It’s been a frenetic month of news in the semiconductor industry, following a flurry of regulatory activity in Washington aimed at boosting American chipmaking capacity and stemming the flow of advanced chips to China. 

As technology competition heats up, policymakers and observers even beyond the chip sector must familiarize themselves with an expanding vocabulary of chip-related concepts. Electronic design automation software, chips for AI, 12 nanometer vs 7 nanometer: what does it all mean, and how does it affect U.S.-China relations? 

This week, The Wire explains key concepts in the semiconductor supply chain and what recent events mean in the competition over its control. 

ON NODES AND NANOMETERS

An integrated circuit — also known as a chip — is a device composed of billions of transistors: computational devices that can be switched on (1) and off (0), which together are able to perform calculations at remarkable speeds. 

Listen to experts compare chips and you will hear plenty about nanometers. This refers to the microscopic distance between individual transistors on a chip — just nanometers (nm) apart, or less than one-ten thousandth of the width of a human hair — with smaller distances allowing transistors to be packed more densely, increasing computing power. 

Marketing and discrepancies between foundries mean the numbers (or nodes) have lost their literal meaning, but by convention firms still refer to their most advanced chips as 14nm, 10nm, 7nm and so on. 

Companies at the forefront of chip development are largely located in Asia. Taiwan’s TSMC and South Korea’s Samsung offer 5nm chips, which Chinese producers have struggled to match. Part of what’s keeping China back is export restrictions on “extreme ultraviolet” (EUV) lithography machines, a critical piece of hardware made by Dutch firm ASML which is needed to produce 7nm chips and smaller at scale.¹Officials with the Trump administration lobbied heavily around 2020 for the Netherlands to block the sale of EUV lithography machines to China. The CHIPS Act, passed in August, forbids recipients of federal funding from investing in new plants in China that produce chips with nodes 28nm or smaller. 

That’s why the news last month that Chinese state-backed chipmaker SMIC had shipped a batch of 7nm chips came as a surprise: Many policymakers had assumed SMIC lacked the knowhow and hardware to produce such advanced chips. Some experts argue that SMIC’s reliance on older technology (deep ultraviolet, or DUV, lithography machines) may still prevent it from cost effectively producing 7nm chips at scale.²U.S. regulators are reportedly now also lobbying their Dutch counterparts to restrict the sale of DUV machines to China.

AIMING FOR AI

A major reason countries are eager to develop advanced chips is for artificial intelligence (AI). “AI chips” are hugely powerful integrated circuits, capable of performing demanding calculations. The process of ‘training’ a leading AI algorithm can require a month of computing time and cost tens of millions of dollars, according to an analysis by the Center for Security and Emerging Technology (CSET), a Washington D.C.-based think tank. It estimates that only nodes of 12nm and below are suitable to train and run the most complex neural networks: Older chips would require orders of magnitude more time, electricity and money. 

China is a major buyer of AI chips, which are coveted by its consumer tech companies and research institutions, but also by its military, where applications include image recognition for scanning satellite imagery for weapons, or surveillance, such as by scanning audio communications for intelligence.

Almost all the AI chips that the People’s Liberation Army (PLA) acquired between March and December 2020 came from American companies — namely Nvidia, Xilinx (now part of AMD), Intel and Microsemi — according to a CSET analysis of its procurement data. In each case, the PLA acquired those chips via Chinese intermediaries, underscoring the shortcomings of U.S. export controls focused on military end-users. The records also revealed not a single contract for Chinese-made AI chips, highlighting the PLA’s dependence on superior U.S.-designed technology. 

Last week, the U.S. government introduced new export restrictions on the sale of high-end AI chips to China and Russia. Nvidia, which is on the verge of releasing a groundbreaking new AI chip called “Hopper,” has warned the ban could cost the firm $400 million in sales this quarter; AMD says it does not expect the restrictions to have a material impact.³Hopper, also known as H100, is designed by Nvidia but fabricated by TSMC using its 4nm process node.

Chipmakers are among the most dependent on China for their sales among U.S. multinationals that disclose their regional revenue breakdowns.

NOT JUST HARDWARE

While Asian economies dominate in chip fabrication, American firms lead in chip design and related tools. Three U.S.-based companies — Synopsys, Cadence and Mentor Graphics — dominate the market for electronic design automation (EDA) software, which integrated circuit designers use to design chips.⁴Mentor Graphics was acquired by Germany’s Siemens in 2017.

Beijing identified the development of China’s own EDA software as a priority in its latest five year plan. Several veteran engineers from Cadence and Synopsys have founded or joined Chinese EDA startups in recent years.

Last month, the Commerce Department announced new restrictions on the sale of EDA software which can be used to design certain high-end chips, in line with an agreement between 42 countries. The new export control affects over 150 countries, including China. But some experts have questioned the degree of consensus among participating countries to the export control agreement, which includes most of Europe as well as Russia and India. 

The U.K. government’s move last month to block the Chinese acquisition of a British EDA software company shows that other countries are taking note. Then-business secretary Kwasi Kwarteng — who is now Britain’s finance minister — blocked the takeover of EDA software provider Pulsic Ltd. by a Hong Kong holding firm controlled by Shanghai-based UniVista Industrial Software Group, a Chinese government-backed two year-old EDA software firm that is partly financed by China’s national chip investment fund, also known as the “Big Fund.” 

---