Electronics Weekly Electronics Design & Components Tech News
But this week’s news that it is putting out leading edge chips to foundry on TSMC’s 14nm process shows how precarious is Intel’s control of its process technology.
The latest best guess for Intel’s 10nm process is Q4 2019 – three years later than originally intended.
With 10nm in its sights, obviously Intel doesn’t want to build any more 14nm capacity, but Intel’s 14nm capacity is full and hence the move to TSMC.
But what happens if, next year, Intel doesn’t get its 10nm process to work?
Next year, TSMC expects that 20% of its output will be on 7nm wafers and TSMC 7nm is generally seen as equivalent to Intel’s 10nm process.
If Intel gets its 10nm process yielding next year then it will only be a year or so behind TSMC but, coming along on TSMC’s roadmap is its 5nm process due for launch in Q4 2019/Q1 2020.
If TSMC 5nm is substantially better than both its own 7nm process and Intel’s 10nm process and/or Intel stumbles further over 10nm, then the process game’s up for Intel.
To stay competitive it will have to put all its leading edge parts out to foundry at TSMC or Samsung.
And it will have no financial justification for building new fabs
Thank you martijn, that resolves the conundrum. Incidentally, it also explains why they don’t make a profit in memories – unlike everyone else.
Intel is known to heavily underutilize their fabs, so you are both correct. They have 50% overcapacity yet still cannot get more wafers out. The effect of being rich (so never had to improve utilization from a cost perspective) and being your own customer (so never had to improve cycle times).
Equipment makers love them though 😀 And now, foundries do too!
Massive over-capacity somehow doesn’t square with the shortage of Whiskey Lake CPUsvreported today or the out-sourcing to TSMC announced earlier in the week, Mike.
Anything is possible, Stooriefit, Intel seems to be a mess. If it can be three years late on delivering a process, it can also be screwing up elsewhere.
That depends on when Intel stopped building 14nm capacity Mike.
It had such massive over capacity not that long ago that there were concerns Leixlip would delay coming on stream. The semiconductor market just seems to carry on growing and growing.
According to the Digitimes report, QuickSilver, the Intel parts being put out to 14nm foundry at TSMC include server processors.
I’m really not convinced on that and suspect it was just a guess. To send processors to TSMC would require a complete redesign as the TSMC finfet is totally different, and also their metalization stack isn’t the same. But I suspect everything else is going out, which if you think about it means Intel must still be doing very well to have all its fabs full.
Or the yield in its own fabs is shocking!
It is not unusual for Intel to use foundries. What would be unusual would be for Intel to use a foundry to fab a processor.
In this case we do not know what this device is but my money is on a peripheral device like WiFi/Bluetooth and not a processor.
Q.