Whether it’s a CPU, GPU, or flash memory, the process technology is a core factor that determines their performance. From the surface numbers, a smaller process number indicates more advanced technology, allowing more transistors to be packed into a smaller space, reducing leakage, improving overall chip performance, and lowering power consumption. However, due to the lack of a unified standard for process technology, the numbers often have discrepancies.

For example, Intel has been stuck at 14nm for 3 years, while Samsung and TSMC have already moved into the 10nm era. So, does that mean their 10nm technology is necessarily better than Intel’s 14nm?

A9 Processor Version Dispute

Let’s recall the version dispute of the A9 processor in the iPhone 6s in 2015. To ensure stable supply, Apple simultaneously partnered with Samsung (14nm) and TSMC (16nm) to manufacture the A9 processor. Theoretically, the A9 produced by Samsung should have better power efficiency because 14nm is superior to 16nm.

Surprisingly, numerous user feedback and tests revealed that the A9 produced by TSMC was actually better than the one produced by Samsung, with longer battery life.

The reason is simple: although 14nm and 16nm differ by 2nm, they both belong to the same generation of technology, and there is no claim that 14nm is more advanced than 16nm. Additionally, the process used by TSMC for the A9 is the second generation 16nm FinFET Plus, while Samsung’s 14nm FinFET LPE is the first generation 14nm process, making the more mature 16nm FinFET Plus naturally superior.

Who Sets the Standards?

Careful readers may have noticed that Samsung jumped from 28nm directly to 14nm for mobile processors. In contrast, TSMC launched a 20nm process after 28nm, but it was only adopted by Qualcomm Snapdragon 810 and MediaTek Helio X20, and quickly transitioned to 16nm.

In other words, there is no clear iterative relationship in the process nodes launched by Samsung and TSMC, giving the impression of a sudden leap forward.

In contrast, Intel has consistently adhered to Moore’s Law: each generation of process technology must double the number of transistors on the chip. Intel always follows a naming convention that reduces the previous generation’s process size by about 0.7 times, from 90nm→65nm→45nm→32nm→22nm→14nm→10nm→7nm, with each generation accommodating twice as many transistors per unit area as the previous generation.

Therefore, Intel executives have repeatedly stated that further miniaturization of processes is becoming increasingly difficult, and some companies are beginning to deviate from the principles of Moore’s Law, even if the transistor density increases little or not at all, they continue to name new process nodes. Indeed, Intel is subtly criticizing Samsung and TSMC, whose 16nm, 14nm, and 10nm cannot accurately reflect their positions on the Moore’s Law curve!

The Secrets Behind the Numbers

So, how can we calculate the standard transistor density metrics? Intel previously dismissed the traditional calculation formulas of “gate pitch × minimum metal pitch” and “gate pitch × logic cell height” because they do not truly measure the actual transistor density achieved, nor do they attempt to quantify the relative density of different types of logic cells in the design library compared to the previous generation.

To this end, Intel began advocating a previously popular but once “disgraced” calculation formula, which is based on the transistor density of standard logic cells and includes multiple weighting factors that determine typical designs.

According to this formula, although Intel’s technology seems “behind” its competitors at face value, it can achieve a leapfrog performance.

Simply put, in core parameters such as fin pitch, gate pitch, minimum metal pitch, and logic cell area, even the first generation 14nm process launched by Intel in 2014 can perform comparably to competitors’ 10nm technology.

In other words, even if competitors inflate their process names, the latest processes have only just reached the level of Intel’s technology from three years ago. It is important to note that Intel’s 14nm has now evolved to 14nm++, and the latest 10nm technology is also about to be launched.

How much of a gap does Intel’s 10nm have compared to its competitors? According to Intel’s comparative parameters, in all projects, Intel’s 10nm is far ahead of TSMC and Samsung’s 10nm, and can directly compete with the 7nm technology that competitors will mass produce next year.

With Intel’s subsequent launch of 10nm+ and 10nm++ optimized versions, they may even rival competitors’ 5nm technology.

Intel’s leading position in technology is also evidenced by the latest materials and technologies used. For example, Intel introduced strained silicon as early as the 90nm era, while competitors only introduced it from the 65nm period; Intel brought high-K metal gates at 45nm, while competitors only started experimenting with it at 28nm…

In translation, Intel wants to emphasize that its process technology is the industry standard, while others may have inflated claims, urging consumers not to be easily misled.

Shifting Learning Costs to Consumers

Although industry insiders know that Intel’s technology is unparalleled, and players with a certain foundation can discern the strengths and weaknesses of various companies’ technologies through formulas and data, for the vast majority of ordinary consumers, complex formulas and lengthy explanations are too esoteric; we only know that 10m is better than 14nm, and 7nm is the next generation of 10nm. Conventionally, newer processes are seen as superior.

Furthermore, Intel’s leading technology has long been used solely for producing X86 processors. In contrast, competitors’ technologies are primarily used to design ARM processors, and there is no direct competition, making it difficult to compare final products.

Therefore, compared to TSMC and Samsung, who update their processes almost every two years (even if just a numbers game), Intel’s practice of refining the same process for at least three years indeed lacks efficiency. Shifting the learning costs to consumers and helping them understand that 14nm+ is better than 10nm, and 10nm is better than 7nm? It is a long way to go.

Having read this, I believe everyone has gained some understanding of the process technology behind processors. Simply looking at the numbers does not fully reflect energy efficiency; in the face of complex professional terms like fin pitch, gate pitch, minimum metal pitch, and logic cell height, Intel always has the ability to challenge ahead due to the industry’s lack of mandatory standard concepts.

The good news is that Intel has now fully opened its foundry services, and the 10nm test chip based on the ARM Cortex-A75 core was completed in just 12 weeks. Moreover, the two mobile processors SC9861G-IA and SC9853I launched by Spreadtrum this year are both manufactured using Intel’s 14nm technology.

If Intel can attract Qualcomm, MediaTek, Kirin, and Xiaomi’s Surge processors to adopt its technology, and then conduct actual comparisons with competing chips of the same level, consumers will truly feel the powerful momentum brought by the benchmark technology.

After reading this article, 99% of readers will also click: