4 ms·

But in consider of time to mass production and density[1]: TSMC 5nm: 171 MTr/mm2 in 2021 (for apple A15) Intel 10nm: 100.76 MTr/mm2 in 2021 (for 12th core-i)

by ytch 5y ago

But in consider of time to mass production and density[1]: TSMC 5nm: 171 MTr/mm2 in 2021 (for apple A15) Intel 10nm: 100.76 MTr/mm2 in 2021 (for 12th core-i) Intel keep saying they will win in the future, while they seems forget TSMC will improve too. Comparing Intel 10nm with TSMC 7nm is outdated. Or is there any objective metrics to determine which one is better, other than MTr/mm2? [1] https://www.anandtech.com/show/16823/intel-accelerated-offensive-process-roadmap-updates-to-10nm-7nm-4nm-3nm-20a-18a-packaging-foundry-emib-foveros https://www.anandtech.com/show/16823/intel-accelerated-offen...

throwawaylinux 5y ago
Sure but at some other point in time Intel had a denser process and TSMC caught up despite Intel also improving, so that fact alone doesn't give you a lot of certainty about the future. > Or is there any objective metrics to determine which one is better, other than MTr/mm2? There are many, many objective metrics. Many are interdependent, many are tunable or selectable within certain ranges, many are in opposition, most are highly secretive, and all of them influence he type of device being made, the circuit design, logic, in different ways. Even the experts with all the data for a particular technology and with a particular device in mind can't necessarily tell at a glance whether a change would be worthwhile. And even the meaning of worthwhile varies for the end-user of the device depending where their workload is. Make the middle-performing transistor slightly faster and leakier? Great you can close timing at a few ps faster but the device uses more power and is hotter, good for some customers bad for others. Even simple density is not that simple in the fine details. Density can be in opposition to leakage, switching speeds, wire RC and propagation delay, etc. There isn't an easy way to compare and there never will be.

by ytch 5y ago

throwawaylinux 5y ago

Sure but at some other point in time Intel had a denser process and TSMC caught up despite Intel also improving, so that fact alone doesn't give you a lot of certainty about the future. > Or is there any objective metrics to determine which one is better, other than MTr/mm2? There are many, many objective metrics. Many are interdependent, many are tunable or selectable within certain ranges, many are in opposition, most are highly secretive, and all of them influence he type of device being made, the circuit design, logic, in different ways. Even the experts with all the data for a particular technology and with a particular device in mind can't necessarily tell at a glance whether a change would be worthwhile. And even the meaning of worthwhile varies for the end-user of the device depending where their workload is. Make the middle-performing transistor slightly faster and leakier? Great you can close timing at a few ps faster but the device uses more power and is hotter, good for some customers bad for others. Even simple density is not that simple in the fine details. Density can be in opposition to leakage, switching speeds, wire RC and propagation delay, etc. There isn't an easy way to compare and there never will be.