Is Moore's Law dead?

David Clark simply could not throw anything away. The engineer from England hoarded all manner of junk in his attic, much to the dismay of his wife. But in 2005, he happened to read that a tech corporation was offering US$ 10,000 for a 40-year-old issue of a magazine – and the hairs on the back of his neck stood up. He had a copy of precisely this magazine!

Today, David’s magazine is no longer in a dusty attic, but in the Intel museum in California, where it is showcased like a work of art. The reason for its high value? It contains an influential essay by Gordon Moore, a co-founder of Intel.

In this essay, Moore not only predicted “home computers”, “automatic controls for automobiles” and “personal portable communications equipment”, but also prophesized that the number of transistors in computer chips would double every year up to 1975.

Moore was proven right far beyond the year 1975 – although he later adjusted the time required to double the number of components to every two years. In the industry, people therefore no longer talk about Moore’s “observation” or “forecast”, but instead “Moore’s Law”.

However, Moore’s Law has repeatedly been declared invalid as long as it has existed. One especially prominent critic was Jen-Hsun Huang, CEO and co-founder of Nvidia, who in 2022 declared: “Moore’s Law is dead!” Just a few weeks later, his company launched a new, ultra-powerful graphics card that demonstrated the exact opposite in the eyes of many.

But what’s the truth? Is Moore’s Law dead 60 years after it was proposed? Or is a popular joke from my university days more accurate: “The number of critics of Moore’s Law doubles every two years”?

Modern transistors are not bigger than the coronavirus

Moore’s Law has two aspects: the growing number of transistors and the falling cost per transistor. Initially, the number of transistors that could be obtained for a dollar doubled every two years. Today, however, it is clear that increasing complexity also has an impact on costs. The development of new chips that are smaller and more powerful is so complex that the manufacturing costs are no longer falling, but rising. 

But is the number of transistors still doubling today? 

State-of-the-art transistors are extremely small, not bigger than the coronavirus. Intel, TSMC and Samsung are currently working on new transistor technologies with numerous structural features in the single-digit nm range. However, the smaller the transistors, the more difficult it is to control them. Many layers are only a few atoms thick and here we encounter a fundamental problem of physics: quantum mechanics. To put it simply, electrons behave differently in the quantum world than in classical physics: They pass through the thin insulating layers and into the transistors. The technical term for this is quantum tunneling. As a result, the chips no longer work as precisely as before; they may lose energy or overheat.

New materials can help to minimize these tunneling effects, for example by using thicker and therefore “tunnel-proof” layer systems that do not lose performance despite their thickness. This is one of the reasons why almost 2/3 of all non-radioactive elements in the periodic table are already in use today. And the journey goes on – we continuously assesses further elements to find out whether they are suitable for improving semiconductor materials. Above all else, this work requires a great deal of precision: High-performance chips such as GPUs or AI accelerators can contain billions of transistors. To ensure that these transistors work reliably, stringent purity standards apply when manufacturing semiconductor materials. If a transistor were the size of Antarctica, only a single soccer ball would be allowed on it – a second would be deemed contamination.

In order to utilize the limited space on the chips more efficiently, many manufacturers now rely on the third dimension. The latest transistor technologies are no longer traditional “flat” 2D structures, but are slimming down more and more vertically with an ever smaller footprint as a result. However, it is not only the transistors that need to become smaller, but also their supply lines. The industry is working on moving the power supply lines from above the transistors to the level below them – underground, so to speak. This is a highly complex process to create more space for the data lines.  In 3D NAND memory technology, which we know from SSD and SD memory cards, the memory cells are stacked more than 200 times. In simple terms, memory cells used to be arranged on chips like houses in a village. They were built expansively, while making the houses smaller and smaller. In the third dimension, by contrast, the transistors are stacked on top of one another over several layers – 3D NAND is like building a city with skyscrapers. 

One step further is the so-called 3D heterogeneous integration, in which functional blocks and even small, entire chips are combined directly with one another. This means that you no longer build a single, enormous chip, but instead combine multiple small chiplets consisting of different materials and technologies. Like a castle made of Lego bricks, each chiplet has a single function but together they make up something bigger. Amazingly enough, Moore had already anticipated this in 1965: “It may prove to be more economical to build large systems out of smaller functions, which are separately packaged and interconnected”.

While this sounds simple, in reality it is highly complex. Each individual component behaves differently in terms of heat, electricity and stability. These properties must be balanced in such a way that they work together without causing problems such as overheating or short circuits.

However, unlike with Lego, you cannot simply join the individual building bricks together. The connection patterns of each chip are not the same as with Lego but are highly specific to the corresponding functions. Of course there are certain standards, but there is so much variety that almost every block has its own “fingerprint”. However, these highly complex connections pose a challenge: once connected, they can no longer be separated. If there is a defect, expensive Lego bricks are lost. With ultra-precise measuring instruments, these defects can be detected before the bricks are joined together.

Without semiconductors, our modern world would grind to a halt

Moore’s Law can be compared to climbing Mount Everest: The higher we climb, the more difficult the next step becomes. The question is therefore not whether Moore’s Law still exists, but rather how much we are willing to spend to maintain it.

This question should not be underestimated; after all, without semiconductors our entire global economy would grind to a halt. Anybody who doubts this simply has to remember that semiconductors are not only fitted in computers, smartphones and cars, but also in pacemakers, MRI scanners, airplanes, aircraft navigation systems, LED lamps, refrigerators, and credit cards.

Or in other words, in almost everything we use every day.

Let’s come back to the initial question: Is Moore’s Law dead? From a pure cost perspective certainly. This does not mean that no added value is being created, but the days of “more transistors for free” are over. 

In terms of the increasing number of transistors, Moore’s Law is alive and kicking. We are still climbing the peak, but it is getting steeper and the air thinner. After all, one thing is clear: If we are to continue, we need ever more complex and ingenious climbing equipment. In doing so, we can rely on the innovative strength of the industry. In the past, we have found ways to scale transistors and make chips smaller, and I am confident that we will continue to find solutions in the future. Or, to conclude with a quote also attributed to Gordon Moore: “If the auto industry advanced as rapidly as the semiconductor industry, a Rolls Royce would get half a million miles per gallon, and it would be cheaper to throw it away than to park it.”