ASML: Innovating at the Speed of Light
By: Leisen Liu & Desmond Li
The Ivey Business Review is a student publication conceived, designed and managed by Honors Business Administration students at the Ivey Business School.
The Moore the Better
Moore’s Law states the number and speed of transistors on a microchip double roughly every two years. However, this historic pace of processing power growth is slowing down. In 2019, Intel’s 10-nanometer (nm) chips were delivered, a full five years after the previous generation of 14nm chips. These measurements approximate the size of one transistor in a microchip — smaller transistors can be packed more densely into a microchip of a given size. For reference, human hair is 80,000 to 100,000nm wide and human DNA is 2.5nm in diameter. Economists at Stanford and MIT determined the research effort required to uphold Moore’s Law has risen by a factor of 18 since 1971. Computational progress will soon reach a point where marginal improvements will be unable to recoup ballooning R&D costs. However, since smaller transistors are critical to reducing the energy, cost, and time required for electronic computations, a slowdown in progress jeopardizes long-term growth for the semiconductor industry. With the increasing adoption of computationally intensive applications like artificial intelligence (AI), processing needs will quickly outpace the power provided by the current generation of chip architecture.
An analysis conducted by OpenAI found the computational power required to train the largest AI models has been doubling every 3.4 months. This trend is likely to continue, with several microchip startups developing AI-specific chips, signalling the birth of a new industry segment where players are racing to find a profitable niche or be acquired. The greatest limitation to this trend is cost, with AI training sessions requiring immense processing power and millions of dollars worth of hardware. With sufficient economic incentives, larger and more frequent training runs will drive the increase in power requirements. This predicted growth signals a need for improved AI chips; however, these chips come at a cost.
A Chip on the Environment’s Shoulder
Coupled with the increased demand for processing power is a tremendous surge in energy usage. Researchers at the University of Massachusetts found that the process of developing and training a research-worthy AI model emits more than 78,000 pounds of carbon dioxide, nearly the equivalent of 40 round-trip flights between New York and San Francisco for one individual. This comes at a time where most industries are making commitments to reduce their carbon emissions in alignment with net-zero emissions timelines. These environmental factors compound the demand for energy-efficient chips.
The semiconductor industry now faces a dilemma. Existing chip hardware will soon be unable to meet processing power requirements and will have to address the environmental costs associated with higher processing power. Without addressing these issues, existing microchips will quickly become obsolescent. The industry must develop a solution for the next generation of microchips that pushes the boundaries of Moore’s Law, satisfies processing needs, and balances the environmental costs.
All About ASML
Founded in 1984, ASML Holding (ASML) is a global supplier in the semiconductor industry. ASML’s core revenue streams consist of the development, production, and maintenance of extreme ultraviolet (EUV) photolithography machines that produce microchips. Using pizzas as an analogy for the semiconductor industry, foundries such as TSMC and GlobalFoundries are the chefs that bake the pizzas, the recipes for different pizzas are written by companies like AMD and Apple, while the pizza ovens are designed, built, and maintained by ASML.
Engage Against the Machine
Each machine is highly-specialized, consisting of over 100,000 globally-sourced parts. ASML’s innovation and introduction of EUV machines revolutionized the industry, enabling patterns to be printed with greater precision utilizing wavelengths of only 13.5 nm, a 14x reduction compared to the previous generation of 193 nm wavelengths. With processing power dictated by transistor size, and new electronic devices using 7 nm chips, only ASML’s EUV machines are capable of producing chips at such a small scale, driving their dominance in the photolithography space. This was made possible by ASML prioritizing cooperation and partnerships with key industry players in the earliest stages of the development of EUV machines. ASML maintains collaborative R&D as a core pillar of the company’s strategy, investing 2.2 billion euros in R&D in 2020, representing 15.7 percent of total sales. ASML’s dedicated research department’s focus is to not only generate and explore ideas internally, but to also search for technological solutions for ASML’s products and applications within the semiconductor industry and beyond. These technological innovations are integrated with ASML’s systems or used to develop new applications, to drive forward ASML’s machines and the industry.
Conducting Energy-Efficient Chips
However, while pushing the boundaries of Moore’s Law, ASML is still making strides to reduce its greenhouse gas emissions by enhancing the energy efficiency of its products. At the front end of the semiconductor value chain, ASML’s innovative lithography systems enable their customers to design and manufacture more powerful chips that consume less energy. A major challenge for ASML moving forward will be to meet customers’ expectations of increasing product performance while simultaneously reducing energy consumption.
