Notable Characteristics of Semiconductor Industry

Semiconductor industry is traditionally and notoriously known to be highly capital and technology intensive.

This is on account of complex and expensive machinery involved in manufacture, need for access to utilities such as clean and abundance of water and uninterrupted clean electric power and existence of efficient supply-chain from production floor to consumer. The other reason is highly skilled manpower involved in designing special purpose semiconductor chips which is expensive to hire. Coupled with this are high risks due to rapid changes in technology, long gestation period for products to surface and consequently long payback periods. This is the main reason why there are only a few successful companies the world over across a few countries that are occupying a leadership position as a global supplier of semiconductor fabrication components of different types.

Table below gives an idea of capital and design costs involved in a semiconductor design and production eco system, as per the figures reported by Mc Kinsey.

This is pushing the semiconductor companies to tirelessly work through innovative processes to minimize chip production processing times and increase productivity using technology inputs in both design and fabrication.

In a traditional technology stack the semiconductors required are for storage, memory, logic and networking. Among these, while the processors continue to occupy the premier position chasing the Moore’s law by shrinking the linewidths to 5 nanometers, as per an IRDS report, storage will witness the highest growth due to humongous amount of data that is stored and is required to be processed for deriving intelligence in the application concerned. But the semiconductor industry will continue to reap most profits in computing, memory and networking solutions creating best opportunity for value creation.

No country is 100 % independent in their semiconductor requirements even as the countries notably leading the supply are Taiwan, South Korea, Japan, USA and China. Following table gives the percentage share of semiconductor foundries revenue globally, and the semiconductor industry global market share country wise, and shows how lobsided the spread is as follows;

However, these very countries are seeking heavy investments in design and manufacturing of semiconductors to capture a larger share of global market and maintain a leadership position resulting in accelerated economic growth and giving a strategic advantage to the country globally.

India currently imports its 100% of semiconductor requirements with an estimated imports of products worth US $ 15 billion in the year 2020, as per sources from MeitY. The irony is that India enjoys the distinct status of a leading global semiconductor chip designer with approximately 20,000 engineers engaged in this profession working for practically all major global semiconductor manufacturers. We will deal with this later in this paper.

Thus far, PCs and Mobiles have been the largest consumer of semiconductors due to their exponential growth in the previous decade. With the plateauing of demand in these sectors the semiconductor industry has to redefine its strategy to focus on other sectors that promise potentially high market growth to sustain return on investment. As per the data released by IRDS, the semiconductor industry is only able to capture 20-30% of the total value of PC stack and merely 10-20 % of the smart phone market. Such a diminished value creation is not enough to justify the capital investment made in design and manufacturing of semiconductor devices unless it brings in innovation to find newer opportunities and market segments.

The revolution in Artificial Intelligence technology and its potential applications comes as a rescue to the semiconductor industry providing it the greatest opportunity in the current time to create value of the kind it has had in the previous decades.

Semiconductor Industry Redefining its Strategy

With the markets for semiconductors saturating in the hitherto fast-moving sectors of PCs and Mobiles, and significant investments already having been made by the industry, the semiconductor industry has already redefined its strategy to work with the areas that will not only continue to return on the investments already made but also justify newer investments in custom designs and manufacture. The focus is now shifting on autonomous vehicles, industrial robots, IoTs, drones etc, many of which are driven by the concept of Industry 4.0.

The applications being viewed with interest here are for semiconductors needed in embedded AI for facial recognition, speech-to-text, personal assistant, navigation and search, and use of AI technologies in high-performance computing for complex simulation & modelling, determining data patterns and prediction, data analytics and decision support functions.

At the core of this new strategy lies the technology of artificial intelligence (AI).

The various emerging AI applications share a one common feature - that is reliance on hardware as a core enabler of innovation, especially for compute and memory functions. The strategy therefore embeds into it a dual role of AI towards new semiconductor requirements, as explained below.

• One is applying AI/ML use cases to semiconductor manufacturing and designs to optimize on costs. As the manufacture of semiconductors is the largest cost driver, as per a McKinsey estimate, AI/ML will accrue most value - upto about 40% - and decrease in costs by about 17% through automation and verification during production cycles, efficient tool design for enhanced performance, use of computer vision for inspection, assess tool fatigue, prevent machine down times and minimize overall chip production processing time. Further, for semiconductor design apply AI/ML use cases to avoid time consuming iterations and ramp up yield to reduce costs possibly by upto 28-32%. Also use ML to identify patterns in component failures and predict likely failures and adopt optimal layouts that prevent or minimize failures. Further, use AI in production planning, supply- chain and optimum pricing based on consumer data analytics on the available data.

