7nm vs 10nm vs 14nm: Fabrication Process

“7nm, 10nm, 12nm, 14nm, 16nm….”

You’ll generally find this number when you’re going through the specs of Semiconductor devices like a Processor.
Product reviewers and Tech Experts talk about this number like its a big deal.

So, what exactly is this nanometer number?

Well, as a consumer, all you need to know that this number represents the size of transistors (or components) in a Processor (We will get back to the actual meaning later). The transistors are the buildings block of a CPU and digital circuits.

When we combine transistors in different ways, we obtain logic circuits like AND, NOT, OR Gates. Then these gates can be used to obtain Adders, Multipliers, and other different types of complex circuits.

A modern processor may contain billions of transistors. For instance, AMD Ryzen’s 1st Generation had 4.8 billion transistors in their 8-core Zeppelin die.

We need to shrink the size of transistors so that we can increase their count in the same unit area.

So, what happens when we shrink the size of a Transistor?

This results in two major improvements:

1) Performance: As the size of the transistor decreases, we can fit a higher number of them in the same unit area. Hence, we can achieve higher processing power from the same sized processor.

2) Power Efficiency: Smaller transistors require less power for their functioning. This reduces the overall power consumption of the chip. Less power also results in the generation of less heat and thus allowing us to increase the clock speeds further.

This was explained in Moore’s Law which states that the number of transistors in an Integrated Circuit doubles every two years because of the advancement in manufacturing technology. We also call this Die Shrink.

Moore’s Law: The Increase in the Number of Transistors

This not only leads to performance improvement in semiconductor devices over the years but also reduces their size. This is the reason why computers keep getting smaller and faster.

Here’s a video from Linus to help you learn more about the same.

Hence, this nanometer value or the Fabrication Process matters a lot, right?
So, the next time you compare two products, pay close attention to this number.

And now for those of you who want to go into the details, here’s the real deal:

And there is no universal standard to calculate this value either. Different brands have different ways to calculate this. For instance, the 10nm from TSMC is not equivalent to the 10nm from Samsung.
What Intel calls as 10nm, is similar to what TSMC calls as 7nm. Thus, there is no universal standard for the same.

So, what matters now?

Transistor Density!

The fabrication process with a higher transistor density is usually better. Move on to the next section to find the ranking of different process nodes based on their density.

7nm vs 10nm vs 12nm vs 14nm: Transistor Densities

We all have been waiting for Intel’s 10nm Processors since 2016. And as of early 2020, Intel has only managed to launch 10nm chips for Laptops (Ice Lake) and these are clocked very low and the yields are still poor.

But one big reason for the delay is the fact that Intel is taking a big leap forward. What Intel’s 10nm brings is even better than the 7nm offerings of Samsung and TSMC. Here’s a look at the transistor densities of different semiconductor chip manufacturers.

Transistor Density Comparison

While there’s hardly any sign of Intel’s 10nm in Mainstream Processors, TSMC’s 7nm has already entered mass production and is used for manufacturing of Apple A12 Bionic, Kirin 980, Snapdragon 855, and the Ryzen 3000 Series (Zen 2).

As Zen 2 is based on TSMC’s 7nm HPC Process, it has presented AMD with a great opportunity to capture some market share from Intel in 2019.

I’ve ranked different fabrication technologies to make it easy to understand which process is the best. As of November 2019, TSMC’s 7nm+ is the most advanced node that has entered mass production.

*Intel’s 10nm’s density is based on their estimation for Cannon Lake in 2018. The actual density for the current Ice Lake chips could be much lower. If Intel ever brings 10nm to their HPC products, the density could be very close to TSMC’s 7nm HPC Process which powers Zen 2 and Navi 10.

You should note that the above densities are approximate & estimated in some cases while some of them are the actual densities. But nevertheless, that won’t affect the ranking. We also use these transistor densities for calculating the Centurion Mark for Mobile Processors.

If we go deeper into the matter, Intel’s 10nm is slightly denser than TSMC’s 7nm for SRAM. But TSMC’s 7nm is actually denser than Intel for logic. This complicates the thing even more, right?

So, let’s just focus on the above table for now.

Density is calculated in terms of MTr/mm² which stands for millions of transistors per square millimeter.

Intel uses the following formula to calculate the density:

Calculation of Transistor Density

Fabrication Technologies of Different Manufacturers

Now let’s have a look at the different process nodes of various manufacturers like TSMC, GlobalFoundries, Samsung, and Intel.

Zen 2 vs Sunny Cove

2019 is expected to be a great year for the Computer Hardware industry. We are expecting stiff competition between AMD, Intel, and Nvidia. It will be interesting to see how well does AMD’s Navi GPU performance. But we will have a discussion about the GPUs sometime later. This section is meant for the CPUs.

