I had this questions come into my head becasue I was watching a video of someone zooming into a microchip and they pass a human hair and continue zooming in an incredible amount. I’ve heard that some of the components in microchips are the size of DNA strands which is mind boggling. I also watched a video of the world’s smoothest object in which they stated that normal objects are no where near as smooth because if you blew them up in size the imperfections would be the size of Mount Everest. Like if you blew a baseball blew up to the size of earth it would have huge valleys and mountains. It wouldn’t be perfectly smooth across. So my question is how are these chip components the size of DNA not affected by these imperfections. Wouldn’t transistors not lay flat on the metal chip? How are they able to make the chips so smooth? No way it’s a machine press that flattens the metal out that smooth right? Or am I talking about two different points and we haven’t gotten that small yet?
In: Engineering
Branch Education has an insanely detailed video on the production process.
As someone else has said, “photolithography” is like film developing. Semiconductor manufacturing is layers upon layers of masking, exposing, etching, and depositing the next layer of material down, built on top of a slice of a perfect crystal of pure silicon (a wafer). There are hundreds to hundreds of thousands of copies of the same die on each wafer. At the scales required, even specs of dust can create a defect, which is why it is done in clean rooms with people wearing head-to-toe suits; the suits don’t protect the person from the product or process (not primarily), they protect the product from the human. Dozens of layers placed on top of each other extremely precisely; any misalignment, or mistake in processing at any step, will mean some to all of the dies will be bad. You can’t eliminate every potential source of failure, and you can’t just look at it visually, so it is electrically tested before it gets shipped to a customer (does it work like it’s supposed to?). What an “acceptable” amount of loss (how many are thrown away) depends on a large number of factors, but primarily money (how much money you make on each one times how many you sell). If you’re Intel, and your part is complex and worth a lot of money, it makes sense to sort out (“binning”) and sell the ones that aren’t perfect (but usable) for less money. But for simpler parts, it either works 100% or it doesn’t. Even then, the aim is <10% of the parts produced are thrown away, and that’s pretty common for a mature device.
there are a lot of imperfections in microchips.
If you look at the AMD Ryzen CPUs for example. Ryzen 7 has 8 cores, while Ryzen 5 has 6 cores. During production they make one type of chip, and if all 8 cores are functional, they sell it as Ryzen 7. if only 7 cores out 8 are working, then they deactivate the 7th and sell it as Ryzen 5 with 6 cores.
In addition to the other answers memory structures are built with redundant rows that can be switched in during product testing so if a fault is found it can effectively be patched.
Error correction is used so that temporary errors caused by cosmic radiation and the like are corrected automatically and this can also correct permanently failed individual memory cells.
Those structure sizes you read (nanometers) used to represent the actual dimensions on the chip. At one point like 10 years or so ago, it became impossible to shrink much further, so they improved the structures. Marketing still labeled it to be smaller to represent the gain in computing power per chip.
How it’s done is called optical lithography, and although I did that and could tell you many things about it, better watch some explanations on YouTube first. It’s a bit beyond a eli5.
Note that there still are quite a number of imperfections. Which is why they manufacture all CPU/GPU chips to the highest spec they offer, and those chips who can’t deliver simply become the lower tier models.
It’s all about statistics. It is technically nearly impossible to make chip-wafers with imperfections. There is always a certain error rate spread all over the silicon wafer throughout the making. BUT if you make A LOT of wafers with a ton of i9 chips on it, there are some perfect ones on it.
These perfect ones are sold as i9, the slightly imperfect ones are downgraded (imperfections marked and programmed as unusable gates) sold as i7, the more imperfect ones as i5 and then there are lots of i9 that are so imperfect that they are downgraded to still make good i3 chips.
One of the methods used for making chips is called photo lithography. In a way, chip manufacturing borrows development practices from photography.
They cut very thin slices of silicone and use it as a semiconductor. Semi-conductor is a material that is semi-okay at conducting electricity. Some metals/materials are better than others at this job. Then they add layers ontop the entire waffer and use a special film that is sensitive to light and flash it to imprint the design. Then they will add another layer of metal and this covers the waffer completely. Then they use gas and a chemical bath to remove the excess metal and leave only traces where the design was imprinted.
Since the work surface is so tiny they add extra material then use fancy methods to remove what’s not needed.
Really basic comparison but say you drew a smiley face picture with a gluestick. Then dumped glitter all over it and flipped the paper over. You’d be left with the glitter that stuck to the glue creating your smiley face design.
They arent.
Let me tell you a cool secret.
Lets say intel comes out with a brand new series of CPUS, and they introduce a new i11 chip, the most advanced chip yet, well, because making microchips is so effing hard, even on a damn automated manufacturing machine, there is a huge amount of errors in them, so any chip that dont fall within lets say 15% of what an i11 chip is suppoed to be gets marked down to an i9, and they repeat the process anything under 15% of the power of the i9 becomes and i7, then and i5, and then an i3.
So whats the difference? None really, they are all made identically, on the same conveyour belt by the same machine, with the same process.
But cause its so difficult to make that jazz, and instead of binning the hella expensive cpus that did not reach the standard, they turn it down a version number.
Again, this is simplified as that just happens if they work in the first place.
A good example of this is back in the 90s, when a lower powered cpu was more popular, and intel was running outta stock, so they took higher powered one, shut down the power to the lower one, and sold it at lowered one prices.
But whats so funny about that? Well a guy figured out they could remove that powering down block, and get the higher power cpu for the lower power cpu price.
It comes down to quality really, and the error rate is way higher then you would think.
15% is also an example number.
Let’s imagine the chip being made is a city.
The city works by getting cars from homes to offices. The homes and offices are connected by many different roads and highways.
The city is split up into many similar neighborhoods.
When there are problems with a road or neighborhood, those areas are turned off or destroyed. The city becomes less efficient.
The city does less work, and is slower.
Since many cities are made all at the same time to save on costs, each city is different in terms of efficiency.
Cities with 16 working neighborhoods are sold for more. While some cities only 1 of 16 work cost much less.
Each city still takes up the same amount of space.
Over time they are able to build more neighborhoods in the same amount of space.
They can have taller buildings, skinnier roads and houses.
To build the city very fast, there is a giant stencil/mold. Each stensil represents a feature like all walls on the first floor. After the stensil is placed, it rains concrete on the ground.
Sometimes you need to dig a trench or a valley in the city to put different material. Then they cover what you want to keep, then flood the city with some acid.
If you want a metal highways, they use a projector and blast the surface with metal until the metal gets stuck in the ground.
There are a lot of different methods. But it mainly involves a lot of layers of adding and subtracting materials.
Hopefully the above explanation gives a different take with less computer jargon.
When Intel makes say a 12900k it might be defective but still work well enough and so that is where i5s and i7s come from. They are just i9s with defective cores sold as i5s
Latest Answers