CPU creation process

As the core component of the computer, the CPU (central processor unit) has been very mysterious in the hearts of users: in the minds of most users, it is just an abbreviation, they cannot even write all of it. In the eyes of some hardware experts, the CPU is at most a block of more than 10 square centimeters, with many feet, the CPU core is less than 1 cm² large. They know how many microns a piece of less than 1 cm² is made of, and that it integrates hundreds of millions of transistors, but seldom understands the manufacturing process of CPUs. Today, let's take a closer look at how the CPU is developed.

Basic materials

Most people know that modern CPUs are made of silicon. Silicon is a kind of non-metallic element. From a chemical point of view, because it is at the junction of the metal element zone and the non-metallic element zone in the periodic element table, it has the properties of semiconductors, it is suitable for manufacturing various tiny transistors and is one of the most suitable materials for manufacturing modern large-scale integrated circuits. In a sense, the main component of the sand on the beach is silicon dioxide, and the silicon material used to produce the CPU is actually extracted from the sand. Of course, some other materials need to be used in the CPU manufacturing process, that is why we don't see intel or AMD just pull tons of sand to their manufacturing plants. At the same time, the manufacturing CPU has a high requirement on the purity of Silicon Materials. Although it comes from cheap sand, but due to the complexity of the material purification process, we still cannot compare the price of one hundred grams of high-purity silicon with a ton of sand.

Another basic material for manufacturing CPUs is metal. A metal is used to make a circuit that connects components inside a CPU. Aluminum is one of the commonly used metal materials because it is cheap and has good performance. Most of today's mainstream CPUs use copper instead of aluminum, because the electric migration of aluminum is too high, it is unable to meet the needs of the current rapid development of the CPU manufacturing process. The so-called electricity migration refers to the migration of some metal atoms from the original place under certain conditions (such as high voltage.

Obviously, if an atom constantly moves out of the metal microcircuit of the connected element, the circuit will soon become riddled with holes until the circuit is broken. That's why when a frequent user tries to significantly increase the voltage of the Northwood Pentium 4, this miserable CPU is often "Sudden Northwood death syndrome (sudden Northwood death syndrome, snds) "causes of shock and even sacrifice. Snds enables intel to apply the Copper Interconnect Technology to the CPU production process for the first time. Copper Interconnect Technology can significantly reduce the phenomenon of electric migration, but it can also be smaller than the aluminum process manufacturing circuit, this is also a problem that can not be ignored in the nano-scale manufacturing process.

More than that, copper is much less resistant than aluminum. Various advantages allow Copper Interconnect Technology to quickly replace aluminum and become the mainstream choice for CPU manufacturing. In addition to silicon and some metal materials, there are also a lot of complex chemical materials involved in the manufacturing of CPU.

Preparations

After solving the problem of manufacturing CPU material, we started to prepare. In the preparation process, some raw materials will be processed to make the electrical performance meet the manufacturing CPU requirements. One is silicon. First, it will be purified by chemical methods, pure to almost no impurities. At the same time, it has to be converted into a silicon crystal, and essentially draw a boundary between the sand on the beach.

In this process, raw silicon will be melted and placed into a giant Z melting furnace. In this case, a kind of crystal is placed into the furnace so that the silicon crystal can grow around the crystal until a crystal silicon is almost perfect. If you did a good job in the crystallization experiment of copper sulfate in high school, or saw how the crystal sugar was made, it is not hard to understand this process. At the same time, you need to understand that many solid substances have crystal structures, such as salt. The same is true for silicon in CPU manufacturing. Careful and slow mixing of the silicon paste, the silicon crystal surrounded by the crystal to grow in the same direction. Finally, a piece of silicon ingot is produced.

At present, the diameter of the silicon ingot is 200mm, and the CPU manufacturer is preparing to manufacture the 300mm diameter silicon ingot. It is obviously more difficult to create a larger silicon ingot without changing the quality, but the investment of CPU manufacturers solves this technical problem. The construction of a manufacturing plant that produces 300mm-diameter ingots costs about $3.5 billion, and Intel will use the silicon material it produces to make more complex CPUs. The construction of a similar manufacturing plant that produces 200mm-in-diameter silicon ingots costs $1.5 billion. As the first person to eat crabs, Intel obviously has to pay a higher price. It seems impossible to build such a manufacturing plant with more than twice the cost, but we can see from below that this investment is worthwhile. There are many other ways to manufacture a silicon ingot. one of the methods described above is the CZ manufacturing method.

The ingot is made and formed into a perfect cylinder. It is then cut into slices, called wafers. Wafers are actually used for CPU manufacturing. Generally, the thinner the wafer is, the more CPU products the same amount of silicon material can make. Then the wafer will be polished and checked for any deformation or other problems. Here, quality checks directly determine the final CPU yield rate, which is extremely important.

Problematic Wafers will be incorporated with appropriate other materials to make various transistors on them. The doped materials are deposited in the gaps between silicon atoms. The most widely used transistor manufacturing technology is CMOS (complementary metal oxide semiconductors, complementary metal oxide semiconductor) technology, which you often see. In a simple explanation, the C (complementary) in CMOS refers to the relationship between two different MOs circuit "N" and "P" circuits: they are complementary.

