When you buy through links on our articles, Future and its syndication partners may earn a commission.
Last week, we reported on a claimed 18-month deadline for Intel to win customers for its next-next-gen 14A chip node, else the company would have to give up on cutting-edge chip manufacturing. This week, it's Intel's actual next-gen node that's in the news for the wrong reasons, as 18A is apparently in trouble.
Reuters says sources inside Intel put the company's all-important 18A node (well, it used to be all-important before Intel shifted that dubious honour to 14A) at 10% in terms of manufacturing yields. That lowly figure is reportedly up from an even worse yield of just 5% late last year.
Reuters says that Intel has generally aimed for yields of 50% or more before ramping up volume production in order to ensure that chip manufacturing is profitable. By that metric, it would appear 18A is nowhere near ready by Intel's usual standards.
It's worth noting that those yield figures apply to Intel's upcoming Panther Lake mobile CPU, which will be the first consumer CPU from Intel based on 18A. Intel is only using this process node for a relatively small part of Panther Lake, in that only the CPU tile will be on 18A. The rest of the chip, including I/O and the iGPU, will be produced on another node.
So, the chiplet Intel is attempting to make on 18A for Panther Lake is very small. But according to this Reuters report, it still can't get yields up. That implies a very problematic node.
The bigger a chip, the more likely it is to have a manufacturing flaw that breaks functionality. That's why huge GPUs are so expensive. They are much more likely to have flaws which prevent them from being sold as fully functional.
The idea with the Panther Lake CPU tile is to make it much smaller and thus reduce the chances of any given tile being broken. But even with that approach, Intel is reportedly struggling to achieve anything even close to good yields.
Intel hasn't commented directly on the claimed yield numbers. However, Intel's Chief Financial Officer David Zinsner previously admitted to Reuters that even when Panther Lake yields were good enough for volume production, they wouldn't be "accretive" and Intel would "still have to make an improvement."
That seems to be a slightly opaque way of conceding that Intel doesn't necessarily expect Panther Lake to be actually profitable at first release as a consequence of low yields. As for how accurate the Reuters report is, that's hard to say. Independent data on the health of Intel's 18A node essentially doesn't exist.