Deep (Semi) Takes: Moore's Law vs. Moore's Precept
A Deep Takes Welcome and Moore's Law vs. Moore's Precept
Welcome to Deep (Tech) Takes! I’ll write a longer post about why you should subscribe. But if I had to make a 10-second pitch: The deep tech industry can no longer afford to stay in the shadows. The industry’s influence goes beyond just creating slimmer phones. In 2024, deep tech took a significant role in protecting our country from malicious adversaries and ensuring abundant and affordable energy. The Deep (Tech) Takes blog will share my deep takes on the entrepreneurs, scientists, engineers, investors, and policymakers going all-in on deep tech.
In today’s first Deep Takes post, I want you to understand the difference between Moore’s Law and Precept. My next post will discuss how 2D materials will shape the future of the transistor in the semiconductor industry.
Moore’s Law vs Moore’s Precept
Moore's Law is dead. False. Moore’s Precept is what is in danger of dying. How can we keep it alive?
In 1965, the original Moore's Law stated that the number of transistors in an integrated circuit doubled yearly. However, Gordon Moore revised his prediction a decade later and said the number would double every two years. It is important to note that this law is not a fundamental law of physics (like the laws of thermodynamics) but an empirical observation that the world’s semiconductor ecosystem races to meet every few years.
Moore’s Law has come a long way since 1965 and has evolved into Moore’s Precept. Moore’s Precept, first coined by Stratechery’s Ben Thompson, states that computing power will increase and become cheaper over time. The precept is an adaptation of Moore’s Law with the added price component.
Moore’s Precept helps explain why foundries and their leading suppliers are in such a cutthroat business. Foundries need to double the density of their chips (or million transistors/mm) faster than their competitors and do it economically to have a chance to stay afloat. This is why foundries care so much about increasing their yields (number of dies per wafer) and equipment productivity (number of wafers processed per hour per equipment). It also explains why Intel is in such a predicament today. Intel’s 18A node is arguably coming closer to fabricating chips with transistor performance and densities similar to TSMC. However, they are still at least two to three years behind in producing them economically with extremely high yields.
We see both Moore’s Law and Precept in action when we look closer at TSMC’s transistor density by node size and TSMC’s price per billion transistors provided by Stratechery. The transistor density in MTr/mm has roughly doubled since 2004, with every new node introduced on average every two years. The graph shows that Moore’s Law is alive and healthy. However, if you look at the bar graph that graphs costs in Price/BTr, costs decreased only modestly or increased from 2019 to 2020. The increase represents TSMC’s transition to ASML’s extreme EUV lithography machines that cost about $150 million each.
Moore’s Precept is at risk of stagnating. But why is it so important to keep it alive? You don't care about transistor density counts if you don’t work in a chip company. Duh. But you still want that iPhone, Chromebook, or Quest 3S to perform even faster, have longer battery life, and have more storage.
If you work in a high-tech industry (many readers of this blog), you also want Moore’s Precept alive for as long as possible. The law continues to improve our electronics, computers, and data centers, which are integral in any technical field you choose, such as medical imaging, robotics, bioinformatics, autonomous driving, etc. It amazes me that CPUs, GPUs, ASICs, and HBMs (High Bandwidth memory) can improve so much in just a few years. These chips are the core building blocks that power our noise-canceling headphones, data centers, AR and AI glasses and headsets, and self-driving electric cars, and they will only become more integral.
Moore’s Precept also helps drive the cost of computational power down, inevitably making the technology more accessible and affordable. If the Precept were to die, say hello to increasing prices when we buy personal electronics.
Finally, maintaining the precept is critical to the US in a new Cold War 2.0 with rising superpowers. One of the reasons the US defeated the Soviet Union during Cold War 1.0 was its ability to develop chips that outperformed the Soviets. As Chris Miller writes in the book Chip War, even with advanced espionage techniques, the Soviets could not build advanced foundries that came close to competing with US foundries. The US is in similar technological brinkmanship with China, and whichever nation can economically produce the world’s best-performing chips will be at a considerable advantage.
What are we doing to keep Moore's Precept Alive? My next Deep (Semiconductor) Takes blog post will talk about the changing architecture of transistors and 2D materials that can hold some of the answers.