Indistinguishable from Magic: Manufacturing Modern Computer Chips (2012) [video]
youtube.com
youtube.com
I've always wondered: could we have discovered/invented photolithography earlier, and thereby started the integrated-circuit revolution earlier? As far as I can tell, there was nothing holding someone back from inventing it in the 1800s, even: they had access to all the relevant chemicals, and (in some countries) lenses were already precision-ground enough to serve the purpose down to some pretty tiny process-nodes. We wouldn't have had digital logic (no transistors), but we could have been making "electronic" watches, tiny AC-to-DC power converters, and other analogue ICs even back then.
But heck: suppose we had known the principles behind semiconductive materials back then, too. We could have passed right by the vacuum-tube era and started in on inventing digital ICs (and boolean algebra.) What could the 1800s have looked like then?
(Is there any well-known science-fiction story exploring this premise? I might give a shot to writing it, if not...)
Anyway, perhaps the problem was more that (almost) nobody back then saw the potential of this technology.
But to answer your second question. I don't think we could have skipped the vacuum-tube era. It's because we already had applications for those sorts of electronics (and unreliable mechanical relays that were replaced with transistor switches) that researchers realized the potential for transistors in the first place. Even after the transistor was invented there was still a lot of research necessary to make them feasible in practice. The first one was germanium if I recall correctly and it took a while before silicon transistors were viable.
Edit: Also, sort of unrelated, non silicon semiconductors are still used a lot because they can switch faster, and different color LEDs use different semiconductors because they emit different colors of light, which is also related to why different color LEDs have different voltage drops.
Edit: what you buy as a typical discrete transistor is in fact integrated circuit which contains tens to thousands of parallel connected transistors.
Also I believe that photolitography-like processes were used before semiconductors (and I would not be too surprised if production of at least some mass manufactured valves involved photolitography).
For typical BJTs it depends on how you define "separate transistor". The structure in typical BJT has one contiguos junction, but its geometry is non-trivial enough that you can well think of it as multiple paralleled transistors.
Even on the basic science level, Maxwell published his seminal treatise on classical electromagnetism in 1873, and it wasn't until 1881 that Heaviside published them in the form we know them today and that sort-of made people understand them. Semiconductors were sort-of known already in the early 1800s, but it wasn't until the, say, first half of the 20th century when we really started to understand them with the development of solid state physics (both theoy and experimental methods such as x-ray diffraction etc.).
So yes, while some aspects of lithography could have been developed in the 1800s, I think there was quite a huge gap between that and a whole lot of other things needed.
The logic is relatively easy, but the materials science needed to decide what to lithograph, using which materials, and in what order, is post-WWII, and probably couldn't have been invented much earlier.
It takes a lot of work to work out ideal doping factors and diffusion rates, estimate charge distributions across junctions, calculate propagation speeds, and eliminate leakage.
But more - the chemistry of semiconductor doping and deposition is often rather nasty, and it would have been a huge challenge to industrialise it before WWII.
1) the theoretical knowledge of how the devices worked, which enabled the creation of steps in the process that resulted in fully working devices
2) the ability do do each step repeatably and precisely and reliably.
If it happened before the Battle of Jena, we might have found ourselves in a world of technologically-ensconced monarchies.
- The electric arc furnace, from the late 19th century, enabled the large scale production of crude silicon.
- The crystal bar process, from 1925, is similar to the Siemens process developed in the 1950s for refining purified volatile silicon compounds to electronic-purity elemental silicon.
- The Czochralski process, from 1915, permitted the growth of large single crystals of purified silicon.
There's a great review article from 1981, available through sci-hub, if you want more details about the history of purified silicon electronic devices:
"Twenty Five Years of Semiconductor-Grade Silicon"
DOI: 10.1002/pssa.2210640102
https://www.youtube.com/watch?v=b--FKHCFjOM
Photolithography in general substantially pre-dates the development of the IC. We were using it to produce printing plates in the 19th century. Breakthroughs in chemistry made photolithographic PCB manufacture possible by the early 1940s. It took us another 20 years and a huge array of discoveries and inventions to finally start producing useful ICs.
Development of IC manufacturing required a very broad tech tree - better chemistry, better optics, better materials science, better process engineering and, most crucially, the invention of p-n junctions. Without the p-n junction, any effort to etch a practical integrated circuit is pretty much futile.
Happy to answer questions, there are a couple of errors and misspeaks. Note I'm not in that field anymore (and,thankfully, I'm in much better physical shape).
Plus most of those guys work 3/4 12hr shifts a week...great schedule. No work goes home with you.
Each is different...a college degree will never hurt and a master's is pretty normal for fab engineers and a PhD for process developers.
