How long will Moore’s Law last?

Moore's Law is a prediction made by Gordon Moore, Co-Founder of Intel, in 1965. He observed that the number of transistors on a microchip doubled approximately every year, leading to a corresponding increase in computing power and a decrease in cost per transistor. This trend has driven the exponential growth in the performance and capabilities of electronic devices, particularly computers, over several decades. In 1975, Moore revised his observation to predict that the number of transistors would double approximately every two years.

History of Moore's Law

In 1965, Gordon Moore published a paper titled "Cramming More Components onto Integrated Circuits" in Electronics Magazine. In this paper, he noted that the number of components (transistors, resistors, capacitors) on integrated circuits had roughly doubled each year since the invention of the integrated circuit in 1958, and he projected that this trend would continue for at least another decade. This observation became known as Moore's Law.

During the 1970s and 1980s, Moore's Law held true, with companies like Intel and IBM continuing to innovate and shrink transistor sizes. This era saw the development of the microprocessor, which revolutionised computing. For example, Intel's 4004 processor, released in 1971, contained 2,300 transistors, while the Intel 80386, released in 1985, contained 275,000 transistors.

As the industry moved into the 1990s, Moore's Law continued to predict the pace of advancement. By the mid-1990s, process nodes had reached the 350nm range, and by the end of the decade, they were at 180nm. The number of transistors on microprocessors like the Intel Pentium series grew from millions to tens of millions.

As transistor sizes shrank further into the sub-100nm range in the 2000s, challenges began to emerge. Issues like heat dissipation, power leakage, and quantum effects became more pronounced. Despite these challenges, companies continued to innovate with new materials, such as strained silicon, and new techniques, such as high-k dielectrics, to keep up with Moore's Law.

By the 2010s, the industry was reaching the physical limits of silicon transistor scaling. Process nodes had reached 10nm and below, but the performance gains from each new generation were becoming less significant. Additionally, the cost of manufacturing at these advanced nodes increased dramatically. Companies began to focus more on architectural improvements, multi-core processors, and specialised processors (like GPUs) to continue improving performance.

Examples of Moore’s Law in action

To put in perspective the impact of a doubling each year, here are a few time stamps and the corresponding processor/number of transistors:

• Intel 4004 (1971) – The world’s first microprocessor, containing 2,300 transistors and built on a 10µm (10,000nm) process. It marked the beginning of the microprocessor revolution.
• Intel 8086 (1978) – The processor that formed the basis for the x86 architecture, which remains dominant in PCs today. It had 29,000 transistors and was built on a 3µm process.
• Intel Pentium (1993) – This processor had 3.1 million transistors and was built on a 0.8µm (800nm) process. It represented a significant leap in computing power, supporting features like superscalar architecture and floating-point unit integration.
• Intel Core i7 (2008) – Part of Intel’s Nehalem architecture, the Core i7-920 had 731 million transistors and was built on a 45nm process. It featured an integrated memory controller and support for hyper-threading.
• Apple A14 Bionic (2020) – This chip, which was used in the iPhone 12, was built on a 5nm process and contained approximately 11.8 billion transistors. It exemplified the continuation of Moore’s Law into the 2020s, albeit with diminishing returns in terms of raw transistor count translating into performance gains.

Terminology in the modern era

In order to understand the trends of Moore’s Law in the modern era, it is first important to understand modern terminology to describe the next generation of chips.

In semiconductor manufacturing, terms like "10nm" and lower (such as 7nm, 5nm, etc.) refer to the process nodes used in the production of integrated circuits (ICs). Historically, these terms were directly related to specific physical dimensions, such as the gate length of transistors. However, as technology advanced, these labels became more of a marketing term than a precise measurement of any physical feature size.

For modern semiconductor processes, the "nm" designation does not correspond to any specific dimension within the chip. Instead, it reflects a general level of technology and performance improvements that come with each new generation. These improvements might include:

• Higher transistor density: More transistors can be packed into the same area, allowing for more complex and powerful chips.
• Lower power consumption: Advances in materials and design techniques reduce the amount of power each transistor consumes, leading to more energy-efficient devices.
• Higher performance: Transistors can switch faster, improving the overall speed of the chip.

The reason why these terms no longer correspond to actual sizes is due to the complex nature of modern transistor designs. Different parts of the transistor (like the gate, channel, and contacts) might have different dimensions, and advancements in other areas, such as materials or 3D stacking of transistors, also contribute to the overall performance improvements.

Thus, "10nm" and similar terms are more like a shorthand for a set of technological advancements and characteristics rather than a literal measurement of any feature on the chip.

Is Moore’s Law dead?

Not exactly, but it is reaching its physical and economic limits. The traditional interpretation of Moore's Law is becoming increasingly difficult to sustain due to several factors:

• Physical limits: As transistor sizes shrink to the nanometre scale (e.g., 5nm, 3nm), fundamental physical challenges, such as quantum tunnelling and increased heat dissipation, make further miniaturisation difficult. These challenges result in diminishing returns in terms of performance and power efficiency.
• Economic factors: The cost of developing and manufacturing at the cutting-edge process nodes has skyrocketed. The research, development, and equipment required for each new generation are becoming prohibitively expensive for many companies.
• Diminishing benefits: As process nodes shrink, the performance improvements and power savings have become less pronounced. The industry has seen a shift towards optimising other aspects of computing, such as architecture, software, and specialised processors (e.g., GPUs, TPUs), rather than relying solely on transistor density.

While traditional scaling is slowing, it does not mean innovation in semiconductor technology has stopped. The industry is exploring several alternatives and complementary approaches:

• 3D chip stacking: Instead of just shrinking transistors in 2D, stacking multiple layers of transistors can increase density and performance.
• Heterogeneous integration: Combining different types of chips or materials on a single package can improve performance without relying on further transistor scaling.
• New materials: Exploring alternatives to silicon, such as graphene or carbon nanotubes, could provide new avenues for miniaturisation and performance gains.
• Quantum computing: Although still in its infancy, quantum computing represents a potential future paradigm that could outpace classical computing in certain tasks.

How long will it last?

It's challenging to predict an exact timeline, but many experts believe that traditional Moore's Law scaling will continue to slow over the next decade, with 2036 being the year traditional approaches collapse. Beyond that, any further advancements will likely rely on alternative approaches, such as those mentioned above, rather than pure transistor scaling.

In summary, while the classical interpretation of Moore's Law is nearing its limits, the semiconductor industry is still innovating and finding new ways to continue improving computing performance. Moore's Law, in its traditional sense, may not ‘die’ but rather evolve into a new paradigm that incorporates a broader range of technologies and strategies.