some work examples of this phase include Facial Recognition Systems, Voice Assistants, Real-time Translation Services, Autonomous Vehicle Navigation, Predictive Text and Autocorrect, Health Monitoring Devices, and Security Surveillance.

According to (IBM, 2023), the artificial neurons in a deep learning model are inspired by neurons in the brain, but they’re nowhere near as efficient. Training just one of today’s generative models can cost millions of dollars in computer processing time. But as expensive as training an AI model can be, it’s dwarfed by the expense of inferencing. Each time someone runs an AI model on their computer, or a mobile phone at the edge, there’s a cost — in kilowatt-hours, dollars, and carbon emissions.

What is Accelerated Data processing by Parallel computing? And where CUDA stands? ?

Accelerated data processing refers to the rapid handling, analysis, and manipulation of data to obtain insights or achieve computational tasks more efficiently. Parallel computing, on the other hand, is a type of computation in which many calculations or processes are carried out simultaneously, leveraging multiple processors or computers working together on a shared problem.

Parallel computing transforms data processing by breaking down large problems into smaller, concurrent tasks distributed across multiple processors, significantly speeding up computation. In another way, it is similar to the philosophy of project management principles of SCRUM. Both approaches emphasize scalability, flexibility, and continuous improvement to achieve their objectives, whether in computational tasks or project management.

NVIDIA's CUDA technology leverages GPU architectures for efficient, general-purpose processing, ideal for AI and machine learning's complex algorithms. This approach ensures scalability and adaptability, enhancing processing speeds as data volumes grow and allowing customization for specific applications. By enabling operations to run independently, parallel computing overcomes traditional bottlenecks, promising continued advancements in AI, analytics, and scientific research. The evolution of parallel computing technologies is critical for meeting the demands of modern, data-intensive tasks.

Summary of the Case Study of Nvidia:

The rapid demand, supply scarcity of raw materials, geo-political tension, and subsequent innovations have created a phenomenal organizational example like NVIDIA. The principles of calculation of data, energy efficiency, and cost competency are ever-evolving creating disruptions and NVIDIA has become the leader and mentor for this AI GPU evolution, at present.

The Future of Computation and Semiconductors:

Beyond this binary computation, there is quantum computation which is expected to be the next disruptor. Google claimed to have achieved quantum supremacy in 2019 with its 53-qubit quantum computer, Sycamore, by performing a specific computation in 200 seconds that would take the most powerful supercomputers about 10,000 years to complete. However, this is still a proof-of-concept and the problem solved was not directly applicable to practical applications.

?Summary

The semiconductor industry underpins the modern digital era, with its ecosystem spanning from material supply to the assembly and testing of semiconductor devices. The criticality of semiconductors lies in their unique electrical properties, which can be finely tuned to serve the backbone of virtually all electronic devices. This paper outlines the industry's complex value chain, including IDMs, foundries, and fabless companies, and discusses the pivotal role of semiconductor manufacturing nodes in driving technological innovation. Furthermore, it highlights the explosive growth and strategic importance of AI chips, using NVIDIA as a case study to illustrate the significant impact of GPUs and the CUDA platform on AI and machine learning advancements. The paper concludes with insights into the geopolitical factors affecting the industry and speculates on the future of computing technologies.

Bibliography

Future, M. R. (2024, March). Gold Mining Market Research Report. (A. Mandaokar, Producer) From https://www.marketresearchfuture.com/reports/gold-mining-market-16112

Insights, T. B. (2024, Jan). Artificial Intelligence Chip Market. From https://www.thebrainyinsights.com/report/artificial-intelligence-chip-market-13921

Investopedia. (2024, February 21). Nvidia Stock Jumps 9% as Surging AI Chip Demand Sends Earnings Soaring—Key Level to Watch. (T. Smith, Producer) From Investopedia: https://www.investopedia.com/nvidia-stock-jumps-as-surging-ai-chip-demand-sends-earnings-soaring-8598414

IBM. (2023, October 5). What is AI inferencing? (K. Martineau, Producer) From https://research.ibm.com/blog/AI-inference-explained#

Deloitte. (2024). Tech Media and Telecom. (C. S. Duncan Stewart, Producer) From https://www2.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predictions/2024/generative-ai-chip-market-to-reach-40-billion-in-2024.html

Semiconductor Engineering. (2024). Knowledge Center. From https://semiengineering.com/knowledge_centers/manufacturing/process/nodes/

?

?