Yield is a term widely used in the semiconductor industry. But not many know what this term really is and how it can be calculated. In this article we are going to analyze all this in detail, so that it stops having secrets for you.
More related terms and that you are sure to be interested in to complement information:
- Dark Silicon
- Lithography
- chiplet what is
- Chiplet and future alternatives
What is Yield?
We know that the economy and productivity of the chips will depend on several factors (dust particles that can ruin the circuit, crystalline defects in the wafer, problems caused in the manufacture, misalignment of the masks, defects and aging of the masks, aging of the chemical processes, etc.) It is time to include the concept of yield (yield or production factor), where Ng is the number of good dies and Nt is the total number of dies per wafer, the formula is:
Nt can be calculated by the diameter of the wafer (D), the distance wasted to the edge (De) and the area of the die (Adie), being the formula:
The yield of the die is calculated using: defects per unit area (DPA, which depends on the manufacturing process, being higher if it is new), the area of the die or die (Adie) and an alpha grouping value (alpha, usually be between 2 and 5). The formula looks like this:
On the other hand, we can calculate other types of yield. Three examples are:
We can also calculate the factor of production using this other formula:
On the other hand, this factor is also related to another concept that you should know, binning.
binning
The quality of a chip is classified according to the binning . CPUs, GPUs, RAM, SSDs, and other similar chips are rated based on this method. This technology is most applicable to CPUs and GPUs around the world. To understand binning, you first have to understand how such a chip is made. The chips are made using wafer lithographic processes, so that a single wafer can contain anywhere from a few dozen to hundreds of chips. A focused ultraviolet laser beam is focused on the center of the wafer, making chips closer to the center better defined and nearly identical to the pattern manufacturers are looking for, while those further from the center are less precise and suffer typical manufacturing defects due to reduced focus of the laser beam.
Today, it is easy to understand why chip producers need to carry out a selection of chips . It’s also easier to understand how chips are made once you understand how they’re made. The core chips on the wafer are often of better quality than the others, as a result of various conductivity tests. Once this process is finished, the chips are classified in different quality trays to avoid discarding those of lower quality. The wafers are sorted into different quality trays by binning, which is beneficial for buyers as the manufacturer does not have to throw away the leftovers.
Graphics card manufacturers, such as NVIDIA, ask chip manufacturers, such as TSMC, for a set number of chips in any production node. TSMC performs the necessary procedures after producing the requested wafers and bundles them to provide information to NVIDIA. This procedure is used on any CPU or graphics architecture. As a result, NVIDIA offers various items with the same graphics chip but with different ranges.
For example, many do not know that Intel does not create a different microarchitecture or a different chip design for all its series or SKUs , but rather simply makes one and then binning catalogs them to label them in one way or another and send them to market. Imagine that a 12th Gen Intel Core i7 processor wafer is being manufactured. Well, imagine also that said wafer is 300 mm in diameter and that about 250 chips can be produced on it. Discarding the defective ones that are completely unusable, there will be others that may have a core or two damaged, but the rest of the cores are intact, or some that can work stable at 2.3 Ghz but not at 3.8 Ghz, etc.
Instead of wasting those defective chips, what Intel does is binning and labeling them. For example, the ones with all the cores intact will be labeled as Intel Core i7, and within these there will be some that can run at higher clock speeds, and others less high, so the SKU will be varied according to their features, that is say, 12xxx, with xxx being the numbers that accompany the generation of this processor.
On the other hand, there will also be chips that have two defective cores or more, and those can be classified as Core i5, Core i3 , and so on, even reaching the Intel Pentium and Intel Celeron, which are the ones with the greatest manufacturing problems. . The same goes for GPUs, or memory chips, etc. All the functional ones are used, even if they do not perform the same or offer the same benefits. In this way it is more profitable to produce these chips in such a complex and expensive industry.
Sometimes there are some specific microarchitectures, for example an Intel Atom is not a broken Core i7, but a totally different microarchitecture. The same goes for the Intel Xeon or the Intel Core i9. And this also applies to any other firm in the sector…
The build quality of a chip greatly influences its ability to overclock , so binning is critical. Overclockers have spent many years searching for the best batch (product identification number), which identifies the highest quality chips. This lot consists of a set of numbers and letters that appear on the IHS of the chip on its packaging. You will be able to recognize a good batch of chips after trying several of them if you have the batch number. The lot usually includes data about the factory in which the chip was made and its date.
We recommend reading our guide on the best processors .
Now you know what Yield is and how it can be calculated, in addition to the concept of binning, to understand a little better what they are selling us..