Electronics Weekly Electronics Design & Components Tech News
TSMC’s 28nm yield is now over 80%, according to Taiwan’s Chinese language newspaper the Commercial Times.
The Commercial Times quoted sources from TSMC’s equipment suppliers for the information. According to another Taiwan newspaper, the Taiwan Economic News, TSMC’s 28nm capacity is now running at 100,000 wafers a month, up from the 25,000 wafers a month capacity in Q2.
TSMC’s fab 15, in the Central Taiwan Science Park, is said to be ending Q3 with 69,000 28nm wafer per month capacity and will expand that to 135,000 wpm in Q4.
TSMC spent $3.6m, mostly on 28nm capacity, in the first half, and has budgeted a capex of $8-8.5bn for the year. So there’s $5bn still be to be spent.
The news makes it more likely that TSMC is going to fulfill its aspiration of getting 28nm supply and demand into balance during Q4.
Although Nvidia is the only one of TSMC’s customers to specifically complain about poor yields, other customers like Qualcomm and Altera, have complained that constrained supplies of 28nm parts have affected their sales figures.
That has led Qualcomm to look for other sources of foundry such as UMC and GloFo. However Taiwan industry sources say that customers which had been looking for alternative foundry suppliers have returned to TSMC following its yield improvements and capacity extensions.
Better still, you just use your PR department and all of your yield issues will simply go away … “9nm planar? No problem we’re in production right now, ramping faster than ever before and with yields to die for”
Better still, you just use your PR department and all of your yield issues will simply go away …
@Pradeep
When you fab memory (lots of that on SoCs), FPGAs and GPUs you can get near 100% through adding redundancy.
Or you can use a process conservatively and only push the design rules where it’s really necessary and get a better yield for a slightly larger die.
Alternatively you can ignore the RDRs and use the same design methodologies you’ve been using since 250nm and complain in private or public when you get near zero yield 🙂
is it really possible to get the yields over 80% at 28nm with present process trends?
Oh dear, MGP-1, so it’s all bollox is it? What a naughty old world we live in
Don’t you just love it … the same Taiwan rumour mill has GloFop’s 28nm yields at Dresden running over 90%. There’s lies, damned lies and wafer fab yields??? 🙂
I don’t think TSMC has ever publicly acknowledged that it has ever had yield issues at 28nm Byungchun Yang, so whether the problems are at BEOL or FEOL is unknown to me.
Hello David and Anon,
Do you know which (BEOL or FEOL) process is causing the yield problem? Do you think both processes are causing it? I am just curious.
More wafer starts is not the same thing as higher yield. More functioning die out for the same number of wafer starts is an indication of increased yields. TSMC may just be starting more wafers to meet demand on a less capable process