The whole process of a pile of sand to a powerful integrated circuit chip

The CPU is a core part of a modern computer and is also called a "microprocessor." For PCs, CPU specifications and frequency are often used as an important indicator of the performance of a computer. The Intelx86 architecture has been going on for more than 20 years, and the CPU of the x86 architecture has a profound impact on the work and life of most of us.

Many friends who know a little about computer knowledge will mostly know that the most important thing in the CPU is the transistor. This will increase the speed of the CPU. The most important point is to say how to put more transistors in the same area of ​​the CPU. The CPU is too small, too precise, and it contains a large number of transistors, so it is absolutely impossible to complete it and it can only be processed by the photolithography process.

This is why there are so many transistors in a CPU. The transistor is actually a two-position switch: on and off. If you recall the era of basic computing, it is all that a computer needs to do its job. Two choices, on and off, are 0 and 1 for the machine. So how do you make a CPU? In today's article, we will step by step tell you the whole process of the central processor from a pile of sand to a powerful integrated circuit chip.

The basic raw material for manufacturing CPU

If you ask about the raw materials of CPU, everyone will give an easy answer - it is silicon. This is true, but where does silicon come from? In fact, those are the most insignificant sand. It is hard to imagine that the expensive, complicated structure, and powerful CPU that is full of mystery come from the fundamentally worthless sand. Of course, this must go through a complicated manufacturing process. However, it is not a matter of grasping a handful of sand to make raw materials. It must be carefully selected to extract the purest silicon raw materials. Just imagine, if the CPU is made of the cheapest and most abundant raw material, then the quality of the finished product will be. Can you use a high-performance processor like this one?

In addition to silicon, an important material for manufacturing CPUs is metal. So far, aluminum has become the main metal material for making internal parts of processors, and copper has gradually been eliminated. There are some reasons. Under the current CPU operating voltage, the electromigration characteristics of aluminum are much better than copper. The so-called electromigration problem means that when a large amount of electrons flow through a section of conductor, the atoms of the conductor material leave the original position by the impact of electrons, leaving vacancies. Excessive vacancies will lead to the disconnection of the conductor, leaving the atoms in situ. Staying at other locations will cause short circuits in other places and affect the logic function of the chip, which in turn will cause the chip to become unusable.

In addition to these two major materials, some types of chemical materials are needed in the chip design process. They play different roles and will not be described here.

CPU preparation stage

After the collection of essential raw materials is completed, some of these raw materials require some pre-processing work. As the most important raw material, the processing of silicon is very important. First of all, the silicon raw material is chemically purified, and this step makes it reach the raw material grade that can be used by the semiconductor industry. In order to make these silicon raw materials to meet the processing needs of integrated circuit manufacturing, they must also be shaped. This step is accomplished by melting the silicon raw material and then injecting the liquid silicon into a large-scale high-temperature quartz container.

Then, the raw material is melted at a high temperature. In the middle school chemistry class we learned that many solid internal atoms have a crystal structure, and so does silicon. In order to meet the requirements of high-performance processors, the bulk silicon must be highly pure and monocrystalline. The silicon raw material was then removed from the high temperature vessel using a rotary draw where a cylindrical silicon ingot was created. From the current process used, the diameter of the silicon ingot has a circular cross-section of 200 mm.

However, intel and other companies have begun to use 300mm diameter silicon ingots. It is quite difficult to increase the area of ​​the cross-section while keeping the various properties of the silicon ingot unchanged. However, as long as the company is willing to invest a large amount of funds for research, it can still be achieved. The factory built by Intel for the development and production of 300-millimeter silicon ingots costs about 3.5 billion U.S. dollars. The success of the new technology allows Intel to manufacture more complex and more powerful integrated circuit chips. The 200-millimeter silicon ingot factory also cost 1.5 billion U.S. dollars

After making a silicon ingot and ensuring that it is an absolute cylinder, the next step is to slice the cylinder ingot. The thinner the slice, the less material it can consume, and the more naturally it can produce more processor chips. The slicing also requires a mirror finish to ensure that the surface is absolutely smooth and then checked for distortion or other problems. The quality inspection of this step is particularly important. It directly determines the quality of the finished CPU.

