Note that classically cylindrical cells dissipate heat through the cylinder walls, not the ends, because there are voids at both ends. 18650 cools very well because it's only 18mm in diameter; heat flow needs to travel only a few mm on average; surface-to-volume ratio is good; hence this was the choice for Tesla. 25650 was available, but the cooling was inadequate. Tesla's first micro-optimization was to find the sweet spot where the cooling is still good enough, but casing and assembly costs are reduced, and that was just at 21mm diameter, between standard 18650 and 25650 size. All this makes sense and, as an anecdote, I actually happened to predict their exact 21700 cell size before they published it.
But this, I couldn't predict. It's funny how trivial idea this is, yet it never crossed my mind!
The new idea is to change the primary cooling path from the cylinder walls, to the endcaps, by removing the void and replacing it with solid connection to the electrodes. This is actually quite a big fundamental change. As a result, the diameter can be arbitrary; increasing the d
increases the cooling area at the endcaps. I suppose, this means a big design change in their liquid cooling system, as well.
Much of the Teslas original choice to go for cylindrical cells was their off-the-shelf availability, superior cost ($/kWh) and superior energy density (Wh/kg) back when they were designing the Roadster. They had 200Wh/kg cells available from the big manufacturers like Panasonic while the other EV pioneers were struggling with in-house custom-specified pouch or prismatic cells, using new weird chemistries, somewhere around 90 - 150 Wh/kg depending on case. Tesla's initial success is because of all the development seen in laptops at the time; they used predominantly 18650 cells. BTW, this market situation was still similar while designing Model S.
The history shows their choice of cylindrical cell was clearly a correct one, and it seems that even when the market situation has changed so that the original main driving force, using a superior COTS product, has become completely irrelevant as they can design anything they want now, they are still not going to abandon it; clearly cylindrical cells work well.
And, finally,
Can't believe Telsa's marketing is so effective that people are talking about such marginal process improvements.
Claimed 15% energy density gain is far from marginal. It's one level up from micro-optimization; and even micro-optimization makes sense at such scales. And they didn't even talk about the actual numbers in process cost savings.
The optimizations need to be done everywhere. As a result of
dozens of different small improvements, li-ion cells are approaching 300Wh/kg as we speak, being some 100Wh/kg originally when introduced in late 90's. There is no single big breakthrough in the history.
PTCs were mentioned earlier, note that since Model S, Tesla has custom-ordered their Panasonic cells
without the PTC, replacing it with their fusible link wires instead. This is likely a significant cost saving, and also improves performance.