Unsung hero here is TSMC and their industry-leading 5nm fabrication node. Apple deserves praise for its SOC architecture on the M1 and putting it together, but the manufacturing advantage is worth noting.
Apple is essentially combining the tick (die/node shrink) and tock (microarchitecture) cadences together each year, at least the past 2-3 years. The question, perhaps a moot one, is how much the performance gains can be attributed to either? The implication is that the % improvement due to tick is available to other TSMC customers, such as AMD, Qualcomm, Nvidia, and maybe even Intel.
We'd have to wait until next year (or 2022) once AMD puts Zen4 on 5nm and see an apple-to-apples comparison on the per thread performance. But of course by then Apple will be on TSMC 3nm or beyond...
Worth mentioning is the insane and disruptive technology making TSMC's 5nm possible. ASML and it's suppliers have built a machine[1] that has a sci-fi feel to it. It took decades to get it all right, from the continous laser source to the projection optics for the extreme ultraviolet light[2]. This allowed photolithography to jump from 193nm light to 13.5nm, very close to x-rays.
The CO2 laser powering the EUV light source is made by Trumpf[3].
Edit:More hands-on video from Engadget about EUV at Intels Oregon facility[4]
Thanks for the great links / resources. Those machines look insanely complicated. I can just imagine how they get shipped to Taiwan and elsewhere (they apparently cost $120M each in 2010 [1]).
A bit offtopic but I've always found it amusing that a form of lithography, of all things, is fundamentally powering our tech revolution for decades. Especially after a girl I knew learned lithography in an art class, watching her do it in a primitive form, which inspired me to read about it's history in art and professional uses (signage, etc).
That combined with vacuum tubes (which also rank high up there in the revolution thing) are the two things I one day wish to learn how they really work. Not just surface level nodding along.
They say “decades in the making” - when did it first become viable, and then how long to master the process and become confident enough to mass produce consumer goods from it? I’d love to see a timeline w milestones.
Apple has built many second-source partners for cost-reduction for a long time. But most of their CPUs are made by TSMC right now.
I'm wondering will Apple find another semiconductor factory partner (they tried to build A9 by both Samsung and TSMC, but Samsung one seems like has heat issues)or stick with TSMC?
Apple is essentially combining the tick (die/node shrink) and tock (microarchitecture) cadences together each year, at least the past 2-3 years. The question, perhaps a moot one, is how much the performance gains can be attributed to either? The implication is that the % improvement due to tick is available to other TSMC customers, such as AMD, Qualcomm, Nvidia, and maybe even Intel.
We'd have to wait until next year (or 2022) once AMD puts Zen4 on 5nm and see an apple-to-apples comparison on the per thread performance. But of course by then Apple will be on TSMC 3nm or beyond...
EDIT: confused myself with tick and tock