In electronics, "N" and "p" are abbreviations of negative and positive, respectively, used to represent polarity. It can be simply understood that the "P" well can be installed on the "N" type substrate to manufacture the "P" type transistor, the "N" well can be installed on the "P" substrate to manufacture the "N" transistor. In most cases, the manufacturer adds relevant materials to the wafer to make the "P" substrate, because the "P" substrate can produce better performance, in addition, the "N" transistor can effectively save space. In this process, the manufacturer will try to avoid the "P" transistor.

The wafer will be sent to a high-temperature furnace. Of course, this time we will not let it melt again. By closely monitoring the temperature, pressure, and heating time in the furnace, the wafer surface will be oxidized into a layer of silica (SiO2) of a specific thickness as part of the transistor gate circuit-substrate. If you have learned about Logical Circuits, you will be very clear about the concept of gate circuits. Through the door circuit, a certain input level will get a certain output level, the output level varies according to the door circuit. The level is represented by 0 and 1, which is why the computer uses binary. Among Intel's CPUs manufactured using a 90nm process, this layer of gate circuit is as thick as five atoms.

The final step of preparation is to apply a photosensitive anti-corrosion film to the wafer, Which is intrusive and the photosensitive part can be cleaned by specific chemicals to separate it from the unexposed part.

Complete door circuit

This is the most complex process in the CPU manufacturing process. This time we used the Optical Micro-engraving technology. In this case, the optical micro-engraving technology has pushed the application of light to the limit. The CPU manufacturer will expose specific areas of the photosensitive anti-etching films covered on the wafer and change their chemical properties. To avoid the interference of light in areas that do not need to be exposed, a mask must be prepared to mask these areas. You must have recognized the concept of mask in software such as Photoshop, Which is similar here.

Here, even if a very short ultraviolet light is used and a large lens is used, that is, the edge of the mask will still be affected for the best focus, you can simply imagine that the edge is blurred. Note the scale we are discussing. Every Mask is too complicated to be imagined. If you want to describe it, you must use at least 10 Gb of data to create a CPU, at least 20 such masks are used. For any mask, try imagining a map of Beijing, including its suburbs, and then scale it down to a small piece of paper worth 1 cm² square meters. Finally, don't forget to connect each map. Of course, what I'm talking about is not as simple as connecting with a line.

After the masks are made, they will be covered on the wafer, and the short-wave light will be exposed through the holes in these Z masks on the photosensitive anti-corrosion membrane. Next, stop the light and remove the mask, use a specific chemical solution to clean the exposed photosensitive anti-corrosion film, and closely follow a layer of silicon.

When the remaining photosensitive anti-etching films are also removed, the wafers leave the ups and downs of silica mountains. Of course you cannot see them. Next, add another layer of silica, add a layer of polysilicon, and then cover a layer of photosensitive anti-corrosion film. Polysilicon is another part of the door circuit mentioned above, which was previously made of metal (M: metal in CMOS ). The photosensitive anti-corrosion membrane is once again covered with a mask that determines the removal of these polysilicon to accept the baptism of light. Then, the exposed silicon will be bombarded by atoms to make N or P Wells. Combined with the substrate made above, the door circuit will be completed.

Repeated

You may think that after the complex steps above, a single CPU has almost been manufactured. In fact, at this time, the CPU completion is less than 1/5. The subsequent steps are as complex as described above, that is, adding the silica layer again, etching again, and adding another ...... Repeat multiple times to form a 3D structure, which is the core of the final CPU. Each layer must be filled with metal as a conductor. Intel's Pentium 4 processor has 7 layers, while AMD's athlon 64 reaches 9 layers. The number of layers depends on the CPU layout and the current size.
Testing, testing, and testing

After several weeks of manufacturing from the initial wafer to the CPU core of a layer of silicon, metal and other materials, it is time to look at the monster. This step tests the electrical performance of the wafer to check for any errors and the steps at which the errors occur (if possible ). Next, each CPU core on the wafer will be separated (not cut) for testing.

The tested Wafers will be split into several separate CPU cores, And the invalid cores found in the test above will be put aside. The core will be encapsulated and installed on the substrate. Then, most mainstream CPUs will install an integrated anti-deformation chip (IHS) on the core ). Each CPU is fully tested to test all its functions. Some CPUs can run at a high frequency, so they are labeled with a high frequency. Some CPUs have a low running frequency due to various reasons, so they are labeled with a low frequency. Finally, some CPU may have some functional defects. If the problem occurs in the cache (the cache occupies more than half of the CPU core area), the manufacturer can still block some of its caches, this means that the CPU is still available for sale, but it may be celon, sempron, or something else.

Before the CPU is put into the packing box, the last test is generally required to ensure that the previous work is correct. According to the highest operating frequency determined above, they are put into different packages and sold all over the world.

After reading this, I believe you have a deep understanding of the CPU manufacturing process. The manufacturing of CPU can be said to be the result of cutting-edge science and technology. The CPU itself is just a little big. If we separate the materials in it and sell them separately, I am afraid it will not cost much. However, the manufacturing cost of CPU is amazing. From here we may understand why it is so expensive.