Semiconductors: Growth Story Milestones by GB Reddy Sir

Semiconductors: Growth Story Milestones

A $500 billion industry that is expected to grow to $ 1 trillion by 2030! In India semiconductor consumption is expected to grow to roughly $70 billion by 2026 and to cross $148 billion by 2030 – mostly from imports, raising the import bill. China buys more than 50% of the chips manufactured globally.

The Chip-makers

Semiconductors were invented in the U.S, but over time East Asia emerged as a manufacturing hub, largely because of government incentives. This allowed the U.S. to develop business ties and strategic alliances in the face of Beijing's growing influence in the Asia Pacific.

Major Semiconductor Companies: U.S.A. - Micron, AMD , Intel, Apple, Qualcomm, Broadcom, NVIDIA and others; Taiwan – TSMC, PSMC; UMC, SK hynix and others: Japan - TEL, Toshiba, Sony, Fujitsu, Mitsubishi Electric and many others; South Korea - Samsung and others; and China – SMIC, Huawei and others.

Types

There are 2 types of semiconductors: p- or n-type. In 1968, the 20pm type was made that was improved by 1984 to 1pm type. Thereafter, the n-type started from 800nm in 1987 to 3nm by Samsung mass producing in mid-2022. As of date, the transition to 2nm by 2025 represents a quantum leap in chip design and fabrication. The latest under development is 1.5nm.

The Nanometer Usage Story

The usage of technology is given below:

· 3nm to 7nm (leading edge nodes) – Smartphones, AI, data centres, Graphic processing units (GPUs), high performance computing (HPC), automotive advanced driver assistance systems (ADAS), defence and mission critical hardware and supersonics.

· 16nm to 28nm – mid range smartphones, laptops and tablets, HPC, IoT devices, sensors, EVs, automotive, consumer electronics

· 40nm to 65 & 90nm - Automotive, computing, data storage, wireless communications, power management, set top box, smart home, smart grid, medtronics, industrial CPUs.

· 180nm, 250nm & beyond (older nodes) – Industrial controls, ol;der telecom equipment

· Analog Chips – communications inside phones, consumer electronics, industrial, analog to digital converter, telecom towers.

· Compound Chips – Radio frequency (RF) and microwave for communication systems, satellite communications, EVs, LEDs

The GLOBAL Semiconductor VALUE CHAIN (GVC)

The global semiconductor value chain is dominated by the U.S., Taiwan, South Korea, Japan, Europe, and China. An iPhone has chips that are designed in the US, manufactured in Taiwan, Japan or South Korea, assembled in China. India, which is assembling iPhones, could play a bigger role in the future.

The Race

The race is on to make the best and most efficient chips at scale - the smaller, the better. Their manufacture is complex, specialist and deeply integrated. The U.S. and China are battling over one another for the precious resource, which is reshaping the global economy. Whoever controls the supply chains - a tangled network of companies and countries that make the chips - holds the key to being an unrivalled superpower.

The challenge: how many transistors - tiny electrical switches that can turn a current on or off - can you fit onto the smallest bit of a silicon wafer? How narrow is that? Much narrower than a strand of human hair, which is about 50 to 100,000 nanometres. Smaller "leading edge" chips are more powerful, which means they go into more valuable devices - supercomputers and AI, the internet of things.

As of date, Cerebras Systems has doubled its existing world record of fastest AI chip - Wafer Scale Engine 3 (WSE-3). The WSE-3 delivers twice the performance of the previous record-holder, the Cerebras WSE-2, at the same power draw and for the same price. Purpose built for training the industry’s largest AI models, the 5nm-based, 4 trillion transistor WSE-3 powers the Cerebras CS-3 AI supercomputer, delivering 125 petaflops of peak AI performance through 900,000 AI optimized compute cores. 24 trillion parameter models can be stored in a single logical memory space without partitioning or refactoring, dramatically simplifying training workflow and accelerating developer productivity.

With every component optimized for AI work, CS-3 delivers more compute performance at less space and less power than any other system. While GPUs power consumption is doubling generation to generation, the CS-3 doubles performance but stays within the same power envelope. The CS-3 offers superior ease of use, requiring 97% less code than GPUs for LLMs and the ability to train models ranging from 1B to 24T parameters in purely data parallel mode.

The CS-3 is designed to train next generation frontier models 10x larger than GPT-4 and Gemini. Training a one-trillion parameter model on the CS-3 is as straightforward as training a one billion parameter model on GPUs.

China too has made chip production a national priority and is investing aggressively in supercomputers and AI. It is nowhere near being a global leader but has been catching up quickly in the past decade, especially in its chip design capabilities.

The U.S. is trying to cut China off from the tech that makes chips. The U.S. also wants to make more chips. The Chips and Science Act offers $53bn of grants and subsidies to companies making semiconductors in the US. Major players are taking advantage of that. TSMC is investing in two plants worth $40 billion in the US, their only facilities outside of Taiwan.

