Saturday, October 24, 2020

Silicon nanoelectronics is a truly extraordinary achievement.

Arguably the greatest technical and manufacturing achievement in all of history is around us all the time, supporting directly or indirectly a huge fraction of modern life, and the overwhelming majority people don't give it a second's thought.  

I'm talking about silicon nanoelectronics (since about 2003, "microelectronics" is no longer an accurate description).  As I was updating notes for a class I'm teaching, the numbers really hit me.  A high end microprocessor these days (say the AMD "Epyc" Rome) contains 40 billion transistors in a chip about 3 cm on a side.  These essentially all work properly, for many years at a time.  (Chips rarely die - power supplies and bad electrolytic capacitors are much more common causes of failure of motherboards.)  No other manufacturing process of components for any product comes close to the throughput and reliability of transistors.  

The transistors on those chips are the culmination of many person-years of research.  They're FinFETs, made using what is labeled the 7 nm process.  Remember, transistors are switches, with the current flow between the source and drain electrodes passing through a channel the conductance of which is modulated by the voltage applied to a gate electrode.  The active channel length of those transistors, the distance between the source and drain, is around 16 nm, or about 50 atoms (!).  The positioning accuracy required for the lithography steps (when ultraviolet light and photochemistry are used to pattern the features) is down to about 3 nm.  These distances are controlled accurately across a single-crystal piece of silicon the size of a dinner plate.  That silicon is pure at the level of one atom out of every 10 trillion (!!).  

This is not an accident.  It's not good fortune.  Science (figuring out the rules of the universe) and engineering (applying those rules to accomplish a task or address a need) have given us this (see here and here).  It's the result of an incredible combination of hard-earned scientific understanding, materials and chemistry acumen, engineering optimization, and the boot-strapping nature of modern technology (that is, we can do this kind of manufacturing because we have advanced computational tools for design, control, and analysis, and we have those tools because of our ability to do this kind of manufacturing.)   

This technology would look like literal magic to someone from any other era of history - that's something worth appreciating.

7 comments:

Anonymous said...

Another good book on the history of the semiconductor industry and its lasting effects on industry and society is Cyrus Mody's "The Long Arm of Moore's Law".

Anonymous said...

The way wall street treats the semiconductor manufacturing companies is also incredible. A software company gets valued at multiple times the value of chip manufactureres.

Anonymous said...

@Anon commenting on valuation of software companies for their trivial pursuits: I propose a solution to that. Just like many are wearing masks in the pandemic, if they were also using ad-blockers whenever they browse the web - the valuation problem would be largely solved.

Anonymous said...

Agreed. It's a shame that the US has lost leadership (permanently?) of silicon nanoelectronics given that Intel is behind and its CEO is considering outsourcing production to Asian foundries due to internal manufacturing issues.

https://www.oregonlive.com/silicon-forest/2020/10/intel-shares-fall-sharply-on-weak-outlook.html

Raj Giri said...

I spent several years in Intel before returning the academia. It was funny how mundane it seemed to everyone inside to talk about 10 nm transistors (that was the node I last worked on), but also just the scale of the tools required is incredible and amazing. I mean things like an entire suite filled with many TEMs running literally 24/7, teams of engineers working night shifts on million-dollar optical microscopes, things like that.

I won't really talk about last anon's points-- aside from my info being 5 years out of date, Intel did (and does) struggle a lot at 10 nm.

Jon Barnard said...

@Raj Giri - two days ago, I was trying to explain to a colleague why TEMs within Materials Science & Engineering (MSE) Departments in academia are generally not automated. Your comment reminded me that, for the nano-electronics industry, this is actually the case. I suppose companies like Intel tend to buy a lot of microscopes that must work 24/7 with industry standard metrology algorithms and processes. The diffusion of automated microscopes down to universities has yet to materialize (unless, of course, you count cryo-EM fro structural biology). Thanks for the reminder!

Anonymous said...

@Jon, the automated TEMs that Intel and others use are highly optimized for a particular kind of material/structural cuts. In academia, you are preparing many different kinds of samples and I am not sure if you can automate more while still retaining flexibility.