• The other is AI workloads driving the semiconductor requirements for storage, memories, logic functions, sensors etc. Many AI applications would require specialized end to end solutions necessitating designs for such range of semiconductors as workload specific AI accelerators, non-volatile memories, high speed interconnects, high bandwidth memory, on chip memory, programmable switches, AI optimized storage and networking chips and system on chip (SOC). Special requirements will also arise for the central processing units (CPUs), graphics processing units (GPUs), field programmable gate arrays (FPGAs) and application specific integrated circuits (ASICs). As per McKinsey estimates, AI accelerator chips will see the growth of 18%, almost five times that for semiconductors used for non-AI application stack.

An immediate area of high growth will be autonomous vehicles covering proximity-based vision for navigation, engine performance monitoring, tire-road grip monitoring, driver habits and fatigue monitoring etc. industrial robots, drones and IoTs. And there are other proven major areas of AI applications in verticals such as healthcare, banking & finance, agriculture, education, environment monitoring, supply - chain, retail, and manufacturing for such functions as facial recognition, voice recognition, fraud detection, image analysis, modelling, data analytics for pattern detection and prediction and decision support which would require GPUs, FPGAs and ASICs in the high-performance computing systems (HPCS) that are deployed for such applications.

Artificial Intelligence Driving Semiconductor Markets

Here are some relevant statistics of the emerging AI triggered semiconductor market. These are based on reports from McKinsey, Redline Group, MACRO OPS and IRDS.

AI spurred semiconductor market will grow @ 18 % for next five years as per McKinsey estimates.

• Accordingly, it is estimated that semiconductor industry will go from the current 20% to control 40-50 % of the total AI technology value stack in the next five years. As predicted by Grand View Research, Inc. in the IRDS report, the global AI market is forecast to grow to US $ 390 billion by 2025 and to US $ 997 billion by 2028 at a CAGR of 40.2 %. Thus, the AI triggered semiconductor market demand, which currently stands at 20 % of total AI value stack, will represent a market value of about US $ 70 billion by 2025, and about US $ 350 billion by 2028 with 40 % of total AI value stack.
• With IoT and 5G enabling the connected world with about 26 billion connected devices (a number likely to go over three times by 2025), AI technology stack will drive the effective utilization of these connected devices through various applications. This has already sparked a semiconductor chips arms race among the two main companies, Amazon and Google, that will accelerate the digital transformation from deployment of smart phones to cloud computing, originally initiated by the likes of Intel, NVidia and Qualcomm by their making significant investments into the design and manufacture of special purpose semiconductor devices needed for AI technology stack.
• Among the various semiconductor chips, AI will drive the most growth in storage, approximately 30 % year on year. This is due to the fact that in every AI application there is a presence of enormous data, much of it is unstructured, to be processed for data analytics using advanced algorithms, and then monetizing the data to be of value to the application concerned.
• The other main and critical area of value creation from AI lies in compute processing chips whose demand is growing @ 10 %. These include GPUs, IPUs, FPGAs, and ASICs. The architecture of these chips is likely to gain ground in data center and edge computing applications. Here ASICs will gain much momentum to as high as 50 % of compute chips for data center and edge computing in a cloud computing to smart phone requirements.
• AI applications have high memory- bandwidth requirements, with the demand expected to rise to US $ 12 billion by 2025.The demand is for both high bandwidth memory (HBM) and DRAMs chips, with preference for HBM chips even as these are three times more expensive than DRAMs. The on-chip memory is a faster way -times 100- to store and access data.
• With AI applications dealing with high volumes of data- going from current approximately 80 exabytes per year to 845 exabytes per year by 2025 registering annual growth of 20-25%, a potential technology to use in the chip is of non-volatile memory (NVM). Here, magneto resistive RAM has lowest latency, hence is preferred. Market prediction for this is US $ 10 billion by 2025.
• AI-specific memory chips command 300 % higher prices than standard memory chips. And for the tremendous benefits that accrue out of the AI application workloads, such a high price is accommodated by the industry.
• Another type of semiconductor chip required in AI application is for programmable switch in networking as it will require many servers spread across the globe to seamlessly connect. Cloud computing infrastructure needs such chips.
• Combining nonvolatile memory on chips with processing logic is making system- on -chip (SOC) possible as a novel design, which will meet demand for AI algorithms.

The trend is now to go for semiconductors for industry specific AI use cases, called Microvertical solution, as distinct from earlier trend to develop broad cross- industry solutions eg. GPUs by NVidia. This approach does come with a price penalty but benefits of AI technology workloads outweigh the high price of semiconductor devices designed for specific tasks.