What about technology obsolescence? The market for "lagging edge" chips (5 nm chips)- which power the more mundane bits of our lives, such as microwaves, washing machines and refrigerators is likely to wither in the future.

Taiwan

The four key contract chipmakers in Taiwan: Taiwan Semiconductor Manufacturing Co (TSMC); Powerchip Semiconductor Manufacturing Corporation (PSMC); United Microelectronics Corp (UMC) and Vanguard International Semiconductor Corp (VMC). TSMC, UMC and PSMC account for 62% of the worldwide market.

By 2018, TSMC, the world’s largest contract chipmaker, first achieved 7nm mass production without Extreme Ultraviolet Lithography (EUV) equipment. Also, it is using EUV machines to mass produce 5nm chips. Presently, TSMC is making significant strides towards its 2nm production capabilities. With a massive investment exceeding USD 12.3 billion in EUV equipment, TSMC is set to begin mass production of its advanced 2nm process in 2025. The TSMC Center is also focused on developing 1.4nm Technology.

TSMC is investing US$ 3.3 billion to build and equip a new global R&D center located in Hsinchu, Taiwan that will employ more than 7,000 people to focus on the development of 2nm, 1.4nm and even more advanced process technologies to keep up with changes. TSMC’s investment is likely to acquire more standard EUV machines to bolster their production capacity for the upcoming 2nm process (N2) in 2025.

Next, PSMC is Taiwan's third-largest contract chip manufacturer and the sixth-biggest in the world. PSMC specializes in mature technologies and owns two 8-inch wafer fabs with a monthly capacity of 110,000 units and three 12-inch wafer plants with a monthly capacity of 110,000. It is expanding with plans to build a series of fab factories outside its base. The company develops, manufactures, and distributes advanced memory components.

China

Semiconductor Manufacturing International Corporation (SMIC) is the largest contract chip maker in mainland China. SMIC has wafer fab sites throughout mainland China, offices in the United States, Italy, Japan, and Taiwan, and a representative office in Hong Kong. It provides integrated circuit (IC) manufacturing services from 350 nm to 7 nm process technologies.

During the 2010s, Huawei was at the forefront of China’s remarkable rise in the smartphone, computer, and artificial intelligence (AI) semiconductor chip design industry. In late 2018, there were only two companies in the world selling smartphones with 7nm mobile application processors, Apple and Huawei, both of which designed the chips in house and outsourced manufacturing to Taiwan's TSMC.

By 2019, the Chinese SN1 facility was producing 14nm chips at the rate of 3,000 finished wafers per month (WPM). And, by June 2020, it was increased to 6,000 WPM. In February 2021 SMIC claimed that SN1 had achieved 15,000 WPM production capacity by the end of 2020. Huawei only announced in March 2023 that it had finally made a breakthrough on 14 nm software tools after years of work.

SMIC’s most advanced chip fab is SN2 is the facility where SMIC conducts advanced node research and development (R&D). SMIC has begun mass production of its 7 nm (N+1 and N+2) processes. SMIC was producing and selling 7 nm chips as early as July 2021, despite having no EUV machines. SMIC’s initial 7 nm chips (N+1 process) were specialized chips for cryptocurrency mining. According to a June 2020 report, SMIC planned for 7 nm production capacity at SN2 to also reach 35,000 WPM.

As per experts, there is no 7 nm fab in the world that does not rely upon controlled U.S. technology, so it is quite clear that SMIC is using U.S. technology in producing these chips for Huawei. Blocking EUV sales required a policy change by the Dutch government since the sole supplier of complete EUV lithography machines is a Dutch company, ASML. The Trump administration reportedly reached an informal agreement with the Dutch government in early 2020. So, blocking China from acquiring EUV technology has complicated China’s path to producing chips at technology nodes more advanced than 7 nm. However, it actually did not block SMIC from legally acquiring all the equipment required to manufacture 7 nm chips, since much of the advanced deposition, etching, inspection, and metrology equipment was not blocked from purchase. Moreover, advanced deep ultraviolet (DUV) lithography equipment can be used as an alternative to EUV for the production of 7 nm chips.

The companies that do have mature EDA software available for designing 7 nm chips are all American. Prior to Huawei’s entity listing, Huawei was legally allowed to license this software from the companies and did so. That was how Huawei designed its first 7 nm chip back in 2018. However, the U.S. providers of EDA software have been prohibited from renewing Huawei’s software licenses or providing software updates since 2020. Chinese software piracy is a well-known problem in the EDA industry. Thus, the fact that Huawei and SMIC used U.S. technology is not controversial. It is clearly the case.

SMIC is likely to reach production capacity of 50,000 WPM across SN1 and SN2 by the end of 2024. SMIC has existing customers for its 14 nm capacity, so presumably it will not immediately reallocate all of its machines to 7 nm. SMIC is also planned as recently as September 2022 to eventually pursue 5 nm production at SN2 despite lacking access to EUV machines.

(Part 2 – India’s Lost Opportunities and Initiatives)