The new slice is to be doped with some material to make it a true semiconductor material and then to be scribed on it with transistor circuits representing various logic functions. The doped substance atoms enter the voids between the silicon atoms, and atomic force acts on each other, so that the silicon raw material has semiconductor characteristics. Today's semiconductor manufacturing is a multi-select CMOS process (Complementary Metal Oxide Semiconductor).

The word complementary refers to the interaction between an N-type MOS transistor and a P-type MOS transistor in a semiconductor. N and P represent negative and positive electrodes respectively in the electronic process. In most cases, the slices are doped with a chemical substance to form a P-type substrate, and the logic circuit scribed thereon is designed to follow the characteristics of the nMOS circuit. This type of transistor has a higher space utilization rate and is more energy-efficient. At the same time, in most cases, it is necessary to limit the occurrence of the pMOS transistor as much as possible, because at the later stage of the manufacturing process, the N-type material needs to be implanted into the P-type substrate, and this process leads to the formation of the pMOS transistor.

After the work of incorporation of chemical substances is completed, standard slices are completed. Each slice was then heated in a high-temperature oven and a silicon dioxide film was formed on the surface of the slice by controlling the heating time. By closely monitoring the temperature, air composition and heating time, the thickness of the silica layer can be controlled. In Intel's 90-nanometer manufacturing process, the gate oxide width is as small as an astonishing 5 atomic thickness. This gate is also part of the gate of the transistor. The role of the gate of the transistor is to control the flow of electrons. Through the control of the gate voltage, the flow of electrons is strictly controlled regardless of the voltage of the input and output ports.

The final step in the preparation work is to cover the silicon dioxide layer with a photosensitive layer. This layer of material is used for other control applications in the same layer. This layer of material has a good photographic effect when dried, and after the photolithography process is completed, it can be chemically dissolved and removed.

Photolithography

This is a very complicated process in the current CPU manufacturing process. Why? The photo-etching process uses a certain wavelength of light to inscribe a corresponding nick in the photosensitive layer, thereby changing the chemical properties of the material at that location. This technique requires very strict wavelength requirements for the light used, requiring the use of short-wavelength UV light and large curvature lenses. The etch process can also be affected by smut on the wafer. Each step of etching is a complex and delicate process.

The amount of data needed to design each step of the process can be measured in units of 10 GB, and the etching steps required to make each processor are more than 20 steps (each step is etched in one layer). And if the etched drawings of each layer are magnified many times, it can be even more complex than the maps of New York City and suburbs. Imagine that the entire New York map has been reduced to the actual size of only 100 square millimeters. On the chip, how complicated the structure of this chip is, can be imagined.

When these etching operations are all completed, the wafer is flipped over. The short-wavelength light shines on the photosensitive layer of the wafer through the etched nicks on the quartz template, and then the light and the template are removed. The photosensitive material exposed on the outside is chemically removed, and the silicon dioxide is immediately formed under the hollow position.

Doping

After the remaining photosensitive layer material has been removed, the remaining trenched silicon dioxide layer and the exposed silicon layer underneath the layer remain. After this step, another silicon dioxide layer was completed. Then, another polysilicon layer with a photosensitive layer is added. Polysilicon is another type of gate circuit. Because metal feeds (hence metal oxide semiconductors) are used here, polysilicon allows the gates to be established before the transistor queue port voltage is activated. The photosensitive layer is also etched by the short wavelength light through the mask. After a second etching, all the necessary gate circuits have been basically formed. Then, the exposed silicon layer is chemically bombarded with ions, the purpose of which is to generate N-channel or P-channel. This doping process creates all of the transistors and the circuit connections to each other. None of the transistors have an input and an output, which is called a port between the two ends.

Repeat this process

From this point on, you will continue to add layers, add a silicon dioxide layer, and then photolithography. Repeat these steps, and then there will be a multi-layered architecture, which is the bud of the processor you are currently using. Electrically conductive connections between layers are made using a metal coating technique between each layer. Today's P4 processor uses a 7-layer metal connection, while Athlon64 uses 9 layers. The number of layers used depends on the original layout design and does not directly represent the performance difference of the final product.

