In the world of computer chips, larger numbers are often better. More cores, higher GHz, more FLOP, all engineers and users want. But there is a hot analogy in semiconductors now, and the smaller the better. Import semiconductor technology node (commonly known as process node).
But what exactly is it and why is it so important? Why is it measured in nanometers and why are we walking all over Sesame Street and presenting you this article with the numbers 10, 7 and 5?
Let's take a trip into the world of process nodes...
But before we get into that, if you take a look at our CPU architecture thread, you'll have a better understanding of things. In part one, we cover the principles of how processors work, and in part two, we study how programmers plan and design chips.
An essential part of this article is to explain how computers with the chip are actually placed. together. If you want to get a deeper understanding of the production process, you need to read the optical lithography section carefully, while in this feature we will focus more on what has been briefly mentioned:
In the chip industry, the size of the attribute is related to the so-called process node. As mentioned in CPU Design Part 3, this is a relatively loose term, with different manufacturers using the term to describe different aspects of the same chip, but recently referring to the smallest distance between the two parts. Transistors Transistors are, however, an important feature of any processor, because their combinations do all the slicing and data storage that takes place inside the chip, and a smaller process node from the same manufacturer is highly desirable. Q: The obvious question here is why? Larger components take longer to change their state, signals take longer, and more power is needed to transmit power through the processor. Larger components take up more physical space without trying to disassemble, so the chips themselves are larger.
In the image above, we're taking a look at three old Intel CPUs. Starting on the left, we have a 2006 Celeron, a 2004 Pentium M and a really old 1995 Pentium. They have 65, 90, and 350 nm processing nodes, respectively. In other words, the critical parts in a 24-year-old design are 5 times more than those in a 13-year-old model. Another important difference is that the newer chip has about 290 million transistors inside, while the original Pentium has more than 3 million transistors. Almost hundreds of times less.
Although process node cuts are only part of the reason new designs are getting smaller and more transistors come in, they play an important role in Intel's ability to deliver. It
but here's the main factor: that the Celeron only produces about 30 watts of heat, compared to a 12-watt Pentium. This heat is due to the fact that by pressing electricity to the circuits in the chip, energy is lost through various processes and the vast majority of it is released as heat. Yes, 30 is greater than 12, but do not forget that the chip has almost 100 times more transistors. Which one can change faster - giving us more computations per second - and waste less energy due to heat, begs another question: Why do we use the smallest possible process node on every chip in the world?Let it be light!
At this point, we need to look at a process called photolithography: light passes through something called a mask image, which blocks light in some areas and allows it to pass through in others. When the light passes through, the light focuses strongly on a small point and then interacts with a special coating used to make the slide, helping to locate the different parts. Imagine it's like an X-ray from your hand: the bones block the rays and act as a mask for images, while the body allows it and an image of the internal structure. Create hands.
Photo: Peellden, Wikimedia Commons
It doesn't actually use light - it's too big, even for chips like the old Pentium. You might be wondering how light on Earth could be of any size, but it has something to do with wavelength. Light is the so-called electromagnetic wave, which is a permanent combination of electric and magnetic fields.
Although we use the classic sine wave to visualize the shape, electromagnetic waves do not actually form. This is mostly the case with the effect you create by interacting with something. The wavelength of this periodic pattern is the physical distance between two identical points: an image of sea waves rolling on the shore shows the wavelength at which the crests of those waves are farther apart. Electromagnetic waves have a wide range of possible wavelengths, so we've put them together and call them a spectrum.Small, Smaller, Smaller
In the image below, we see that what we call light is just a thin image. There are other familiar names in this spectrum: radio waves, microwaves, X-rays, etc. The size of the light is about 7-10 meters which is about 0.000004 inches.
Scientists and engineers like to use slightly different ways to describe small, nanometer lengths, or "nanometers" for short. If we look at the widest part of the spectrum, we notice that the light ranges between 380 nm and 750 nm.
Image: Philip Ronan, Grainger
Return to this article Touch and re-read the part on the old Celeron chip - which was produced at the 65nm node. So how do you make smaller optical pieces? Simple: The lithography process doesn't use light, it uses ultraviolet light (commonly known as UV).
In the spectrograph, UV radiation starts at about 380 nm (where the light ends) and shrinks down to about 10 nm from manufacturers such as Intel, TSMC and GlobalFoundries of an electromagnetic wave. Called EUV (UV UV), they use a size of about 190 nm. This small wave means not only that your components can be made smaller, but that their overall quality will likely be better. This allows the different parts to be grouped together and helps reduce the overall chip size.
