Who ate all the ‘chips’?

Digitization, autonomous driving, artificial intelligence – none of these happen without semiconductor chips. Advances in AI/machine learning capabilities doubled every 3.4 months between 2012 and 2019,¹ in large part due to semiconductors. To facilitate this golden age of technology is going to require a lot of chips embedded in a huge array of devices.

While we all take the seamless use of our smart devices for granted, there is a hugely technological and heavily automated manufacturing process sitting behind the amazing little chips enabling the digital world of today.

A semiconductor chip (integrated circuit) is an electronic circuit built on a “semi-conducting” material. This material has an electrical conductivity that sits between a strong conductor (copper) and an insulator (rubber). Adding other elements allows the conductivity to be controlled.

It turns out that the best material for this is silicon, the second most abundant element in the earth’s crust (27.7%)⁴ and found in sand. Pure silicon ingots are sliced into wafers, on which the chips are developed. The name Silicon Valley derives from the pioneering work in semiconductors in the Santa Clara Valley, California.

Modern day semiconductors can trace their roots back to 1947 at Bell Labs, New Jersey, when Bardeen, Brattain and Shockley developed the transistor. In 1959, the transistor was “flattened” to create a “planar” manufacturing process, and in the same year, Kilby and Noyce developed an “integrated circuit” allowing multiple electronic components to operate on a single chip.⁵ These breakthroughs were the first steps in enabling mass production of integrated circuits.

In 1964, the chip with the greatest “transistor density” carried just 64 transistors. In 2020, the iPhone 12’s A14 Bionic chip contains 11.8 billion transistors, in a chip just 88 mm² in size.⁶ That is an astonishing amount of componentry in a space roughly the size of a fingernail, and there is a lot more to come.

Figure 1: More brainpower than ever is fueling the AIoT⁷ era

In 1965, Gordon Moore noticed that the number of transistors on the chips that Fairchild Semiconductor was building seemed to double roughly every two years.⁸ This rate of increase became known as Moore’s Law. It is central to the development of computational power. In essence, the industry is able to double the computational power per chip. To do this, all of the components shrink at each technology upgrade, or node as the industry calls it.

The node size has largely become a marketing feature but can be thought of as a guideline to the smallest feature size in the chip. In the early 1960s, chips were produced at the 50-micron node (0.000,05 meter). Chips in the recent iPhone12 were produced at the 5nanometer (nm, 0.000,000,005 meter) node. A transistor in this chip is about 25 atoms wide.⁹ Imagine the technological complexity in designing, producing, testing, and packaging these chips.

Semiconductor chips are made in fabrication plants, or “Fabs”. They are produced on silicon wafers. TSMC’s “GigaFABS” can produce over 100,000 300 mm diameter wafers per month.¹⁰ Remember our 88 mm² A14 Bionic iPhone chip – a “GigaFAB” can produce 80 million per month. A production cycle takes about four months, which means there are 400,000 wafers in production at any time in the Fab. There are many steps to making the chip.

At all times data are flowing to the order controlling software. Here, in this automated world we can break things down and see in one minute how much automation is occurring. For example, every minute 2.31 wafers enter production and 240 process steps complete.

Figure 2: What happens in a Gigafab minute?

The answer, of course, is every step is vital, but lithography holds the key to the technology nodes at 5nm and below. Broadly, the key steps in manufacturing are depositing layers of material onto the wafer, applying a photoresist, patterning the wafer with lithography, etching the chip features on the pattern, and implanting the silicon with additional chemical elements. There can be hundreds of process steps in the production of a chip.

Not only that, the internal layout and cleanliness of the building are vital. Even a particle of dust on the lens can be a catastrophe. It will destroy the pattern for hundreds of wafers, a very expensive mistake. For this reason, Fabs maintain extremely controlled environments with some of the highest standards of air quality, up to 10,000 times cleaner than the outside air.¹¹

Lithography in semiconductors is somewhat akin to a (complex) photographic process. To pattern the wafer with the correct design a machine called a stepper exposes each individual die (chip) on the wafer to a flash of light through a mask. The mask is the negative image of all the components of the chip and the machine steps from die to die until the whole wafer is exposed.¹² At the 7nm node, the process has become so complicated that it requires a new approach.