Key to success in semiconductor industry is human and the brain behind, as it is a highly skilled job to create designs that are best fit for the application and consume least amount of time in the cycle from breadboard design- to- manufacture. There is thus a need to create an eco-system of software application developers and chip designers to build an all-inclusive environment for bringing out new designs that will enable AI technology stack to produce world class application.

There is a woeful shortage of skills across the spectrum of semiconductor designers to production floor engineers to supply - chain specialists. As per Semiconductor Engineering (SEMI) estimate there are more than 10,000 job openings in global semiconductor industry. Hence it is a much sought after and lucrative job of today. Companies of the likes of Amazon, Google, Apple, Intel are instituting special programs to recruit engineers as chip designers needed for the AI driven semiconductor industry.

Areas Where Special Semiconductor Chips Require to Handle AI Workloads

In the following paras we describe the areas where specially designed semiconductor chips are used to handle AI workloads giving tremendous benefits of AI applications in the concerned verticals.

• Use of GPUs has been made to accelerate speed of Intel designed semiconductor processors that are required in a number of applications dealing with simulation & modelling, pattern recognition & prediction, algorithms optimization, high end data processing & analytics etc. in various verticals where AI/ML technologies are being used.
• Beyond the GPUs is the development of Intelligence Processing Unit (IPU), a massively parallel processing unit developed by Graphcore for use in massively parallel processing architecture based high performance computing systems. Its multiple instruction multiple data architecture reflects  the requirements of modern and advanced AI workloads.
• With the AI technology applications gaining traction in health care, agriculture, education, manufacturing, supply- chain, retail, social networks monitoring, military etc, the demand for specialized sensors, ASICs and improved memory is increasing for use in high performance computing platforms needed for these applications. The example AI usages in some of these verticals are:

Healthcare - Automating medical diagnostics, speedy and accurate image analyses, assisted clinical analysis, health data management, robot assisted surgery, genomic  sequencing, molecular structure modelling for drug discoveries

Education - AI enabled tutoring, ITS, digital libraries, pedagogy and content management, cognitive learning.

Finance - Fraud detection, loan/ investment processing, credit risk assessments, companies’ data monitoring and reporting.

Agriculture - Soil monitoring and disease monitoring based on satellite data, crop yield optimization and cycle management, predictive weather impacts, farm inputs optimization.

Supply-chain - Optimizing the processes for transferring products from production to consumer, predicting any bottlenecks in supply and identify alternate routes, inventory optimization.

Retail - Analyzing trends and consumer preferences, predicting supply needs, creating data for future procurement planning, pricing etc.

Social Media - Chasing the social media networks for data accuracy and authenticity and taking corrective actions, chasing for preferred contents and advertisements etc.

Urban environment monitoring - Modelling to cover environmental pollution, weather forecasting, transportation, security, medical and utility services.

• As opined by Kai Fu Lee, founder of Fund Sinovation, in his recent book on AI -2041, AI from being used as a technology for solving complex data problems, will increasingly handle our day to day lives in this and next decade. We already are using AI behind such common usages as Google search, Alexa and Siri for our day-to-day queries, voice commands and facial identification replacing keyboard, touch screen etc, road guidance and navigation in real-time, short-range weather prediction, online shopping with customer preferences, email system filtering SPAMs, social media sites to tag pictures etc. Behind such usages are complex algorithms and semiconductor chips for ultra-high-speed processing, fast memory access and storage.
• There are other usage of AI as well like traffic monitoring and guidance, urban sewer usage, smart city infrastructure operation & maintenance, utilities like water and electricity distribution and performance monitoring and maintenance calls. All these are right candidate for using high performance computing and data to run efficient AI algorithms that will require high quality semiconductor chips.
• Nonvolatile memories are needed for data analytics in all those applications that deal with processing the huge amounts of data.
• Semiconductor chips used for AI applications that stimulate the demand for automation in capital goods and create opportunities for semiconductor manufacturing equipment suppliers and also for the computing machines used by the chip designers.
• Semiconductor chips used to build hyperscale servers (high performance computing machines) required for Cloud computing and Data Centers that are needed to run AI application.
• AI technology workloads for a range of functions such as facial recognition, image processing, speech- to- text, personal assistance, data base search that need special semiconductor processors as GPUs.
• Platforms such as autonomous cars, industrial robots, drones, smart phones, 5G connects, IOTs, security cameras require semiconductor chips that run AI applications for achieving their enhanced functionality. These platforms are now increasingly driving the semiconductor markets after plateauing of the PCs and Mobiles requirement of chips.
• Making AI functionality available to all by providing high end GPU based computing capability on Cloud so that small units and startups can develop smart AI applications without their investing in owning high performance computing infrastructure.

