>Once you see a modern fab with your own eyes, it will change you in a deep way. I felt a sense of compassion for this incredibly complex and valuable thing that humanity is just barely able to scrape together. Nothing you see in media can prepare you for the real thing.
I interned at Micron as an industrial engineer looking at capacity and equipment purchases. I can very much relate to that. The whole summer was my mind being blown by the orders of magnitude across the board and it's somehow it's an economically viable process.
First, we take this giant silicon crystal taller than humans and cut it into wafers. Then we take these pure materials (like 99.99999% pure) and transfer a tiny big on to the wafers in a successive layering process. Oh no, the deposition process wasn't perfectly even across the whole wafer (because of annoying laws of physics), so we'll throw it into the chemical mechanical planarizarion process to skim off a tiny layer keep the internal mechanical stressed down. Add in other mechanical, etching, lithography, and measurement processes and it gets crazy.
Wafers go through hundreds of manufacturing steps depending entirely on purpose, and with each step there's a possibility of messing up part or all of the wafer.
99% yield on a per-machine basis is sub-par in most manufacturing environments, but achieving that would be devastating in semiconductors. For sake of demonstration:
0.99^100{manufacturing steps} = 36.6% total process yield.
0.995^200 = 36.6% total process yield.
0.9975^400 = 36.7% total process yield.
Each additional 9 on yield is really expensive to add.
In school we talked about 1/1000th of an inch being kinda tight to hold on a CNC mill, with 1/10,000th needing a lot more specialized processing and time. Suddenly I was hearing about nanometer thick layers with tolerances measured in angstroms.
And the capital expenditure was nuts! 6-figures hardly gets you anything in a fab, it's really in the 7-8 figure range where you see most of your equipment landing. That will be old but still viable equipment in a few years.
Somehow depreciating all that equipment designed to make chips sold at pennies to dollars each is profitable.
As far as I'm concerned, it's black magic and truly an incredible achievement for humanity.
> Assembling a motherboard is to semiconductor fabs as flying a toy drone is to landing a Boeing 747.
I disagree. I live next to a fab and have coworkers that have worked the fabs.
Humans in fabs are taken out of the loop as much as possible. That's because when dealing with nanoscale structures, human error is simply too common.
One of my coworkers worked at the fab during a period where they had humans running the forklift that moved the wafers from one stage to the next. That was cut out because the tiny bumps caused by a human operating the controls caused imperfections in the chips that decreased yield (the metric that matters most for a fab). They ultimately removed that work and job and replaced it with robots to carefully move the wafers.
What's complex about a fab ends up being not the frontline work, but rather the layer or 2 in the back (like designing the lithography filter for a given chip). That stuff happen outside the actual plant.
> Humans in fabs are taken out of the loop as much as possible.
This. We are talking about Atom scale here. I have as much admiration for skilled mechanics as the next guy. I heard about a lathe operator at Patek Phillip who could turn an arbor to within 1 micron precision, just by listening to the pitch the cutting tool made when trimming it down.
And when chip features were on the micron scale, humans in the loop made sense. But chip feature are ten thousand times smaller than that now--4 orders of magnitude. Anything that doesn't need a Ph.D. in solid state physics to do is going to be automated.
> We are able to get silicon wafers from TI and then package them...
Wow. I've started to become mildly interested in chip manufacturing tech, and the fact that you can do that is really amazing. I'm very curious how this actually works!
> Once you have the desired circuit, you don’t have to build it out of discrete components, you also can send it to a fab
You are still going to use a very old and obsolete process, compared to the microcontroller.
As a rule of thumb, every generation of lithography that has made transistors smaller and more efficient, has also roughly doubled the NRE costs. As you move down the feature size slope, you get all kinds of useful properties, but the tradeoff is that you have to manufacture more of any given design for it to be able to make any economic sense. To the point where you can get an amazing chip that has an arm core, storage and memory in a single package that costs pennies (well, not right now it doesn't, but it did in the past and will again) and uses almost no power, so long as you can use the exact same device that is also shipped in the millions for other things too.
> Semiconductor engineering is complex, but probably not so complex that your average Math Olympiad couldn't pick it up in a month or two.
We have a chip shortage going on for months now, if all it took was two months and a bunch of smart people, those billions of dollars in chip orders would've made it happen. That tells me it has to be a bit more complex than you think.
>I'm not exaggerating when I say that designing a new technology node for mass production is like a moon program.
