Inside TSMC, The world's largest chip factory...

[BrianHG]

Just listen to those insane specs of the company's size... 
Not to mentions the insane small size of the chips...

[Red Squirrel]

Pretty insane engineering is involved and they only really scratch the surface.

It's crazy to think that some of the scales being worked at are smaller than the wavelength of light.   

For something this complex though you would think there would be more contingency setup in the industry.  If something were to happen to that factory the world would practically halt for years.  The chip shortage is already bad enough but from my understanding they are still producing chips just not at full capacity, so there are probably still some critical products being made. Guessing priority is going to things like medical equipment, critical infrastructure equipment etc right now until they can get production back to normal levels.   Things like GPUs and cars will have to just wait.

[tszaboo]

"And as the last step in the production, we take specifically the chips that you need for your design, and place it into this industrial grinder to safely shred it into small pieces. The chips that are not shredded this way are sold to unauthorized resellers, who will charge 50 times the market price for it."

[coppice]

"And as the last step in the production, we take specifically the chips that you need for your design, and place it into this industrial grinder to safely shred it into small pieces. The chips that are not shredded this way are sold to unauthorized resellers, who will charge 50 times the market price for it."

It wasn't technology that allowed TSMC to dominate the foundry industry. It was integrity. In its early days it was technologically behind people like UMC (its main rival in Taiwan). However, customers kept finding the other foundries using some dodgy business practices. They might not isolate customer's designs from each other, allowing one company's technology to leak into another. The might just take a customer design and start making it as their own. TSMC's integrity created big sales. Big sales created a big R&D fund. Big R&D has taken them to the forefront of the industry. There are instances where some dodgy things have happened in TSMC fabs. However, the ones I know about were where TSMC was fooled into doing something bad, and as soon as they found out they took swift action. Some of this kind of thing will always happen with production services. For example, record pressing plants (vinyl, CD, DVD, BD, Laserdisc, etc) have always had problems of unwittingly producing batches of pirate discs. Its not whether it happens that is really important. Its how you handle it when it is discovered that really matters.

[mawyatt]

Intel and IBM were the silicon semiconductor technology leaders a decade and half ago. Poor management and lack of significant government supplements without "strings" allowed others to take the lead. TSMC has done superb job of pushing the technology cart, taking over from IBM and Intel, so hat's off to them

The wavelength of green light is ~550 nanometers, so at 5 nanometer feature size this is 110 times smaller than the wavelength of green light

The technology behind the ability to do photolithography at this feature size is amazing, massively defying conventional optical physics (diffraction limited). Other areas are also looking into "defying physics" (not as much as with photolithography tho), research into things like breaking the Nyquist limit with faster than Nyquist communications for example.

Before I retired a few years ago TSMC had a university program where qualified universities got access to a simple 65nm CMOS process. USC and Cornell made good use of this producing the "Physics Defying" Non-Uniform Sampling ADC (NUS ADC) and PolyPhase Mixer (also called N-Path or Mixer First) respectively. The NUS ADC sort of defeats Nyquist limits by allowing the input waveform to define Nyquist rather than the ADC clock rate, no Pre Filter is involved but a Post ADC filter is employed. The PolyPhase Mixer defies conventional Mixer Physics by posting Noise Figures (<2dB) below the theoretical conventional mixer limit of 3.92dB without a preamp. See chip images below in TSMC 65nm CMOS Technology (NUS ADC, PolyPhase Mixer) from nearly a decade ago.

Anyway, having the later part of my career directly working with these advanced SOTA semiconductor technologies has given a different perspective on what we were all taught in grad school as "physics limits & barriers" 

Best,