075582814553
The CPU, sometimes known as a "microprocessor," is the heart of modern computers. The specs and frequencies of the CPU are frequently regarded as crucial indicators of a computer's performance in PCs. The Intel x86 architecture has been around for more than two decades, and the x86 architecture's CPU has had a significant impact on most of our work and lives.
The transistor is the most critical component in the CPU, according to many acquaintances who know a bit about computers, The most important point to improve the speed of the CPU is to figure out how to fit more transistors into the same CPU area, because the CPU is too small and precise, and there are so many transistors in it, that humans will never be able to finish it, and it can only be processed by photolithography.
This is why a CPU can have so many transistors. A transistor is a switch with two positions: on and off. If you go back to the early days of computing, that was all a computer needed to get the job done. The machine has two options: on and off, which are 0 and 1. So, how would you go about making a CPU? In today's essay, we'll take you through the entire process of creating a central processing unit, from a pile of sand to a powerful integrated circuit chip, step by step.
If you ask what the raw material of the CPU is, almost everyone will say silicon. True, but how did the silicon get there? It is, in reality, the least noticeable sand. It's difficult to believe that the CPU's pricey, complicated, powerful, and enigmatic nature emerged from such worthless sand. Of course, somewhere in the midst, there must be a difficult production process. To manufacture raw materials, though, you can't just grab a handful of sand; you have to pick carefully and extract the finest silicon raw material from it. Consider the quality of the completed product if the CPU is manufactured of the cheapest and most abundant raw materials; can you still use such a high-performance processor as it is now?
Metal is another significant component of a CPU, in addition to silicon. Aluminum has largely replaced copper as the primary metal used in the manufacture of processor internal parts. This is due to a number of factors. Aluminum has much better electromigration characteristics than copper at the current CPU operating voltage. When a significant number of electrons flow through a conductor, the atoms of the conductor substance are hit by the electrons and leave their original places, leaving voids. This is known as the electromigration problem. Staying in other positions will induce short circuits in other places and disrupt the chip's logic function, rendering it useless.
Aside from these two basic elements, several chemical raw materials are also necessary in the chip design process, and they serve various functions that will not be discussed here.
After the appropriate raw materials have been collected, some of these raw materials will need to be preprocessed. Silicon processing is critical since it is the most significant raw element. The silicon feedstock is first chemically purified, bringing it to a grade suitable for usage in the semiconductor industry. These silicon raw materials must be shaped in order to meet the processing requirements of integrated circuit production. Melting silicon raw materials and injecting liquid silicon into massive high-temperature quartz containers are used in this process.
The raw material is then melted at a very high temperature. Many materials, including silicon, have interior atoms with a crystalline structure, as we taught in middle school chemistry class. The monolithic silicon raw material must be exceedingly pure and monocrystalline in order to meet the criteria of high-performance CPUs. The silicon raw material is then rotated stretched out of the high-temperature container, and a cylindrical silicon ingot is formed at this point. The diameter of the circular cross-section of the silicon ingot is 200 mm, according to the current technique.
However, Intel and a few other businesses have begun to employ silicon ingots with a diameter of 300 mm. It is tough to expand the cross-sectional area while maintaining the various features of the silicon ingot, but it is still possible if the corporation is ready to invest heavily in research. Intel spent around $3.5 billion to build a factory to design and create 300mm silicon ingots, and the success of the new technology has allowed Intel to produce more complicated and powerful integrated circuit circuits. The factory for 200mm ingots cost $1.5 billion as well.
The next step is to slice this cylindrical silicon ingot after it has been made and verified to be an absolute cylinder. The thinner the slice, the less material you'll use and the more processor chips you'll be able to make. Sections are also mirror-finished to provide a perfectly smooth surface before being evaluated for deformation or other issues. This step's quality inspection is very crucial because it directly affects the finished CPU's quality.
The new slices are doped with chemicals to turn them become true semiconductors, and transistor circuits representing various logic functions are then scribed on them. The atoms of the doped substances penetrate the spaces between the silicon atoms, where they interact through atomic force, giving the silicon raw material semiconductor properties. CMOS processes are used in today's semiconductor manufacturing (Complementary Metal Oxide Semiconductor).