As AI applications are growing with the adoption of technology components of machine learning (ML), deep learning (DL), computer vision, natural language processing (NLP), robotics in different application areas, it is throwing new challenges and opportunities to the semiconductor industry for designing and offering special purpose chips that match the requirements of the application.

The various emerging AI applications share one common feature. That is reliance on hardware as a core enabler of innovations, especially for logic and memory functions. It is this eco system of software application and hardware components together that is set to create unprecedented opportunities and economic value from the new and huge semiconductor triggered AI markets.

India’s Aspirations for Semiconductor Industry

Recognizing the importance of semiconductors as the life force of electronics industry and the consequent digital transformation in the economy, India has been endeavoring to create an eco-system of semiconductor design, manufacturing, and supply due to its strategic importance in the country as a key to the security of critical information infrastructure.

As a first notable initiative in this, the Government had set up Semiconductor Complex Ltd. (SCL), a company under the Ministry of Electronics and Information Technology (MeitY), several decades ago. It created awareness about the important attributes of this industry like the choice of suitable manufacturing machines, maintenance of micron level clean environment for the wafer fabrication machinery, technology of packaging and the relevance to the linewidths of the semiconductor designs, and its eventual application in an electronics product. This experience brought in the exposure to the complexity of the processes involved and a need to control the evolving technology of production. This initiative however could not provide an encouraging ground for the industry to take on commercial scale semiconductor manufacturing operation.

A significant gainer behind this initiative was the emergence of India as a leading house of semiconductor designs for global semiconductor companies. Today India boasts of a powerhouse of a large number of small and medium size companies and startups engaged in design of semiconductors for nearly 90% of the semiconductor manufactures globally, producing 2000 new chip designs every year employing over 20,000 quality engineers, as per an estimate by MeitY.

In a software obsessed India globally known for its quality and cost -competitive software, which shares a significant part of overall exports from India, there has been a continuing felt need for undertaking semiconductor manufacture in order to have complete autonomy in the value chain of the semiconductors industry thereby achieve full control on production of electronics products.

There have been attempts by the Government to woo the global manufacturers of electronics products by giving incentives of tax credits, customs duty exemptions and other fiscal benefits. This has showed the way to an extent that India today is the second largest producer of mobile phones globally, after China. One would like to see such a position emerging in many other fields of electronics products like telecommunication, IoT, office automation, computers and white goods. This can lead to ‘advantage India’ if we have a control on semiconductor industry.

No country has been able to establish a monopoly in the semiconductor value chain, however, Government of India recently announced its decision to engage in having a complete eco system of semiconductors, from design, fabrication, packaging and testing to supply, put in place.

This is a tall order and would require huge investments, a long-term strategy and capitalize on the existing strengths.

Following are main components of this very ambitious semiconductors plan unveiled by India in December 2021, in furtherance of ‘Atmanirbhar Bharat’ pronounced by the Prime Minister of India. The plan is expected to pave the way for India’s technological leadership in the area of electronics and information technology as of being strategic importance and economic self-reliance.

Semiconductors and displays being the foundation of modern electronics, the plan will have a multiplier impact on significantly enhancing domestic capacities of electronics industry in creating a US $ 1 trillion digital economy within a US $ 5 trillion GDP by 2025 (as announced by the Minister of Electronics and Information Technology, Government of India, and mentioned in the PIB release);

i. In order to drive this plan, set up India Semiconductor Mission (ISM), a body of experts, to act as a nodal agency for efficient and smooth implementation of the program and build a long-term 20-year road map strategy for a sustainable eco system of semiconductors and displays design, manufacture and integration with the electronics industry needs.

ii. Plan for development of semiconductors and display manufacturing eco system at an out lay of Rs 76,000 Cr (US $ 10.2 billion) for the next 5 years.

iii. Set up two green fields semiconductor and display fabs each with a fiscal support of 50% of project cost on a pari- passu basis to eligible selected partners having capital and technology resources to execute this project.

iv. Commercialization of SCL with a fab partner in a joint venture to modernize this brown field fab facility and commence operation.

v. Set up 15 units of compound semiconductors/photonics components/sensors including MEMS fabs/semiconductor (ATMP/OSAT) packaging facilities with a fiscal support of 30% of total project cost.

vi. Extend design linked incentives of up to 50 % of eligible expenditure and product deployment linked incentive of 4-6% on net sales for 5 years to 100% owned domestic companies for semiconductor designs for growth of not less than 20 companies which can have a turnover of 1500 Cr in next 5 years.

vii. Additionally, provide incentives for complete supply chain covering electronics components, sub-assemblies and finished goods and ancillary products like batteries, auto components, telecom, networking, solar PV modules and white goods, totaling an amount of Rs 230,000 Cr (US$ 30 billion).