It's insane. I had a friend tell me about what goes on in these asml machines. Essentially the targeting system that moves the wafer around cannot have any vibration so I kid you not the platform is floating and controlled by magnets. And this is just the etching machine.
> Computer parts are always high-margin. They're made out of purified sand after all, the raw materials of chips is excessively cheap.
The purification and especially the lithography (aka making actual chips out of a silicon disk) is what is expensive. The machines are pricey as fuck - a single line from ASML alone costs 200 million dollars, and that's just the machines for the lithography, not including the machinery that makes the ultra pure monocrystals, that slices them, or that cuts and packages them, or the cleanrooms and the remaining support infrastructure. Or the extremely specialized staff you need to actually operate them.
And it's a ton of money involved with very long lead times - say you're TSMC, you get an order for a batch of chips, and even if you had capacity starting tomorrow, the 700 steps to the final product take three months [1]. That's a massive amount of capital to be stuck for months, and errors can waste all of that.
Excerpt: "the relatively low cost accessibility to semiconductor fabrication using gate arrays. Chip manufacturers who needed high volumes to recoup their capital investments of hundreds of millions of dollars (now billions) had figured out a way of producing standard product families of chips called gate arrays, which were identical except for the last few steps of the production process which defined the interconnections. This enabled computer systems companies to buy state of the art fabrication by effectively buying batches of wafers on a time-share basis from a few thousand dollars upwards, instead of the millions of dollars required previously."
> I studied microelectronics. I am aware of the technical challenges. Can you explain those challenges that are primarily non-technical?
We spent far more time buying EDA software, installing it, talking to foundries, getting the PDKs, signing NDAs, dealing with buggy EDA software, dealing with slow EDA response times, etc. than actually working on our chip.
>Things like Silicon-on-insulator, high-k dielectrics, finfets, extreme ultraviolet lithography are not innovative or new ideas?
I'm not saying they aren't, but I have noticed that the general level of openness, and following that, innovation and open-mindedness has dropped dramatically in the past decade or so, and I do have to say that the general semi industry has stayed generally innovative, and much of my criticism is directed towards the rest of the industry primarily. That being said, there is a major glacial pace.
Example of a real conversation I had with an engineer at one of the major (can't name the exact one) foundries about a device that's actually pretty close to reality:
Me: "Why don't you use this X device?"
Him: "Because it's still research"
Me: "Sure, but it's very promising, why aren't there at least any industrial research efforts to commercialize it?"
Him: "Because it's still research"
Me: -__-
SOI is innovative, but it's been held back by cost and the self-heating effect, both things that really aren't that much of a problem.
FinFETs were launched by a DARPA initiative.
High-k dielectrics I will say are the single most interesting (if not innovative) innovation in the last decade in the semi industry, although I have some bias there.
EUV is a feat to engineering no doubt, but again, my grievances aren't really focused in that area.
> unless you are willing to create your own fab or build your CPU up out of discrete transistors [...]
It’ll happen someday. I think there are enough hobbyists interested in home manufacturing (of all sorts of kinds) that we’ll eventually have low barrier to entry home semiconductor fabs. They’ll probably sacrifice performance for simplicity — I can’t imagine a home fab ever being cutting-edge — but for most applications that’s fine.
>If you want to fix electronics, find a way to make VLSI fabrication cost $500 and have a 5 day turnaround.
Agreed. The root problem is that producing chips has become so expensive and proprietary that it has become extremely centralized. I would bet that no more than a few hundred people decide the big design questions of the chips in nearly all computers produced today.
> I wonder if you can make micro machines at this level? The MEMS thing.
At this size range, though state-of-the-art MEMS (mechanical vibrating frequency filters for RF receivers in phones, accelerometers) can have sub-100nm dimensions, basic accelerometers, pressure sensors, and inkjet heads are absolutely doable.
> Not with this PDK or process, no. MEMS processes are quite specialised.
But yeah, this is the problem. Although ICs and MEMS devices are made with similar tools, MEMS usually needs processing steps that don't play nicely with the steps in an IC process (e.g., etching away huge amounts of silicon to leave gaps and topography, or using processing temperatures and materials that mess up ICs). This SkyWater process cannot do MEMS.
A more general problem is that different MEMS devices often need different incompatible process steps, so a standardized process is infeasible (though http://memscap.com/products/mumps/polymumps tries).
However, there is a tiny chance that, if we get enough detail on the process steps and leeway in the design rules, a custom layout could implement a rudimentary accelerometer or something that works after post-processing (say, a dangerous HF bath), but only with intimate knowledge of said process steps (e.g., internal material stress levels) and a lot of luck.