As a Light-matter of Fact
To strategically position itself against the slowdown of Moore’s Law, ASML should acquire Lightmatter. Founded in 2018 by two MIT graduates, Lightmatter is a startup that makes photonic chips. The company has raised $113 million in funding from companies such as Google Ventures and prominent venture-capital firms Spark Capital and Matrix Partners. In doing so, ASML gains the necessary internal capabilities to develop photolithography machines capable of manufacturing photonic microchips. Using Lightmatter’s innovative technology as a foundation, ASML will be able to develop machines capable of producing microchips that meet the growing computational requirements of end consumers while improving energy efficiency.
Case Studies: M&A-SML
ASML has successfully completed several recent M&A transactions to fuel growth and technological advancement to complement their current product offerings. In 2013, Cymer, a manufacturer of light sources, was acquired to accelerate the development of EUV source technology. This acquisition allowed ASML to solve several technical challenges, allowing EUV infrastructure to meet customers’ high volume manufacturing requirements, ultimately enabling ASML to become the world’s only manufacturer of EUV lithography systems. In 2017, a 24.9 percent interest in Carl Zeiss was acquired to further the development of ASML’s EUV systems and align both companies’ long-term roadmaps, including the development of ASML’s next-generation High-NA photolithography machine. Most recently in 2020, ASML acquired Berliner Glas Group, one of the world’s leading providers of optical components and high-quality refined technical glass. This acquisition provided ASML with the technical capabilities crucial to securing the future roadmap for EUV and DUV products.
Lightmatter as a Heavyweight
Lightmatter’s chips differ fundamentally from traditional microchips. To perform logical operations, traditional microchips utilize silicon transistors to manipulate the flow of electrons, and process information by reducing electrical pulses to binary signals through copper circuits. Contrastingly, Lighmatter’s chips use photonic circuits to perform calculations by manipulating the path of light. Lightmatter’s photonic chips are meant to perform calculations specifically for running AI programs. Lightmatter’s internal testing has shown that its chips are able to run 5x faster than Nvidia’s leading A100 chip on the BERT natural language processing model, all while using 85 percent less energy. This is because information can be encoded in different wavelengths of light, and manipulating light requires less energy than manipulating electrons through transistors.
Lightmatter has a long path ahead in order to solidify its photonic chips as the universal answer to the increasing demand for AI. To produce a chip using photons rather than electrons requires a completely different skill set and a complete retooling of development practices. Nick Harris, CEO of Lightmatter indicates the company plans to use its latest April 2021 funding round to fuel the production of hardware and sell it by next year. Lightmatter plans to outsource its chip manufacturing to GlobalFoundries, a leading American foundry which produces other companies’ chip designs. Getting a production device into the market is key for Lightmatter to convince skeptics of a cutting-edge design and extend its first mover advantage over potential competitors. Lightmatter has taken steps to help itself cross the chasm in the technology adoption lifecycle. Its server offerings fit into standard data center racks and support PyTorch, TensorFlow, and Onyx — languages and formats used to build neural networks.
A Merger that Matters
The fundamental decision in M&A is whether to develop something internally or purchase it from an external provider. Lightmatter’s photonic chip architecture differs from ASML’s chip manufacturing infrastructure, requiring a complete product offering overhaul. In addition, lengthy development time, extensive R&D costs, and a potential rise in competition creates an urgency for an acquisition rather than internal development. Photonic microchips are in an infant stage of the adoption lifecycle and are poised to become the industry’s dominant technology in the AI computing space. ASML, once again, possesses the opportunity to establish a first-mover advantage in an emerging technology by acquiring Lightmatter’s IP, just as ASML had similarly capitalized on EUV technology. By acquiring Lightmatter, ASML’s R&D department can optimize Lightmatter’s chip designs for mass production and make headway towards widespread adoption. ASML’s substantive record of successful acquisitions is a strong indication of the company’s ability to integrate Lightmatter, which in turn can fuel long-term growth and the technological advancement of photolithography machines. Beyond acquisitions, ASML’s financial resources, expertise in manufacturing complex systems, network of business and technology partners, and experience in successfully supporting startups will play a considerable role in propelling Lightmatter’s offering to new heights.
Continuous innovation and R&D has been a key success factor for ASML, enabling the development and commercialisation of EUV photolithography machines. However, as a result of growing demand for superior chip efficiency, ASML’s dominant market share is threatened by competitors who are able to capitalize on the transition to photonic microchips. The acquisition of Lightmatter will allow ASML to maintain dominance in the semiconductor industry and continue to innovate at the speed of light.