The interaction between the N-type MOS transistor and the P-type MOS transistor in the semiconductor is referred to as complementary. In electronic technology, N and P stand for negative and positive electrodes, respectively. The slice is doped with chemicals to generate a P-type substrate in most cases, and the logic circuit scribed on it is meant to mimic the characteristics of an nMOS circuit. This form of transistor uses less space and consumes less energy. At the same time, the emergence of pMOS transistors must be limited as much as possible in most circumstances, because N-type material must be implanted into the P-type substrate later in the manufacturing process, and this procedure will result in the production of pMOS transistors.
The conventional sectioning is completed when the work of integrating the chemicals is completed. The slices are then heated in a high-temperature furnace, with the heating time-controlled to generate a silicon dioxide deposit on the slice's surface. The thickness of the silicon dioxide layer can be regulated by closely monitoring temperature, air composition, and heating time. The gate oxide width in Intel's 90-nanometer manufacturing process is as thin as 5 atoms thick. The transistor gate circuit includes this layer of gate circuit. The transistor gate circuit regulates the flow of electrons between transistors. The flow of electrons is strictly controlled by regulating the gate voltage, regardless of the voltage of the input and output ports.
Overlaying a photosensitive layer on top of the silicon dioxide layer is the final stage in the preparation process. In the same layer, this layer of material is employed for different control purposes. When dried, this layer of substance has an excellent photosensitive effect, and following the photolithography process, it may be dissolved and removed using chemical procedures.
In the present CPU manufacturing process, this is an extremely difficult stage. What makes you say that? The photolithography technique involves carving appropriate notches in the photosensitive layer with a certain wavelength of light, consequently altering the chemical characteristics of the material. This technology places a high demand on the wavelength of the light employed, necessitating the employment of UV light with a short wavelength and lenses with a large curvature. Smudges on the wafer can affect the etch process. Etching is a sophisticated and sensitive procedure with many steps.
Each CPU requires more than 20 etch steps, and the amount of data required to design each step of the process is measured in tens of gigabytes (one layer of etching per step). Furthermore, if the etched drawings of each layer are enlarged many times, they may be compared to a map of New York City and its suburbs, and they are considerably more sophisticated. Imagine reducing the entire map of New York to just 100 square millimeters in size. You can see how intricate the chip's structure is just by looking at it.
The wafer is switched over when all of these etch processes are completed. Through the hollow notches on the quartz stencil, short-wavelength light shines on the photosensitive layer of the wafer, and subsequently, the light and the stencil are removed. The material from the exposed photosensitive layer is chemically removed, and silicon dioxide is produced directly below the unoccupied location.
The silicon dioxide layer that fills the trenches and the exposed silicon layer underneath this layer remains after the leftover photosensitive layer material is removed. After that, another layer of silicon dioxide is applied. After that, a photosensitive layer is put to another polysilicon layer. Another form of gate circuit is polysilicon. Polysilicon allows gates to be formed before the transistor queue port voltages take effect because of the metal substance employed (thus the phrase metal-oxide-semiconductor). Short-wavelength light eroded the photosensitive layer through the mask as well. All of the requisite circuits have been essentially constructed after another etching. After that, the exposed silicon layer is chemically attacked with ions in order to form either an N-channel or a P-channel. All of the transistors and their electrical connections are created as a result of this doping process. Each transistor has two inputs and outputs, and the space between them is known as a port.
You'll keep adding layers after this, then a layer of silicon dioxide, and finally lithography once. If you repeat these processes, you'll end up with a multi-layered stereoscopic architecture, which is where your current processor is at. The technology of metal coating film between each layer provides the conductive link between the layers. P4 CPUs have seven layers of metal connections, whereas Athlon64 has nine layers. The number of layers utilized is determined by the original layout design and has no bearing on the final product's performance.
The wafers will be inspected one by one over the following few weeks, including electrical characteristics testing to check if there are any logical problems and, if so, on which layer they appeared, and so on. Following that, each defective chip unit on the wafer will be individually examined to see if it has any unique processing requirements.
The wafer is then separated into individual processor chip pieces. Failure to pass the initial test will result in the unit being deleted. These chopped chip units will be packaged in such a way that they may be easily put into a motherboard that meets certain interface requirements. A thermal layer covers the majority of Intel and AMD processors.
A broad spectrum of chip function testing is also necessary after the finished product of the processor is accomplished. This section will create a variety of product grades. Because some chips operate at relatively high frequencies, they are labeled with the names and numbers of high-frequency items, whereas chips that operate at lower frequencies are changed and labeled with other low-frequency models. This is the processor responsible for various market positions. Furthermore, some CPUs may have flaws in their chip operation. For example, if it has a bug in the cache function (which is enough to paralyze most CPUs), it will disable some cache capacity, reduce performance, and, of course, lower the price of the product; this is Celeron and the origin of Sempron.
