Bottom Line Up Front
I rate this 1.2.
Unreadable formatting. Too much jargon for an amateur. The paper is mostly describing the vomit of figures found on the last pages. It also introduces roughly a million acronyms (RaMA) that are never used again.
More complex chips take more to build. Only considers manufacturing cost, not lifetime cost.
Summary
Intro
Semiconductor manufacturing ignores life cycle assessments in favor of more performant technologies. This paper will review the transition from iN28 to iN3 with different scenarios for the most recent nodes. Then discuss the growing importance of exotic minerals.
Tech scaling for area and performance
Logic cell area is decreasing. To print these, lithography techniques are using smaller and smaller wavelengths of light, which use more energy. More exotic metals are needed as well. Funky shapes let us cram even more in there.
Manufacturing Metrics
Electrical energy
EUV is thirsty for that zip zap. Old lights used 10kW, we’re at 100kW now, new ones use 1000kW. kWh/wafer has doubled. iN6 and below have some particularly electrically hungry new steps like pitch scaling.
Because we’re burning so much energy, we’ve got to burn more energy to cool it off. 40% of energy goes to running the AC. Doesn’t help the foundries are on a tropical and coal burning island. This will probably get better though.
Ultra Pure Water use
Ultra Pure Water is a use it once material, and energy intensive to create. Some use waste for cooling. Consumption will double going from iN28 to iN3.
Greenhouse gasses
Fluorinated compounds used for dry etching, cleaning CVD chambers, and epitaxial growth are orders of magnitude worse than CO2. OUV helps, and abatement systems catch most now. This will stay flat.
Raw materials
Adopting Ruthenium as the local and back-end-of-line interconnect will increase the CO2 footprint. Germanium, Ruthenium, Cobalt, and Tungsten supplies are all at risk, conflict minerals. Some fabs are starting to recycle scrap.
Power-performance-area-cost-environmental score
iN28 to iN3 will more than triple electricity use, more than double ultra pure water, and more than double greenhouse gasses per wafer. Transistors are still shrinking though, so per transistor, the environmental cost will be a wash.
Thoughts
PPACE isn’t a real acronym as far as I can tell. Or is it PPAC? They only use it twice and can’t agree with themselves. This paper is so short and so dense with acronyms, if they would have just inlined all figures and expanded every acronym it’d be far more readable basically for free.