When AMD introduced Ryzen in 2017, we got some amazing processors at a great price value. But even after the launch of their 2nd Generation, AMD is significantly behind in Single-Threaded performance. Though AMD is currently a better the best choice for workstation builds and for content creators, Intel continues to be the preferred choice among gamers.

Will it change with the launch of Zen 2?

Probably yes.

Ryzen 3000 Series (Matisse) based on Zen 2 Architecture will be available during the mid of 2019.

If we look at the Process Node, Zen 2 will be manufactured using TSMC’s 7nm HPC process that has a density of 66.7 MTr/mm² which is almost twice that of Zen+. When compared to Intel’s 14nm, Zen 2 is 53% denser.

Image Source: AMD

With Zen 2, I’m expecting higher IPC and clock speeds. Furthermore, the requirement for a high-speed memory in Ryzen Processors (for better performance), would be reduced. If the leaks are true, Zen 2 will also have more cores but will be priced the same as the existing Ryzen 2000 series.

AMD may or may not surpass Intel in Single-Threaded performance but it will definitely come close to a point that Zen 2 would become a great choice for Gamers as well. So my advice for AMD is to simply stick with their schedule and keep launching products at competitive prices. AMD has a great opportunity to grab a decent market share from Intel in the upcoming two years.

Now let’s talk about Intel. They first showcased their new CPU Core Roadmap at Intel’s Architecture Day on December 11th, 2018. That day, Intel showcased many different technologies such as 3D chip-stacking.

Image Source: Intel

The new Sunny Cove microarchitecture is expected to bring higher throughputs and better scalability. It will be built on Intel’s 10nm Process which is 2.7 times denser than their 14nm process.

My advice for Intel is that they really need to deliver it this time or else AMD would totally dominate 2019. Intel also needs to bring back the Hyper-Threading Technology for their Core i5 & i7 Series.

Hopefully, Intel will release their 10th Generation Ice Lake Processors by the end of 2019.

If everything goes well for Intel and considering that their EUVL based 7nm Process is on track, they could also possibly launch Golden Cove in 2021 with 7nm Technology.

The roadmaps of both Intel & AMD seem exciting. I cannot wait to see how powerful the processors might get by 2021 if both these companies deliver as promised.

So, what are you more excited for? AMD Ryzen’s 3rd Generation (Matisse) or Intel’s 10th Generation (Icelake)?

Let me know in the comments below.

As for me, I’m more excited for Ryzen 3000 Series as I’ll build my next workstation PC with the Ryzen 7 3700X. But I do hope Intel begins the mass production of 10nm in 2019 and makes the competition stiff.

Good competition between brands results in great products but when we have a monopoly, brands end up refining their ‘Skylake’ architecture thrice.

The End of Moore’s Law & Future

With the delay on Intel’s 10nm, it was clear that the end of Moore’s law is nearing. But rather than calling it an end, we should better call it a “slow down”. The number of transistors is still increasing every two years, but not at the same rate.

I believe that we can shrink the transistor size for another 5 to 10 years before Quantum Tunneling comes into play. Here’s a video from Kurzgesagt – In a Nutshell that explains Quantum Tunneling and Quantum Computing.

But for the next few years, we still have 5nm & 3nm Nodes from TSMC & GlobalFoundries to watch out for. Also, Intel’s upcoming 7nm Process will have higher transistor density than TSMC’s 5nm.

Guess what? TSMC’s 5nm is not too many years away. In fact, TSMC will begin the initial production in the 2nd half of 2019. Both the TSMC’s 5nm and Intel’s 7nm will use Extreme ultraviolet lithography. You can follow this article to learn more about Intel’s 7nm plans.

Here are some estimates from SemiWiki about the upcoming 7nm, 5nm, and 3nm nodes.

So, what happens when Moore’s Law is dead?

We will have to switch to alternative elements or wait for some kind of technological breakthrough. One way to examine these rare architecture resets is to step back from individual process nodes. Our 250-year hardware innovation timeline through Rahu’s cycle connects the transistor, Moore’s Law, 64-bit x86, and AI accelerators, then considers where post-CMOS computing could emerge next.

You can learn more about future technologies in the following video by Seeker.

Here’s an article on Digital Trends about how performance will continue to improve even after the end of Moore’s Law.

References

• Let’s Clear Up the Node Naming Mess – Intel
• 14nm, 16nm, 10nm, 7nm: What we Know – SemiWiki
• Samsung Electronics Starts Production of EUV-based 7nm LPP Process – Samsung
• 7 nanometer – Wikipedia
• Technology Node – WikiChip
• Samsung’s 8nm 8LPP, a 10nm extension – WikiChip Fuse
• Analysis of Intel’s 10nm Process – Wccftech