Figure 3: Example of an EUV Lithography machine

Tiny droplets of tin shoot from a generator at 70 meters per second. Lasers flatten the droplet and vaporize it into a plasma which emits EUV light with wavelength of 13.5nm (UV light 365nm). This needs to happen 50,000 times per second to pattern the wafer.¹³ It has taken Dutch company ASML 20 years to bring this from concept to reality.

Figure 4: EUV TWINSCAN NXE: 3400C – some interesting facts

For the equipment ecosystem EUV opens up the need for new processes, new inspection tools, re-design of chip architecture and new materials. Clearly, scaling to 5 nm and below presents challenges but it is this culture of innovation and drive to keep Moore’s Law on track that has facilitated an amazing technological progression in semiconductors. The death of Moore’s Law has long been rumored. Nothing lasts forever, but EUV is going to keep things pointing in the right direction for the next decade and will lift the whole equipment ecosystem with it.

…I realize there are questions about whether the unprecedented demand we are seeing today is secular or cyclical. When I listen to what our customers say, I hear a firm belief that the data economy is real and driving secular growth well into the future. This perspective is being reinforced by plans for substantial multiyear capital investments which are needed to support demand and fuel profitable growth...

May 20, 2021

Dan Durn, CFO of Applied Materials, Q1 2021 Investor Call

Figure 5: Top US exports in 2019

Source: Semiconductor Industry Association (2021). 2020 State of the US Semiconductor Industry Report (PDF). Accessed on 27.05.2021. Used with kind permission of SIA.

Semiconductors are a strategic asset. The US has the leading technology. It spends over 16% of semiconductor sales on research and development,¹⁴ more than any other country. Yet it has become dependent on production capacity in Asia. A recent FT¹⁵ report highlighted the US share of semiconductor manufacturing capacity fell from 37% in 1990 to 12% in 2020. At the same time, Europe declined from 44% to 9%.

We should expect to see more government incentives to invest in manufacturing facilities. Building multiple facilities in many countries is not an efficient use of capital, but as the CEO of lithography equipment maker ASML said at the Q1 2021 earnings release "[...] well there is a beneficiary of capital inefficiency, and that’s us."¹⁷

One could be forgiven for thinking that the action is all concentrated in the design and production of leading-edge chips. However, “front end” improvements need mirroring elsewhere. In particular, there is an opportunity for the “back end” to add significant value to the process. The finished chips need packaging, in a way that allows them to maximize performance when installed in a device. This requires innovative packaging solutions. Further collaboration between leading front and back-end players is likely.

A recent example is Applied Materials joint venture with advanced packaging company BESI Semiconductor. Older nodes (28nm+) have a big role to play as demand for data gathering and analysis ramps. Hence, equipment demand is likely to have strong underlying support away from the latest and greatest developments. Automated testing is also an interesting space where Teradyne is a big player. Other adjacencies in semiconductors include Electronic Design Automation companies that provide the software used to design the chips.

The hunger for computational power is nowhere near the end. A mathematical model known as the Gompertz curve suggests that we are only in the infancy of silicon transistor production. This model predicts that the rate of growth of shipments of silicon transistors will continue to increase until 2038, not reaching saturation until 2050.¹⁸ Such models should be treated with caution, but they do serve to highlight that the industry still has a lot of room to grow.

A modern smart phone has about 100,000 times the processing power of the computer that navigated the Apollo missions to the moon.¹⁹ New applications breed innovation, new companies, and new solutions. We are at such a point, with major new applications likely to emerge from the age of digitization and data generation.

As an example, many software companies are adding, compute heavy, machine learning algorithms to give computers and robots the “intelligence” to analyze huge data sets, make useful predictions such as when a key component may fail, how to route energy in a grid or recognizing different items in a picking line.

In any real time application, like autonomous driving, there is no room for data delivery or system reaction delays. Hardware across the system needs a lot more computational power and memory. This challenges both logic chip and memory chip developers to come up with new solutions. Equipment to produce whatever new chips emerge will continue to attract investment.

Our conviction in the robotics and automation theme as a long-term investment opportunity is based on the fact that we see a rising tide of innovation, an increasingly varied range of use-cases and applications, and a large number of new ventures entering the market.

Semiconductors are a more mature market but one that is critical to the development of the technologies of tomorrow. They are a key building block in robotics and automation solutions, and they are themselves produced on highly sophisticated automated production equipment.

Demand is likely to increase over time as digitalization, artificial intelligence, autonomous driving, and the industrial internet of things become embedded into society. In addition, due to the strategic importance of semiconductors today, governments are lending their support to ensure that supply is available to them, by encouraging the geographic diversification of semiconductor production sites.

Semiconductor production equipment is already a fantastically advanced example of automated production systems. As the industry pushes further into complex structures at the atomic level, we expect to see greater innovation and the production of even more powerful chips.

The individual companies mentioned on this page are meant for illustration purposes only and are not intended as a solicitation or an offer to buy or sell any interest or any investment.

To the extent that these materials contain statements about the future, such statements are forward looking and are subject to a number of risks and uncertainties and are not a guarantee of future results/performance.

1. FierceElectronics (2021). Memory is the key to future AI and ML performance. Retrieved from https://www.fierceelectronics.com/electronics/memory-key-to-future-ai-and-ml-performance, accessed on 28.05.2021
2. Media Tek, Dr. Kou-Hung Lawrence Loh, “Fertilizing AIoT from roots to leaves”, International Solid-State Circuit Conference, San Francisco, Feb 2020 cited in ASML (2021). An introduction to ASML and the Art and Science of Lithography
3. ASML (2021). An introduction to ASML and the Art and Science of Lithography (ASML, e-mail, [powerpoint], 21.05.2021)
4. Britannica (2021). Silicon. Retrieved from www.britannica.com/science/silicon, accessed on 27.05.2021
5. Nenni, D., & McLellan, P. (2013). Fabless: The Transformation of the Semiconductor Industry, p. 12-13, SemiWiki.com LLC
6. Macworld (2021). A14 Bionic FAQ: What you need to know about Apple’s 5nm processor. Retrieved from https://www.macworld.com/article/234595/a14-bionic-faq-performance-features-cpu-gpu-neural-engine.html, accessed on 27.05.2021
7. Artificial Intelligence of Things
8. Nenni, D., & McLellan, P. (2013). Fabless: The Transformation of the Semiconductor Industry, p. 13, SemiWiki.com LLC
9. Kelion, L. (2020). Apple iPhone 12: The chip advance set to make smartphones smarter. BBC News. 2020, October 13 Retrieved from https://www.bbc.com/news/technology-54510363, accessed on 16.06.2021
10. Taiwan Semiconductor Manufacturing Company (TSMC) (2021). GIGAFAB® Facilities. Retrieved from https://www.tsmc.com/english/dedicatedFoundry/manufacturing/gigafab, accessed on 16.06.2021
11. ASML (2021). How microchips are made. Retrieved from https://www.asml.com/en/technology/all-about-microchips/how-microchips-are-made, accessed on 14.06.2021
12. Nenni, D., & McLellan, P. (2013). Fabless: The Transformation of the Semiconductor Industry, p. 16, SemiWiki.com LLC
13. ASML (2021). Light and lasers. Retrieved from https://www.asml.com/en/technology/lithography-principles/light-and-lasers, accessed on 28.05.2021
14. Semiconductor Industry Association (2021). 2020 State of the US Semiconductor Industry Report. Retrieved https://www.semiconductors.org/wp-content/uploads/2020/06/2020-SIA-State-of-the-Industry-Report.pdf, accessed on 27.05.2021
15. Irwin-Hunt, A. (2021). In charts: Asia’s manufacturing dominance. Financial Times. 24.03.2021 Retrieved from https://www.ft.com/content/2b0c172b-2de9-4011-bf40-f4242f4673cc, accessed on 27.05.2021
16. The White House (2021). FACT SHEET: Securing America’s Critical Supply Chains. Retrieved from https://www.whitehouse.gov/briefing-room/statements-releases/2021/02/24/fact-sheet-securing-americas-critical-supply-chains/, accessed on May 27, 2021
17. ASML (21.04.2021). Investor Call Q1 2021 [Video File]. Retrieved from https://www.asml.com/en/investors/financial-results/q1-2021#InterviewPeterWennink, accessed on 23.06.2021
18. BTIG, e-mail, [PDF - Wally Rhines Semiconductors], 13.01.2021
19. ASML (2021). The basic of microchips. Retrieved from https://www.asml.com/en/technology/all-about-microchips/microchip-basics, accessed on 02.06.2021