Dec 27, 2021 7:00 AM

The Year Everyone Remembered That Chips Matter

Shortages of semiconductors messed up production of everything from cars to toys. Plans are in the works for more factories, but they'll take years.

A collage with an electronic chip and art by Jenny Sharaf

The most important technology of the year was not Mark Zuckerberg’s Metaverse, Jack Dorsey’s blockchain, or Elon Musk’s, err, dancing robot.

It was more likely the same thing that has propelled progress in the tech industry for decades. The one that lets machines juggle and manipulate information, faster and more efficiently every year. It is, of course, the silicon chip.

The importance of semiconductors may have faded from view over the last decade as the web, social media, and apps came to the fore. Silicon Valley is arguably now more synonymous with Google’s inescapable web search, Amazon’s ecommerce empire, or Facebook’s FOMO-fueled feed than with Intel’s newest chip. But the past year has provided plenty of evidence that chips are, in fact, more important than ever.

Rising demand for chips from new types of customers combined with the pandemic and geopolitical tensions to put extraordinary pressure on the supply of both simple and advanced computer processors in 2021, resulting in shortages of everything from cars to game consoles. Control over the production of advanced silicon is shaping competition and conflict between the world’s two dominant superpowers. And many governments are set to pour huge sums into bolstering their manufacturing capabilities.

Custom chip designs are rapidly becoming increasingly vital to emerging fields such as artificial intelligence, self-driving cars, and 5G. These specialized architectures, rather than general purpose chips, will help shape the course of innovation.

“There is really nothing humble about the silicon chip,” says Jésus del Alamo, a professor at MIT who teaches courses on advanced microelectronics. “They are at the heart of everything, and do so many critical things in so many valuable places, with incredible benefits to society.”

Del Alamo notes that chips are now found in a bewildering array of products, making kitchen appliances, industrial machinery, and even objects as mundane as light bulbs connected and programmable.

The pandemic quickly revealed how vital chips have become to the economy. When carmakers shut factories in early 2020, in anticipation of an economic slowdown, they canceled orders for low-cost chips increasingly needed in engines, safety systems, and infotainment displays. Even the most basic gas-powered car now has over 100 chips, while the latest electric vehicle may hold more than 1,000. When car buying recovered unexpectedly, manufacturers didn’t have enough silicon, and were forced to halt production, dampening the economic recovery in many nations.

The computerization of everything from industrial robots to medical devices meant the chip shortage was felt far and wide. Meanwhile, the huge cost of building new chip plants, and the cyclical nature of the industry, has caused the drought to drag on … and on. “Companies that considered chips as just another bullet item on a bill of materials now realize the relevance of semiconductors,” says Gaurav Gupta, a vice president at Gartner, a research firm, who tracks electronics. “Now everyone has to focus on, and strategize, the procurement of semiconductors.”

The chip crunch has been exacerbated by tension between the US and China. The Trump administration barred the sale of the most advanced computer chips—those powering the latest smartphones or cloud servers—to Chinese companies accused of close ties to the government or of aiding in human rights abuses of Muslims in Xinjiang province.

The blockade, which the Biden administration has kept in place, also reflects an awareness that these chips are crucial for progress in fields such as AI, 5G, and robotics, which are increasingly seen as key to economic and military competition. And because China lacks the capacity to manufacture the most advanced chips for itself, leading tech companies such as Huawei, which once led the world in smartphone sales, have seen part of their business crippled. Some Chinese firms reportedly stockpiled chips in anticipation of the ban, stretching available supplies still further.

The importance of advanced chip manufacturing has thrust some lesser-known companies into the geopolitical spat. The Dutch firm ASML is the world’s only maker of the $150-million extreme ultraviolet lithography machines needed to etch the most miniscule features into silicon chips. The US also has blocked ASML from exporting to China, effectively hobbling the country’s domestic chip industry.

But Washington’s chip gambit also reflects America’s weaknesses. In recent decades it has become incredibly complex and expensive to manufacture the fastest and most efficient chips, which feature atomic-scale features. Only three companies in the world are now capable of producing the most advanced components: TSMC in Taiwan, Samsung in South Korea, and Intel in the US. But Intel has fallen behind TSMC and Samsung, and the US share of global chipmaking has slipped to an estimated 12 percent this year, from 37 percent in 1990, according to the Semiconductor Industry Association (SIA), an industry group.

In June the US Senate approved a bill with $52 billion aimed at reviving the US semiconductor industry; it is awaiting action by the House. John Neuffer, CEO of the SIA, which is backed by US chip companies, says the money would help the US regain its edge and also ensure a robust supply of less advanced chips for the automotive industry. “Inaction is not an option” in the face of current trends, says Neuffer.

Other nations are likely to invest similarly in an effort to prop up their chip industries. South Korea says it will invest more than $55 billion over the next three years to update and expand its manufacturing. European Union nations plan to invest $25 billion to $35 billion and will introduce tax and other incentives in an effort to snare 20 percent of global chipmaking by 2030. The Chinese government has signalled that it will spend $150 billion over the next 10 years in an effort to achieve chip manufacturing self-sufficiency.

Kevin Krewell, an analyst with Tirias Research who tracks the industry, notes that not all of the money will be invested in cutting edge chip making capacity, and shortages could still stretch into 2023 and 2024. “Demand shows no sign of slowing as cars get smarter with even more electronics and consumer devices keep growing in sophistication and number,” he says.

One reason it is so vital for the US to be able to make cutting-edge chips is that progress in chip manufacturing has begun slowing in recent years, making custom silicon a lot more important. For decades, the number of transistors on a chip, which roughly corresponds to its speed, doubled about every 18 months, an observation first made by Intel cofounder Gordon Moore; that’s no longer happening.

Custom chips offer a way to boost performance. Apple, for instance, now designs its own chips for its smartphones, tablets, and laptops, as part of a strategy that gives it greater control over the performance of its products. It represents a striking turnaround for an industry where once only chip giants like Intel and AMD designed and made silicon. Now, a growing number of companies see specialized chip designs, including their own, as a way to gain an edge in areas like AI.

The chip at the heart of most computers, known as the central processing unit (CPU), is designed to be flexible enough to carry out any logical operation. But AI algorithms increasingly rely on more specialized chips, mostly the graphics processing units (GPUs) originally developed for video gaming, because they are well-suited to running the necessary mathematical calculations.

As AI becomes more important, and as algorithms become larger and more expensive to run, many companies are exploring customized designs. Amazon Web Services (AWS), the world’s largest provider of cloud computing services, has developed several chips specialized for AI, including Trainium, for training AI models and Inferentia, for running pre-trained models.

“GPUs in general are intended to solve a whole bunch of workloads,” says Raj Pai, a vice president at AWS involved with the company’s chip efforts. “With our own custom silicon we are able to really look at the sorts of models that our customers run and optimize the chip and the power and everything.”

Other companies now designing their own chips to accelerate AI include Google, Facebook, Microsoft, and Tesla.

Neil Thompson, a professor at MIT who studies technological change, says the trend “is a very natural reaction to the slowing of Moore's law.” But he also believes it could lead to less innovation in algorithms over time, as tech becomes more “siloed.”

Now, AI is also helping design chips, making it easier for others to join the parade. Bill Dally, chief scientist at Nvidia, which became the world’s most valuable chipmaker by selling its GPUs to AI engineers, says the company is using AI to design chips that are faster or more efficient than ones designed by hand. The technology will speed up chip design and lead to new designs, he says. “AI will have a profound impact on the process of chip design,” Dally says.