Digital Chinese New Year Challenge For the old Diyer, it may be the classic overclocking products of the year, such as the first-generation Celeron 300A, the famous "Tualina" ......However, for new PC enthusiasts, overclocking is not only getting harder and harder, but also more and more processor models, why?
As a result, senior DIYER will call the current situation "ashes", thinking that from CPU to GPU, this phenomenon of becoming more and more difficult to overclock is because the chip began to "leave the factory that is ashes", there is no potential to be mined, and the odds of getting an ordinary version of the chip with a good "physique" are almost the same as winning the lottery?
The reason for this comes from wafers - all chips come from semiconductor "discs" of different sizes and manufacturing processes, which have created the miracle of chips in the past and caused the current breakthrough dilemma.
Semiconductor wafers are the foundation of electronic design and manufacturing. These thin, disc-like substrates are not only components, but also canvases for unfolding semiconductor manufacturing. Although the materials of semiconductor wafers are roughly divided into four categories: silicon, gallium arsenide, silicon carbide, and indium phosphide, silicon raw materials are mainly used to manufacture various processor chips.
Wafers are made of monocrystalline silicon material of high purity. The cylindrical ingots of high-purity silicon monocrystalline semiconductors are made by extracting seed crystals from the melt. Impurity atoms that act as donors are precisely added to the molten intrinsic material to ensure the doping of the crystals, thus transforming the semiconductor into an n-type or p-type semiconductor. The ingots are then sliced to form wafers.
Then, according to the design of the chip, the chip or chip core is formed on a piece of high-quality wafer of different sizes through photolithography imaging etching and segmentation.
Next, the machine cuts the chip area, assembles it (plugs in other semiconductor components such as the cache depending on the architecture), fits it on the PCB, and then encapsulates the enclosure to form the finished processor.
One piece of Intel's 9th Gen Core processor 118-inch (300 mm) wafers.
In order to remove the CPU GPU core components from the wafer, it is necessary to use a diamond saw to cut the semiconductor wafers into thin slices, but because the cutting of the edge of the wafer is incomplete, a certain percentage of the wafers will be completely scrapped. 5% to 25% of the wafers are thrown away – the exact number depends largely on the size of the chip's core wafers.
In the early stages of wafer production, the manufacturing cost per wafer is thousands of dollars, and the entire manufacturing process from ingot to product can take months from start to finish. How many chips can be removed from these wafers is critical for principals (e.g., Intel and AMD) and foundries (e.g., Samsung, TSMC) to recoup manufacturing costs.
In the middle of production, due to the increasing number of transistors in chip designs, the entire wafer area had to be reduced if more processors were to be produced on the same size wafer platter, and to improve performance. The principle of shrinking is the same for everyone – to reduce the size of a single transistor as much as possible. At present, the mainstream process is already 5nm, that is, each transistor is 5nm wide.
For such a small transistor, the processing etching must rely on extremely high precision lithography machines, which is a huge fixed cost. Moreover, the lithography machine does not guarantee complete accuracy and error. These errors can lead to differences in component processing accuracy and quality in different parts of the same wafer – small differences that can lead to huge performance differences.
And the wafer matrix itself will also bring differences: there will always be some nanoscale impurities in the silicon raw material and deep in the metal. No matter how hard a manufacturer tries, it can't be completely clean and pure. The quality of raw materials is so varied that even the most accurate lithography machines cannot achieve complete consistency.
In the nanoscale world, quantum behavior becomes more pronounced, with randomness, noise, and other minor glitches doing their best to disrupt the subtle chip game. All of these issues are detrimental to the processor manufacturer, and the end result is classified as a chip defect.
Before the wafer is diced, the manufacturer performs a "full inspection" and marking through several test methods in the diagram above.
Not all of the chips that are scanned are bad, they may just cause a certain part of the chip to run too hot, but if it's really serious, then the entire wafer will be completely scrapped.
Once the wafer is cut from the wafer and mounted on the package, each chip undergoes more testing. When checking the quality of the processor, the chip is set to run at a set voltage and a certain clock speed; The chip undergoes a series of benchmarks designed to apply pressure to all the different parts, while careful measurements are taken of the amount of power consumed and the heat generated. Some chips work exactly as required, while others are better or worse.
A similar check is performed on the processor where the defect is found, but before that, additional checks are carried out to see which parts of the chip are still working and which bits have been scrapped.
The end result of this is that the wafer'sUseful output (known as yield).A series of "dies" are generated, which can be classified according to the effectiveness of their functional parts, the stable clock frequency that can be achieved, the required voltage, and the amount of heat generated. This is called chip sorting.
Similar checks are performed on chips that are found to be defective before packaging to determine whether they are genuine scrap or can be resorted.
Let's assume that a full test of a Core i9-13900K chip reveals several of the critical flaws mentioned above. A certain number of these cores and integrated graphics are corrupted to the point where they cannot function properly. At this point, it will be disabled"Stuck"section and label it as a Core i5-13400F series chip. Clock speed, power, and stability are then tested. If the chip achieves the required targets, it will continue to be used as an i5 non-"kf" chip.
The next package is to connect it to the peripheral components of different capacities according to the specifications of the 13400F: a lower capacity SRAM cache, a smaller number of high-speed bus channels, and other ......Complete final testing and packaging.
In short, the more grades are divided,The higher the wafer utilization, the lower the overall cost!
So, in fact, your 13700 may be produced on the same wafer as your friend's 13100!
In all respects, chip sorting can be greatly improvedProcessorbecause this means that more chips can be utilized and sold. Without this technique,FoundryIt's going to be full of scrap wafers, and the processor you get could be three to five times more expensive than it is today!
Therefore, processors are becoming more and more difficult to surpass, on the one hand, it not only reflects the increasing limitations of Moore's Law in the semiconductor industry, but on the other hand, it also provides traffic conditions for more cheap ordinary chips to enter the market and provide cheaper products for the public!