> ...you're analyzing in a way that's both in-depth and shallow at the same time.
I can't really dispute that. I'm no expert in the field.
> Just use diamond prices and dimensions.
Isn't that more or less what I did?
Diamond price per gram depends on the quality of the diamond. If we're gonna address an opinion that includes statements like "Think of the cost of a modern high-performance IC as if it was made of diamonds, because diamonds and silicon are both crystalline structures, and silicon is chemically much like carbon, therefore the substrate manufacturing costs are bound to be very similar." [0], then it seems that we need to look at the cost of high-quality diamonds that are used for their crystalline properties, rather than just for their hardness.
I'm not at all sure, but I would suppose that it would be far more expensive to make one high-quality diamond sheet the size of a silicon wafer than it would be to make a bunch of high-quality diamonds each the size of a CPU die, or maybe cut down a larger one. If it is, then an analysis based just on like-sized crystals would be dramatically unfair. Perhaps you know far more about this than I do? Industrial crystal production is not exactly in my wheelhouse. :)
[0] Direct quote: "Did you know that a silicon wafer is a perfect crystal, structured like a diamond? Silicon is right underneath Carbon in the periodic table, which means it shares the same outer electron shell configuration. Making that ain't cheap." via [1]
> BayBal, since you're an expert in Semi, I wonder if you can fill in the blanks?
An expert? Ahahh, I never even had formal education in the field, just been trying to enter it, and start studies in it for a few years.
My only real experience with ICs was with a company developing a fancy synchronous rectifier chip what was capable of doing few more tricks with the output waveform besides rectification, and that was mostly just hanging around, and doing complete trivialities like routing, or minimal layout wiggling. I was more useful there as a coffee porter.
> How much do you estimate the full chip manufacturing cost for this would be ?
I don't know how many wafers they buy. I don't know whether they ordered masks from TSMC, or somebody else. I don't know how short they want lead times to be. I don't know if they want to have any device inspection provided. I don't know if they have any agreements on repeated runs, or a flexible capacity purchase. I don't know if they order test, and packaging from TSMC.
From a man who was on Allwinner's original A10 chip team, I heard that the most bare bones 65nm run without mask cost, inspection, or packaging was possible at 1k wafers at $2400-$2500 in 2013-2014 by paying cash 1y in advance.
Today, I'm not even sure if clients are even allowed to, or can order masks on the side these days for latest processes.
The universal advice I heard is that you don't get into 300mm game without at least $10m, or better $20m if you have a brand new, untested design.
Before I became a developer, I worked QA in chip fabs (both in northern and southern california) in the 90s to the 00s, the writing was on the wall, so to speak.
I interned at Micron as an industrial engineer looking at capacity and equipment purchases. I can very much relate to that. The whole summer was my mind being blown by the orders of magnitude across the board and it's somehow it's an economically viable process.
First, we take this giant silicon crystal taller than humans and cut it into wafers. Then we take these pure materials (like 99.99999% pure) and transfer a tiny big on to the wafers in a successive layering process. Oh no, the deposition process wasn't perfectly even across the whole wafer (because of annoying laws of physics), so we'll throw it into the chemical mechanical planarizarion process to skim off a tiny layer keep the internal mechanical stressed down. Add in other mechanical, etching, lithography, and measurement processes and it gets crazy.
Wafers go through hundreds of manufacturing steps depending entirely on purpose, and with each step there's a possibility of messing up part or all of the wafer.
99% yield on a per-machine basis is sub-par in most manufacturing environments, but achieving that would be devastating in semiconductors. For sake of demonstration:
0.99^100{manufacturing steps} = 36.6% total process yield.
0.995^200 = 36.6% total process yield.
0.9975^400 = 36.7% total process yield.
Each additional 9 on yield is really expensive to add.
In school we talked about 1/1000th of an inch being kinda tight to hold on a CNC mill, with 1/10,000th needing a lot more specialized processing and time. Suddenly I was hearing about nanometer thick layers with tolerances measured in angstroms.
And the capital expenditure was nuts! 6-figures hardly gets you anything in a fab, it's really in the 7-8 figure range where you see most of your equipment landing. That will be old but still viable equipment in a few years.
Somehow depreciating all that equipment designed to make chips sold at pennies to dollars each is profitable.
As far as I'm concerned, it's black magic and truly an incredible achievement for humanity.
reply