A final test is normally performed before the CPU is installed in the box to check that the preceding work is accurate. They are packed and marketed all over the world based on the maximum operating frequency calculated earlier.
I believe you now have a good understanding of the CPU manufacturing process after reading this. The production of a CPU can be considered the pinnacle of much cutting-edge science and technology, and the CPU itself is quite large. However, the CPU's manufacturing costs are astronomical, and it's from here that we can begin to grasp why it's so pricey.
This is a crucial aspect of the testing process. For example, whether your processor is 6300 or 6400 will be divided in this section, and 6300 is not 6300 by nature, but after testing, it was discovered that the processor can only work stably under the 6300 standard, so the processor is defined, frequency locked, ID defined, packaged, and printed with 6300.
Take AMD for example all CPUs with the same core come from the same production line. They are defined as 5600+ if they work reliably at 2.8GHz with a 1M*2 cache, If the cache is defective, the problem is cut in half, becoming 5400+; if there is no problem with the cache but the frequency can only pass the test at 2.6G, the problem is cut in half, becoming 5200+; if the cache is defective, the problem is cut in half, becoming 5000+......If a batch can't work under 3800+ conditions but can under 3600+ conditions, it's on the market. If batches can work under 3G, 1M*2 circumstances, there will be 6000+ units available. That's why mid-range processors are always first to hit the market, followed by high-end and low-end processors, and last, the factory. It may be able to save money by establishing a bottom-end assembly line for the production of bottom-end processors. Various Celeron and Sempron models have been released one after the other, with high-end pipelines being changed into bottom-end processors due to individual processor instabilities. Sempron 64 is an Athlon 64 cache reduction.
Sand: Silicon is the second most prevalent element in the Earth's crust, and deoxygenated sand (particularly quartz) contains up to 25% silicon in the form of silicon dioxide (SiO2), the semiconductor industry's foundation.
Silicon smelting： It is done at the 12-inch/300mm wafer level, and the same is done below. Through multi-step purification, silicon can be utilized for semiconductor production; the technical designation is electronic-grade silicon (EGS), and there is only one impurity atom per million silicon atoms at most. This illustration depicts the process of obtaining big crystals from silicon through purification and smelting, with the final product being a silicon ingot (Ingot).
Monocrystalline silicon ingot: It is essentially cylindrical in shape, weighs around 100 kilos, and has a silicon purity of 99.9999 percent.
Silicon ingot cutting: transversely cut into circular single silicon wafers, often known as wafers, is done transversely. By the way, you've figured out why the wafers are all the same shape, right?
Wafers: The sliced wafers are polished to a near-mirror-like finish, and the surface can even be used as a mirror. Intel. in fact, does not manufacture such wafers; instead, it obtains finished goods from third-party semiconductor companies and processes them on its own manufacturing lines, such as the current common 45nm HKMG (high-K metal gate). It's worth noting that the wafer size employed by Intel Corporation when it first started off was merely 2 inches/50 mm.
Photo Resist: The photoresist liquid injected during the wafer rotation process, similar to that used to manufacture classic films, is shown in blue. The photoresist may be placed down very thin and flat thanks to wafer rotation.
Photolithography: After being exposed to UV light through a mask, the photoresist layer becomes soluble, and the chemical reaction that ensues is analogous to the change in the film that occurs when the shutter of a mechanical camera is squeezed. A pre-designed circuit pattern is printed on the mask, and ultraviolet light beams through it onto the photoresist layer, forming each layer of the microprocessor circuit layout. The resulting circuit pattern on the wafer is around one-fourth the size of the mask design.
Lithography: This is used to create transistors with a size of 50-200 nanometers. Hundreds of processors can be sliced from a single wafer, but we'll focus on just one here, demonstrating how to produce transistors. A transistor controls the direction of current flow by acting as a switch. Today's transistors are so tiny that a single pin may hold up to 30 million of them.
Dissolving photoresist: The photoresist exposed to ultraviolet light during the photolithography process is dissolved away, leaving a pattern that is identical to that on the mask.
Etching: Chemicals are used to dissolve the exposed wafer sections, while photoresist shields the parts that should not be etched.
Remove the photoresist: After the etching, the photoresist's mission is accomplished, and the planned circuit pattern may be seen after all of the removals.