Different companies provide different names for the process node scale they are using. Intel is quick to call one of its latest P1274s, or "10nm," for the general public, while TSMC simply calls it "10FF." Processor designers like AMD create designs and architectures for smaller process nodes and then rely on people like TSMC to produce them.
TSMC is working hard on smaller nodes (7nm, 5nm and soon). 3nm) and making chips for its biggest customers including Apple, MediaTek, Qualcomm, Nvidia and AMD. At this production range, some of the smallest features are only 6 nm wide (however, most are much larger). To understand how small 6 nm really is, the silicon atoms that make up the bulk of the processor are about 0.5 nm apart, and the atoms themselves are about 0.1 nm in diameter. Therefore, TSMC factories treat transistor sides with a width of less than 10 silicon atoms as a cannonball. EUV photolithography created a series of serious engineering and production problems due to the properties of only a few atoms. GlobalFoundries encountered problems with 7nm production systems and smaller production systems. Although the Intel and GF problems may not be due to problems inherent in EUV optical lithography, they cannot be completely unrelated.
The shorter the wavelength of the electromagnetic wave, the more energy it carries, the more likely it is to damage the chip during production; Production on a very small scale is also very sensitive to contamination and defects in the materials used. Other issues, such as diffraction limits and statistical noise (the natural variation as the energy transmitted by an EUV wave settles on the wafer layer), also conspire against the goal of achieving 100% full chips.
Two production flaws in the image chip: Solid-state technology
There is also the problem that in the strange world of atoms, it is no longer possible to follow the flow of electricity and the transfer of energy from classical systems and laws. It is relatively easy to hold electricity, in the form of moving electrons (one of the three particles that make up atoms), to conduct conductors close to each other at the usual scale - just connecting one layer at a time. Thick dielectric wrap. .
At the level where Intel and TSMC are working, this is more difficult to achieve because the insulation is not thick enough. However, at the moment, production issues are almost entirely related to the inherent problems with EUV photolithography, so it takes several years for us to discuss in the forums that Nvidia manages quantum behavior better than AMD or any similar nonsense!
This is because the real problem, and the ultimate cause of production problems, is that Intel, TSMC and all their manufacturers are companies and they are just looking for atoms to make money in the future. In a research paper by Mentor, an overview of how much the chip costs for smaller process nodes... p>
For example, if we assume that the 28 nm process node is the same node that Intel used to produce the Spectrum A group of Haswell processors (such as the Core i7- 4790K), then a 10nm system would cost twice as much per chip. The number of chips each chip can produce depends largely on the size of each chip, but doing a smaller scale means the chip can have more chips to sell and help offset the increased costs. However, in the end, as the retail price of the product increases, the cost to the consumer is reduced as much as possible, but this must be balanced against industry demand.
The increase in smartphone sales over the years, combined with the explosive growth of smart technology in homes and cars, has meant that chip makers have been forced to face the financial blows of moving to smaller contract operations. The entire system is mature enough to produce highly efficient chips. defective items as much as possible) in bulk. Given that we're talking billions of dollars here, this is a risky business and a good reason why we saved GlobalFoundries from the escape race.Future prospects
If all this sounds a bit agonizing, well, let's not forget that the immediate future looks positive. Samsung and TSMC have been improving their 7-nanometer product lines for some time now, and chip designers plan to use multiple nodes in their products. AMD's chip design and strategy that began with 3rd generation Ryzen processors is being replicated by other chip makers. In this case, the AMD desktop processor used two chips made at the 7nm TSMC node and a 14nm chip made by GlobalFoundries. The first is real processor components, while the second is controlled by DDR4 memory and PCI Express hardware connected to the CPU.
The above graph shows the node changes of the Intel process over the past 50 years. The vertical axis represents the node size by a factor of 10, starting at 10,000 nm. This chip has a node half-life (the time required to reduce the node size by half each time) to 4.5 years. Does this mean we will see 5nm from Intel by 2025? Probably yes, although they have a problem with 10nm, they are working hard the way back. Samsung and TSMC have made progress with 5nm production and beyond, so the future looks good for a variety of processors.
They are smaller and faster, consume less power and provide better performance. It paves the way for fully self-driving cars, smartwatches with the power and battery life of today's smartphones, and in-game graphics that go beyond anything seen in multi-million dollar movies ten years ago. The future is really bright, because the future is small.
The Purpose of Atoms: The Art of Making Chips Smaller