You have just bought a new CPU or graphics card and started it on your PC. It seems to be going pretty cool, so try to overclock a bit. The GHz keeps rising and it looks like you have something special. Isn't it supposed to be that way?
So you're hurrying to the Internet to share your excitement about reaching the silicon jackpot, and within a few posts someone announces that you've secured a binning chip.
If you imagine an engineer rummaging around in a trash can and proudly pulling out a gold ticket, you really need to read this explanation! Welcome to the magical world of processor manufacturing and chip binning.
Waffles to die for
All chips are made of ultra-pure silicon wafers coated with metals, insulators and semiconducting materials, regardless of whether it is a standard CPU, a special graphics processor or a DRAM to become system memory. The whole process is extremely complex and the manufacturing equipment required to manufacture the latest chips in large quantities costs billions of dollars.
These discs are known as wafers, and Intel, GlobalFoundries, and TSMC produce millions of them every year. High quality tools are required to ensure that the end product meets the extremely precise plans of the engineers who designed the chips.
To keep everything as perfect as possible, the production areas of the factories are put under slight pressure to keep bacteria and dust particles from the air out of the rooms. Workers wear protective clothing to ensure that so little skin cells and hair can get into the machine.
A finished wafer is a thing of beauty and also incredibly valuable. Manufacturing costs thousands of dollars each, and the entire manufacturing process – from silicon block to product – takes months to begin. Every chip (also known as a cube) that can be removed from the CD and sold is crucial to recover the money spent on it.
To get them out, the wafer is sliced with a diamond saw, but a reasonable percentage of it is scrap, since the chips along the edge are just not complete. 5 to 25% of the wafer (the amount depends heavily on the size of the chip) is thrown away.
The rest of it is then mounted on a PCB package and possibly covered with a heat spreader to eventually become the CPU we are all familiar with.
Core (in) equality
Let's take a look at the latest processor from Intel in the core model family. The most powerful is the Core i9-10900K with 10 cores and an integrated GPU.
The photo below shows how we normally know and see such PC components, but if we could estimate the heat spreader and use a battery of tools to dive into the guts of the chip, it would be very different.
The actual CPU is a cityscape made up of logic blocks, SRAM memory, interfaces and communication buses – there are billions of individual electronic components on one chip alone, all of which work in synchronized harmony.
This labeled image highlights some of the key areas – on the far left is the I / O system, which contains the DDR4 SDRAM memory, PCI Express and display controller. It also contains the system that manages the communication ring for all cores. The interface for the system memory is located directly above the I / O area, and on the other side of the chip is the integrated graphics chip, the GPU. No matter what Intel Core processor you get, these 3 parts are all there.
The CPU cores are located between all of these. Each is a copy of the other, full of units to crack numbers, move dates, and predict future directions. On either side of a core are two strips of the Level 3 cache (the lower levels are deep in the core), each of which has 1MB of high-speed memory.
You might think Intel makes a new wafer for every CPU it sells, but a single "i9-10900" disc produces chips that may end up in one of the following models:
|model||# Cores||# Threads||Basic clock||All core turbo||Turbo boost||Total L3 cache||PL1 TDP|
The & gt; base clock & gt; 39 measured in GHz is the lowest guaranteed frequency at which the chip operates, regardless of the load it is under. The & # 39; All Core Turbo & # 39; is the maximum frequency at which all cores can run together, but do not necessarily have to stay very long. The situation is similar with & # 39; Turbo Boost & # 39;, except that these are only 2 cores.
PL1 TDP stands for Power Level 1 – Thermal Design Power. It is how much heat the CPU generates when it runs on its base clock under any load. It can do much more than that, but it will limit the speed at which the chip runs, and when plugged into a motherboard, developers can limit how much current the chip can draw to prevent this from happening.
Models with codes that end with a F have a disabled GPU. K indicates that it has an unlocked clock system (so you can easily overclock it), and T indicates poor performance. These are just the desktop CPUs – some will end up as Xeons, processors for the professional market, in the form of workstations or small servers.
So there are 19 models from just one design – how and why do so many different types become from a single chip?
It is an imperfect world
As incredible as the chip manufacturing facilities are, neither they nor the technologies and materials used are 100% perfect. There will always be some nanoscale detritus motifs, either inside the plant or deep inside the raw silicon and metals used. No matter how hard you try, the manufacturers cannot make them absolutely clean and pure.