Test the package test process

The next few weeks will require a series of tests on the wafer, including testing the electrical characteristics of the wafer to see if there is a logic error, if so, which layer it is on, and so on. Then, each defective chip unit on the wafer will be individually tested to determine whether the chip requires special processing.

Then, the entire wafer is cut into individual processor chip units. In the initial test, those units that failed the test will be abandoned. These cut-out chip units will be packaged in some way so that it can be smoothly inserted into a motherboard of an interface specification. Most intel and AMD processors are covered by a thermal layer.

After the finished product of the processor is completed, a full-scale chip function test is also performed. This department will produce different grades of products, some of the chip's relatively high frequency of operation, then put the name and number of high-frequency products, and those relatively low frequency of operation of the chip will be modified to put on other low-frequency models. This is the processor of different market positioning. Some processors may have some deficiencies in chip functionality. For example, if it has a defect in the cache function (this defect is enough to cause most of the CPU), then they will be shielded some cache capacity, reduce performance, and of course reduce the price of the product, which is Celeron And the origin of Sempron.

Before the CPU is placed in the box, the last test is usually performed to ensure that the previous work is accurate. According to the previously determined maximum operating frequency, they are put into different packages and sold around the world.

After reading this, I believe you have some in-depth understanding of the CPU's manufacturing process. The manufacture of CPU can be said to be a great achievement of cutting-edge science and technology in many aspects. The CPU itself is also very big. If the materials inside are sold separately, I am afraid that I can't sell a few dollars. However, the manufacturing cost of the CPU is very alarming. From here perhaps we can understand why this thing is so expensive.

It is very important to test this link. For example, if your processor is 6300 or 6400, it will be divided in this link. 6300 is not born 6300, but after testing, it is found that the processor cannot work under the 6400 standard, only Can work stably under the 6300 standard, then define the processor, lock frequency, define the ID, package, and print on the 6300.

We use AMD for example: the same core processor is a production line down, if the stable work in 2.8GHz, 1M * 2 cache, it is defined as 5600 +, if the cache is defective, the cutting problem Half, as 5400+, if the cache is ok and the frequency can only pass the test at 2.6G, then it is 5200+, if the cache has flaws, it will be cut into 5000+.........it has measured it to 3800+, if not yet Stability, either to find ways to become Athlon 64 single-core or single-core Sempron, or is the emergence of the ES version of the dual-core Sempron, if the batch does not work in 3800 + conditions, and work in 3600 + conditions, then The 3600+ is on the market. If there are batches that can work under 3G, 1M*2, then the 6000+ will be on the market. This is why the processor is always listed on the mid-size model, and the high-end and the bottom-end models will be listed later. It may be cost-saving to create the bottom line, specifically to produce bottom-end processors. Various models of Celeron and Sempron will be on the market successively, and the high-end pipeline will change to the bottom due to unstable individual processors. Processor, for example, cutting becomes Athlon 64 Sempron cache 64.

Intel Core i7 production process diagram

Sand: Silicon is the second most abundant element in the earth's crust, and deoxidized sand (especially quartz) contains up to 25% of silicon, in the form of silicon dioxide (SiO2), which is the basis of the semiconductor manufacturing industry.

Silicon smelting: 12-inch/300-mm wafer level, the same below. The silicon used for semiconductor manufacturing quality, scientifically known as electronic grade silicon (EGS), is obtained by multi-step purification and there is at most only one impurity atom per one million silicon atoms. This figure shows how large crystals can be obtained by purifying silicon, and finally the silicon ingot (Ingot) is obtained.

Monocrystalline silicon ingots: The overall basic cylindrical shape, weighing about 100 kilograms, silicon purity 99.9999%.

Ingot Cutting: A single silicon wafer that is cut horizontally into a circle, also known as Wafer. By the way, why do you know why wafers are round?

Wafers: The cut wafers are almost perfect after polishing and the surface can even be used as a mirror. In fact, Intel does not manufacture such wafers itself. Instead, it directly purchases finished products from third-party semiconductor companies and then uses its own production lines for further processing, such as the current mainstream 45nm HKMG (high-k metal gate). It is worth mentioning that the wafer size used by Intel at the beginning was only 2 inches/50 mm.

