Friends who have paid attention to our bad reviews will definitely remember that last month we discussed the glue chip made by Apple - M1 Ultra.
At that time, we said that Apple's glue chip was a compromise, because the larger the chip, the more "scrap material" was wasted around the wafer after cutting.
It's like a knife 999 cake, this thing is absolutely painful if it is wasted.
Having said that, I found that there were small friends in the comment area to discuss:
So why does the chip have to be squared? If it is made into a triangle and a hexagon, it will not cause the wafer to be wasted???
I have to say that the big guys' ideas are very creative, but in fact, doing so will increase the difficulty of cutting and affect the yield of the chip.
When the time comes, it is not as cost-effective as wasting a little scrap material.
Therefore, the idea of "special-shaped chip" is actually not feasible.
However, at this time, there may be a small partner to continue to find another way:
In order to ensure that the chip is well cut and kept square, what if the silicon wafer at the bottom is directly made square?
Turn the "wafer" into the "crystal square", and there is no waste when cutting the chip!
Well... This method is also feasible if it is feasible, and it is not feasible to say that it is not feasible.
But to make it clear, I have to ask the big guy this question first:
Have you ever seen a square cucumber?
> / There is a wafer before there is a wafer
As we all know, chips are etched by wafers, and wafers are made of high-purity sand...
Oh no, high purity silicon is composed of elements.
After a series of high-temperature reduction reactions on ordinary quartz sand, after chemical purification, we can obtain a high-purity silicon rod such as the following figure.
But this is only the first step, such a silicon rod is made of polysilicon, which cannot be used to produce wafers at this time.
As shown in the middle of this figure, because of the various rough chemical reactions that have been undergone before, the internal silicon crystal structure framework is uneven.
Single crystal Polycrystalline indefinite ▼
It is full of various asymmetrical structures.
In order to eliminate these asymmetrical internal disorders, we also need to treat polysilicon and transform it into a structurally stable, electrically good monocrystalline silicon that can be used in chip production.
At this step, it is actually to produce the wafer of the Eight Classics of Zheng'er.
At present, the commonly used process in the industry is called the Tchai kraski method, and there is also a very popular name - straight pulling method, the market share is about 95%. Our common logic chips and memory chips are basically produced by this method.
The specific changes that have occurred, we can see a rough idea in this picture.
The first is to heat the polysilicon just obtained in quartz to a molten state, and then implant a monocrystalline silicon "seed".
As soon as this melted silicon solution touches the monocrystalline silicon seed, it can begin to grow in an orderly manner at the tail of the silicon seed.
By controlling the speed of rotation and the rate of lifting, we can get cylindrical monocrystalline silicon rods of different widths and lengths.
Attention! It's spinning!
Still a little friend with no concept, let's imagine a marshmallow stall on the side of the road.
The process of pulling the crystal rod is similar to the starting style of turning marshmallows.
That is to say, the current mainstream monocrystalline silicon production process determines that the silicon rod has a high probability of being round. I had never seen a square marshmallow anyway.
As for the back, it is to pinch the head and tail of the crystal rod, and polish the side smoothly.
Then slice it little by little like a sausage, the raw material of the wafer, and the silicon wafer is born.
>/ Cut over, but not over yet?
In fact, at this point, the question about the shape of the wafer is not fixed.
Because although the straight-pull method is the mainstream monocrystalline silicon manufacturing scheme, in fact, in addition to it, there are also schemes such as the zone melting method, and the square crystal rod is still theoretically possible.
But why do you have to use a circle? In fact, the problem also involves a more follow-up design process.
After we polish and polish the sliced silicon wafers, we also need to apply photoresist to them for formal lithography.
In general, the thickness of the photocoagulant film varies from 0.5 to 1.5 um, and the uniformity must be within plus or minus 0.01 um.
This precision can certainly not be solved by our hands...
Now the commonly used scheme in the industry is "glue flinging". That is, add photoresist to the center and then rotate the wafer.
Then by constantly controlling the speed to shake off the gel, we can finally get a uniform flat layer of photoresist.
That is to say, if we make the silicon wafer into a square and then rotate it, it may be that some corner glues accumulate more, and some corner photoresists are less.
The uniformity of the corners is greatly reduced at once, even if it becomes square, but the corner part may still have to be thrown in the end...
And, in addition to the problem of photoresist, there is a more fatal thing.
That is, under everyone's "convention", circular silicon wafers have long been the industry standard.
The corresponding lithography machine, automatic production line, etc., are based on the premise of "wafer" to design.
If anyone wants to complete the square crystal rod now, he not only has to face the problem of "pulling out the square crystal rod", but also has to renovate and redesign the entire subsequent production line.
So, can the wafer be made square? Yes, but not worth it.
>/ Strictly speaking, this is not wasteful
In this way, the "wasted" areas on the wafer seem to have to exist.
But let's just say, is there a possibility that this concept of "waste" is our preconceived notions.
Is it possible that the edge part of the wafer itself should be wasted?
In fact, in the process of cutting, chamfering, grinding and so on in the production of silicon wafers, the edges of the wafers will accumulate a lot of edge stress.
As a result, the structure at the edge of the wafer is relatively fragile.
Even if the edge area is used to make chips, the yield rate is not very guaranteed.
So let's say... Don't look at the wafer to be round, the chip to be square.
But in a sense, the combination of the two pairs is still a very good match.
But of course, even with this explanation, there is still some waste in the area inside the edge of the wafer - but the waste area is not as large as we think.
And if you really want to solve this small part of the waste, in fact, everyone is now using the same method.
That is to align with calculus.
The wafer is getting bigger and bigger, the chip is getting smaller and smaller, and this square is not... Put together like a circle?
As for how to make the wafer whole large, and how to make the chip small under the premise of ensuring performance, this kind of head-sized problem, or leave it to the engineer to find a way...