Qualcomm to give all its 3nm AP foundry work to TSMC instead of Samsung (opens in new tab)

>> What is also missing in the report is that TSMC 3nm also performs better.

Was recently reading some non-authoritative news that TSMC is also having yield issues on their 3nm. From the sound of it probably not as bad as Samsung. Taken together this doesn't look good for Intel's optimism toward catching up in just a few years.

It can’t come soon enough after rounds of terrible Samsung fab results. Their latest and greatest node is still somewhere around TSMC 7nm or intel 10nm, but with worse performance characteristics and current leakage.

huh? Read somewhere, I think from wikipedia, that Samsung 3nm is not ready, and not expected to be ready before 2024.

35% would be bad yields with a 650mm^2 GPU.

These are likely only around 80-100mm^2 (likely because it’s impossible to find die size numbers for recent snapdragon processors — reviewers need to solve this).

Abysmal would be damning their process with faint praise. I wonder if Samsung is even turning a profit or just trying to offset losses.

>Samsung's fabs that inefficient?

This has been the case for sometime, I pointed it out way back in 2017 on SemiWiki. They are fine on mature node though, this is only a problem on leading edge.

But credit where credit are due, Samsung didn't give up. Leading edge is hard, and that is saying the least. If you have been wondering where those Samsung NAND and DRAM profits gone, Foundry services. And if you have been asking why TSMC just doesn't hike price. Samsung has been keeping them at bay some what. And if you think TSMC is a monopoly and has no competition. You should spend a weekend doing more reading on Semi-Conductor industry.

In 6 NM crazy low in 3 Maybe reasonably, each time the process shrik tend to be more harder to sustain the yield

Note that this is for 2023, so Samsung still can improve this for mass production.

There is a positive feedback loop in fab progress: the more customers you have the more yield experiments you can do, the better the yields get, the more customers you get. This explains why the number of leading edge fabs is decreasing, ending up with a monopoly:

https://en.wikichip.org/wiki/technology_node#Leading_edge_tr...

I really hope Intel can get back on track and find success with IDM 2.0. Not long ago they had a process advantage… up until 10nm yield woes.

Just wait for TSMC to become fully Chinese.

The only thing we learn from history is that we learn nothing from history.

Fabs have been consolidating for thirty years as each node has become exponentially more expensive than the last.