Photo Resist: The blue part of the picture is the photoresist liquid poured during wafer rotation, similar to the traditional film. Wafer rotation allows the photoresist to be very thin and flat.

Photolithography: The photoresist layer is then exposed to ultraviolet (UV) light through the mask and becomes soluble. The chemical reactions that occur during this period resemble changes in the film at the moment the mechanical camera shutter is pressed. A pre-designed circuit pattern is printed on the mask. Ultraviolet light shines on it and forms a circuit pattern for each layer of the microprocessor. In general, the circuit pattern on the wafer is a quarter of the pattern on the mask.

Photolithography: This leads to a 50-200 nm transistor grade. Hundreds of processors can be cut from a single wafer, but from here, the vision is reduced to one of them, showing how to make transistors and other components. The transistor is equivalent to a switch and controls the direction of the current. The transistors are now so small that about 30 million can be put on one needle.

Dissolving photo-resist: Photoresist exposed to UV light during photolithography is dissolved away, leaving the same pattern and mask after removal.

Etching: Use chemicals to dissolve the exposed portions of the wafer while the remaining photoresist protects the portions that should not be etched.

Remove the photoresist: After the etching is completed, the mission of the photoresist is announced. After the clearance is cleared, the designed circuit pattern can be seen.

Photoresist: The photoresist (blue portion) is again coated, then photolithographed, and the exposed portion is washed away. The remaining photoresist is also used to protect the part of the material that will not be ion-implanted.

Ion Implantation: In a vacuum system, ions are irradiated (injected) with ions of accelerated atoms to be doped to form a special implanted layer in the implanted region and change the silicon of these regions. Conductivity. After the electric field is accelerated, the injected ion current can exceed 300,000 kilometers per hour.

Removing the photoresist: After the ion implantation is completed, the photoresist is also removed, and the implanted region (green portion) is also doped, and different atoms are implanted. Note that the green color at this time is different from before.

Transistor ready: At this point, the transistor is almost complete. Three holes were etched in the insulating material (magenta) and filled with copper to interconnect with other transistors.

Electroplating: Electroplating a layer of copper sulfate on a wafer to deposit copper ions onto the transistor. Copper ions move from the positive (anode) to the negative (cathode).

Copper layer: After plating, copper ions are deposited on the wafer surface to form a thin copper layer.

Polishing: Polishing excess copper, which is to polish the wafer surface.

Metal layer: Transistor level, a combination of six transistors, about 500 nm. A composite interconnect metal layer is formed between different transistors, and the specific layout depends on the different functionalities required by the respective processor. The surface of the chip looks unusually smooth, but in fact it may contain more than 20 layers of complex circuits. After amplification, it is possible to see an extremely complex circuit network, shaped like a futuristic multi-level freeway system.

Wafer test: Core level, approximately 10 mm/0.5 inch. The figure is part of the wafer and is undergoing the first functional test using a reference circuit pattern to compare with each chip.

Slicing: wafer level, 300 mm/12 inches. The wafer is cut into blocks, each of which is a processor core (Die).

Discard the core: wafer level. The flawed kernel found during the test was abandoned, leaving it ready for the next step.

Single kernel: kernel level. A single core cut from the wafer shows the core of Core i7.

Package: Package level, 20mm/1 inch. The substrate (substrate), core, and heat sink are stacked together to form what we see as a processor. The substrate (green) is equivalent to a base and provides the processor core with an electrical and mechanical interface that facilitates interaction with the rest of the PC system. The heat sink (silver) is responsible for the heat dissipation of the core.

Grade test: The last test can identify the key features of each processor, such as the maximum frequency, power consumption, heat, etc., and determine the processor's level, such as the Core i7-975 Extreme, which is suitable for the highest end. Low-end model Core i7-920.

Packing: Ship the same level of processors together according to the rating test results.

Retail packaging: The manufactured and tested processors are either delivered in batches to OEMs or placed in boxes to enter the retail market.