Photoresist: pour the photoresist (blue portion) once again, then photolithography, and wash off the exposed section; the remaining photoresist is still employed to preserve the material that will not be ion-implanted.
Ion Implantation: A solid material is bombarded (implanted) with accelerated ions of atoms to be doped in a vacuum environment, generating a specific implantation layer and modifying the silicon in the implanted places. Conductivity. The injected ion current can reach speeds of over 300,000 kilometers per hour after being accelerated by the electric field.
Clearing the photoresist: After the ion implantation, the photoresist is removed, and the implanted area (in green) is doped with various atoms. It's worth noting that the green color is different now than it was previously.
Transistor Ready: The transistor is almost finished at this time. For connectivity with other transistors, three holes were etched into the insulating layer (magenta) and filled with copper.
Electroplating: Electroplating is the process of depositing copper ions onto transistors by electroplating a coating of copper sulfate on the wafer. Copper ions will flow from the positive (anode) to the negative (cathode) electrode (cathode).
Copper layer: Copper ions are deposited on the wafer surface after electroplating to form a thin copper layer.
Polishing: The extra copper is removed by polishing the wafer surface.
Metal Layer: Transistor level, a six-transistor combination measuring around 500 nanometers. The specific arrangement depends on the distinct capabilities required by the separate CPU. Composite interconnect metal layers are built between the different transistors. The chip's surface appears to be extraordinarily smooth, yet it may contain 20 layers of intricate circuits, and when zoomed in, an immensely complex network of circuits that resembles a futuristic multi-layer highway system can be seen.
Wafer Test: Core level, approximately 10mm/0.5". A section of the wafer is shown here undergoing its initial functional test, which compares each chip using a reference circuit layout.
Wafer Slicing: 300 mm/12-inch wafer-level The wafer is sliced into parts, each of which is a processor's core (Die).
Discard defective kernels: wafer level. During testing, any flawed cores are discarded, leaving the preparations for the following stage intact.
Single Core: Kernel level. The core of a Core i7 is represented here as a single core cut from a wafer.
Package: 20 mm/1 inch package level. The processor is made up of three parts: the substrate (substrate), the core, and the heat sink. The substrate (green) serves as a foundation for the processor core, providing an electrical and mechanical interaction with the rest of the PC system. The heat sink (silver) is in charge of the core's heat dissipation.
Level test: The final test can identify essential processor features such as the highest frequency, power consumption, heat generation, and so on, and determine the processor's level, such as the highest-end Core i7-975 Extreme or the lowest-end Core i7-920.
Packing: According to the grade test results, processors of the same grade are put together for shipment.
Retail Packaging: Processors are either supplied in bulk to OEMs or in boxes to the retail market after they have been manufactured and tested.
Frequently Asked Questions about CPU:
1. What does CPU "manufacturing process" mean?
It means that in the process of CPU production, various circuits and electronic components are processed, and wires are manufactured to connect various components. Usually, the precision of its production is expressed in nanometers (previously micrometers), and the higher the precision, the more advanced the production process.
2. What does the micron of the CPU manufacturing process mean?
The micron of the manufacturing process refers to the distance between circuits within an IC. The trend of the manufacturing process is to develop towards higher density, and the IC circuit design with higher density means that in the same size of IC, it is possible to have a higher density and more complex circuit design.
3. Why don't you use silver as a conductor for CPU production?
The conductivity of pure silver is the best, but it is easy to oxidize in the air, and the price is also expensive, and silver oxide has to be short-circuited under weak electric conditions.
The earliest chip interconnects were made of aluminum. The simplest reason is that the metal wire is to be made on silicon, and the compatibility of aluminum atoms with the silicon lattice is better, and the process is easier to implement. It took many years for advanced processes to realize copper interconnects. Now the interfaces are all gold-plated, gold has good conductivity and is resistant to oxidation.
4. What materials are CPUs made of?
Computer CPU chips are made of a material called "monocrystalline silicon". The single-crystal silicon material before being cut is a thin circular piece called a "wafer".
5. Why are CPUs so expensive?
CPUs are expensive, so naturally, the cost will not be below. The wafers are purchased from a fab and consist of a single optical plate with epitaxial layers of a few microns on the surface. Of course, hundreds of steps are required to go from wafer to CPU. If there is a slight deviation in the whole production process, we have to start over. This process is money.
Catalogue
FPGA / CPLD
Memory
MOS 
MCU 
DSP
OCEP
Secondary 
Other 