As for schools it definitely depends in a similar way, but most top technical schools will get you recruited. A standout is Rochester institute of technology which has an undergrad degree in microelectronics.
Now I'm working for a tech company in Taiwan, who do the next process step: dicing the wafers into individuals dies, and packaging the dies into chips. There's a trolley downstairs with over 1 petabyte - stacked high with trays of 16GB memory cards from the testing machine.
Unfortunately I only managed to get a job writing monitoring software, equipment control drivers, and business planning software, so I don't get to handle the wafers myself.
In my opinion, you'll need to learn Chinese if you want to work for the lithography companies, or work for the German/Japanese/Swiss companies making the machines that take in the FOUPs. If you want to simply be a technician, most of the migrant workers who operate the machines here come from the Philippines and get paid very little while working very long hours with no path to promotion (all the line managers are Taiwanese). On the bright side though, the technicians speak better English, so it's easier to talk to them directly!
I enjoyed the circuit design/FPGA programming courses in university, but the projects to design ASICs usually take several years, so I couldn't get work experience thorough summer jobs (which is how I ended up in software).
Right now, finishing up a PhD in engineering education where I study the development of engineering students' beliefs about knowledge and how those interact with their development of entrepreneurial competence.
Also you mentioned a background in metrology. How do you inspect something a few atoms thick?
A bunch of different ways. Depends on the material being measured but measuring sheet resistance (effectively, you can back calculate the thickness by measuring the resistance of a square of known size) and light scattering off the surface are the most common.
I uploaded it at the time as I wanted to subtitle this for my deaf (and also non English speaker) teacher, but I lost the password for that YouTube account mid progress. Maybe I should try to reset it.
Thank you!
1) thank you for the effort to make cons more inclusive
2) would love to help...would having the slides help?
What I remember from class is that it's a fan on the roof, blowing air through a very fine filter.
The most magic part for me is that this still works even though the PM2.5 pollution outside is terrible (the AQI is over 150 today, which causes me to sneeze and get red eyes, and is known to increase the risk of cancer, so everyone wears dust masks outside).
Most cleanrooms are now class 10 (meaning 10 half micron sized particles per square meter) compared to a class 1,000 operating room. The wafers are all sealed in the 'clearnoom within the cleanroom' spaces of tools and FOUPS.
I have heard of fabs having problems fro msources as strange as nearby farms and earthquakes half way around the world. The filtering (ultra-HEPA filters) will absolutely clear up your alergies almost over night.
Incredible.
Or writing an essay with the pencil tool in mspaint while using a 50-radius sized brush.
To me the most magical part is not the manufacturing process, although that is very special, but how a team of engineers designed something which is made up of hundreds of millions of transistors, each in a specific location for a specific reason. How is all of that done? How do they keep track of things? Is there special software?
Engineers at the foundry will range from those focused keeping the equipment aligned to produce functional structures on/in silicon over time, to engineers that design standard libraries of components to work at a given 'process' (the steps in going from wafer to chip for a given set of parameters/attributes).
The structures in silicon (or other semiconductor material) are the transistors mainly. These are created by embedding impurities in geographic locations and patterns in the silicon that change it's conductive properties. Embedding these impurities is called doping.
The structures on top of the silicon are conductors and insulators used for interconnecting all the transistors.
There will be engineers at the foundry (or a company that works with that foundry) that take a "netlist" from the customer and convert that into files that can be used by mask company to make masks for the layers of conductors and insulators the lay on top of the transistors, and also the patterns for doping the silicon.
Engineers at "tool vendors" will build software that takes a high level description of the desired functionality of the chip and turn it into the netlist that foundry engineers can use. The netlist is a decomposition of the high level description (such as x <= a * b) into the components in the "standard library" for that foundry. This library will contain basic components like AND gates and OR gates, but also more complicated things like IO pads and RAMs, and things like multipliers.
Then you have design and verification engineers at a given company that wants to build a chip. These tend to stay at the higher level, working in VHDL, Verilog, and SystemVerilog. Which are all pretty archaic by software language standards. And what is considered high level in this arena is about the C language level or below. So not so high from a software point of view.
I just spat this out off the top of my head, so apologies if I made mistakes.
Both the presentation style and the content itself is really engaging.
But watching this made me feel like the computer I'm typing this on is some space age alien tech. It's a little mind blowing how inexpensive it is to purchase stupidly complex hardware.
After watching it, I concluded we are gods for manipulating matter at this level :)
Its something I'm a big fan of. A talk like this is inherently a ton of information in a short time and I try to be inclusive and accessible by being conscious of how others process information.