As a step towards execution of this plan, opportunities are open effective January 1, 2022, to all semiconductor design and manufacturing units to submit applications, as reported by MeitY. As also reported, it has also announced its program, Chips to Start up (C2S), which addresses each entity of the value chain in electronics, viz. quality manpower training, research and development, hardware IPs design, application-oriented R&D, prototype design and deployment, open to all academia, industry, startups, and R&D organizations. Applications have been sought from 100 such institutions under which it aims to train 85,000 high quality engineers in the area of VLSI and embedded system design, as well as result in development of 175 ASICs, working prototypes of 20 system on chip (SOC) and IP Core repository over a period of 5 years. This is visualized to be a step towards leapfrogging in the electronics system design and manufacturing (ESDM) space by way of inculcating the culture of SOC/system level design at bachelors, masters, and research levels and act as a catalyst for growth of the startups for fabless design.

The Center for Development of Advanced Computing (C-DAC) in India  has been given a nodal- coordinating responsibility for the C2S program. It is interesting to note here that C-DAC is currently executing a National Supercomputing Mission (NSM), which is a major program of the MeitY and the Department of Science and Technology to build an exascale supercomputer covering all the underlying technologies, in specific the compute processor and accelerator semiconductor chips, memory and networking chips, systems software, libraries and utilities for applications of AI and ML in solving various science and engineering problems. C-DAC already has delivered its PARAM range of tera and peta scale high performance supercomputers for research and development on various advanced applications by research and academic institutions.

The overall plan is designed to position India as a global hub of electronics manufacturing with semiconductors as foundational building block. It is expected that the program will create specialized jobs of 35,000 with an additional 100,000 indirect employment, and push electronics production to US$ 300 billion in next 5-6 years. A return on this investment of Rs 170,000 Cr is also expected.

With the AI Mission already put in place by the Government of India and an emerging plethora of its applications in the societal and economic sectors, the above program to initiate the ISM will be a significant boost to accelerating the usage of AI in businesses to industry, to corporates, and to homes in our day to day lives with a very large percentage of domestic content.

To Conclude

With semiconductors recognized as the life force of electronics industry, it plays an eminent role in digital transformation of the society and a significant value creator for the economy. This has been further fueled by the revolution caused by the Artificial Intelligence technology and its plethora of applications emerging in all businesses and our day to day lives. Such an ingress of AI in optimizing the design of both semiconductors and the production machineries of this capital and technology intensive industry, and in AI workloads applications to practically every sector is claiming to accelerate demands of semiconductors of special designs.

The main conclusions of semiconductor driven AI applications as forecast are;

i. AI spurred semiconductor market is set to grow @ 18 % over the next 5 years,

ii. AI applications would push semiconductor companies to capture 40-50 % of total value from AI technology stack,

iii. Storage will experience the highest growth, but the semiconductor companies will capture most value from compute, memory and networking chips,

iv. As opposed to a limited value capture in the past from generic design components, semiconductor companies will adopt high value creation strategy that will enable offer customized chip solutions.

Semiconductors are so sensitive to the requirements of electronics industry that in recent time their shortage primarily from the Asian chip production plants due to the pandemic caused a significant monetary loss to the world economy. This is evident from a recent News report that in US alone the loss to the economy was to the extent of US $ 240 billion last year, a significant part of it was in automobile production that forced companies to step up their new chip production plants in US.

The countries world over are clamoring to seek heavy investment in semiconductors for research and in capital goods to capture a larger share of global market and acquire a leadership position resulting in accelerated economic growth and a strategic advantage to the country globally. India has taken a belated but a welcome step to unveil an ambitious plan for complete value chain of semiconductor design, fabrication, packaging and testing and supply for products to the markets. This is designed to catapult India as a global player in the field and pave the way for its technological leadership in electronics and information technology industry as being of strategic importance and economic self-reliance.

i. Artificial-Intelligence hardware: New opportunities for semiconductor companies - McKinsey Report

ii. International Roadmap for Devices and Systems (IRDS) Report - https://irds.ieee.org/topics/semiconductors-and-artificial-intelligence

iii. Government of India announcement on India Semiconductor Mission in PIB Release - https://pib.gov.in/PressReleasePage.aspx?PRID=1781723