And when you try to build components that are so small that you can only see them with powerful electron microscopes, nothing behaves exactly as it should. In the nanometer world, quantum behavior becomes much clearer, randomness, noise and other disturbances do their best to disrupt Chip-Jenga's delicate game. All of these issues conspire with processor manufacturers, and the end results are classified as failures.
Not all defects are serious – they can cause a certain section of the chip to run hotter than it should, but if it's really bad, an entire section can be completely dirty. The first thing manufacturers do is scan the wafers to examine the defects.
Machines designed to detect these problems are used after a wafer has been made, but before it is cut into individual chips. The matrices or entire wafers with problems are marked so that they can be set aside for further inspection.
But even these steps won't catch every little mistake and every malfunction. After the silicon pieces are cut from the wafer and mounted on their packaging, each one is run for even more testing.
Not all trash cans store garbage
When Intel and others sit down to check the quality of their processors, they set the chips to run at a specified voltage and clock rate. As the die goes through a series of benchmarks to load all of the different sections, the amount of electrical energy consumed and the heat generated are carefully measured.
You will find that some chips run exactly as required, while others are better or worse.
Some chips may need a higher voltage to be completely stable, the inside of other chips may generate too much heat, and some may just not meet the required standards.
Similar tests are carried out on the processors where errors have been found. Before doing this, however, additional checks are made to determine which portions of the chip are still working and which bits are junk.
The end result of this is that the useful performance of a wafer, called yield, creates a series of chips that can be categorized based on their working parts, stable clock frequencies, required voltage, and heat dissipation. The name for this sorting process? Chip binning.
In fact, matrices are not thrown into large plastic containers – the term comes from statistics, in which a distribution of numbers can be organized into groups called containers. For example, population surveys on age distribution use the containers 0 to 5 years, 6 to 10, 11 to 16, and so on.
The same applies to wafers, and in the case of our i9-10900K example, some of the bins relate to the number of working cores, the clock frequency range in which the CPU is stable, and the heat output at a specific clock.
Let's say a Core i9-10900 chip is thoroughly tested and has some serious shortcomings, as stated above. Two of the cores and the GPU are so badly damaged that they simply no longer work properly.
Intel would then disable the broken sections and mark them as chips for the Core i7-10700 series, especially for an F model. But then it needs to be tested for clock, performance and stability. If the chip achieves the required goals, it remains an i7, but if it cannot fully achieve these goals, another 2 cores can be deactivated and the die can be used instead for a Core i5 model.
All in all, chip binning massively improves the yield of a wafer because it enables more chips to be used and sold.
For 10th generation core processors, Intel has a separate wafer design for the Core i5, i3 and Pentium / Celeron series. These start as 6-core chips and are then combined in 2-core offers.
Product demand can often exceed production capacity, which is why the 10 core wafers are used to fulfill orders. When tools work perfectly, sections are sometimes turned off to ensure that factories are performing well. That means it's a silicone lottery that is about which dice you actually get when you buy a particular model.
All in all, chip binning massively improves the yield of a wafer because it enables more chips to be used and sold. Without them, Intel's dumpsters would be overcrowded with silicon scrap.
Aren't bundled CPUs special?
Like so many terms in the computer, chip binning has become synonymous with something other than its original meaning. Online shops sometimes sell handpicked special CPUs (those that go to insane levels or run cooler than Pluto's surface) as "binned CPUs". The reality is that all chips are grouped together simply because they have to.
Of course, nothing prevents a retailer from bundling the chips they buy: binned binned CPUs, anyone?
AMD and Intel processors have to be bought in bulk (compartments with dozens, if not hundreds, of chips), and you can sit down with a test computer and check each one – overclock or under voltage, record their temperatures, and soon. The best of the batch could then be resold as something special, and the retailer could rightly classify them as "binned CPUs". Of course, all of these additional tests cost time and effort, so the product's retail price is increased to reflect this.
Are these binned chips special in any way? Yes and no. Every single chip used in your PC, phone, car, etc. has gone through a selection process. This is just another phase in the manufacture of all microprocessors and DRAM chips. That means your beloved CPU or GPU, which runs surprisingly cool or overdrives like crazy, is just another cube of one of the hundreds of thousands of wafers produced by factories around the world.