There’s an interesting piece on SemiWiki, written by Scotten Jones, founder and president of IC Knowledge, saying that high volume 450mm manufacturing won’t happen until 2023.
Jones said he picked up this snippet ‘from the conference floor’ at the SPIE Advanced a Lithography Conference in the last week of February in San Jose.
What Jones heard was that Intel had pulled all resources off 450mm and the view was that high-volume 450mm manufacturing won’t happen until 2023.
Pushing Intel’s decision was, said Jones, “low fab utilization and the empty fab 42 shell”.
Last month, Mark Durcan, CEO of Micron, the world’s fourth largest chip company, said that 450mm may never happen.
“I am not convinced that 450mm will ever happen but, to the extent that it does, it’s a long way out in the future,” said Durcan, “there is not a lot of necessity for Micron, at least over the next five years, to be spending a lot of money on 450mm. There is a lot of investment that needs to go on in the equipment community to make that happen. And the value at the end of the day – so that customers would buy that equipment – I think is dubious.”
Four years ago, research by SEMI suggested that 450mm may only reduce die cost by 8% – not worth doing at an R&D cost of $25-30 billion and the $10 billion-per-fab outlay required by device manufacturers.
However, in January, Malcolm Penn CEO of Future Horizons which is heavily involved in the EU’s 450mm plans said that 450mm reduces die cost by 30%.
Fujitsu’s Ian Dedic questions that: “Saying that 450mm reduces die cost by 30% is being economical with the truth – the missing phrase is “all other things being equal”,which they aren’t… Once the very high cost of developing and installing 450mm is paid off the cost may indeed be 30% lower – but who is going to pay back the multi-billion dollar cost of development and over what timescale? The semiconductor equipment supply industry took many years to recoup the development cost of switching to 300mm wafers, and there’s still a lot of 200mm kit out there.”
“If 450mm only gets installed in a few new state-of-the-art fabs (which will be at 7nm or below by then) the market share will be lower than 300mm (which will still exist, see below) so the economies of scale will be less,” adds Dedic, “on top of this, price-sensitive applications (of which there are very many!) will probably stay in 28nm processes using 300mm wafers because the cost of these will be written off by the time 450mm emerges and the die will probably be cheaper than 7nm processes on 450mm.”
“So the only people who want 450mm are those designing/making a very small number of high-volume high-priced designs like CPU or ultra-high-volume density-driven designs like memory, where the enormous NRE cost of doing a 7nm chip on 450mm – probably something of the order of $200m-$500m depending on complexity – can be paid back by even more enormous revenue,” concludes Dedic.
It was reported before Christmas that ASML’s effort to develop 450mm has been ‘reduced’. ASML has confirmed this.
The equipment companies en masse have been reluctant to do 450mm. “If the fab owners think that the equipment manufacturers are going to make and ship 450mm equipment cheaply so they can make money on the wafers, I think they’re living in cloud cuckoo land,” says Dedic, “the semi equipment industry was badly bitten by 300mm, they invested a lot into development on the promise of big orders which didn’t happen, and it took years for them to even break even on these costs. I doubt that they’ll fall for the same promises again…”
As Dennis Ames puts it: “People have to wonder if the semiconductor equipment manufacturers want to see another 50% reduction in the amount of equipment sold and the doubling of NRE included in every purchase that was seen in the conversion from 200mm to 300mm.”
The stalling of 450mm development leaves the semiconductor industry in stasis because there seems no way open to it to reduce die cost. The only other route, apart from 450mm, to die cost reduction is shrinking. And the only known way to shrink cost-effectively after 28nm is EUV.
But EUV is not going to happen soon, and may not happen ever.
As Ian Dedic says: “As far as I can tell lack of EUV (or some other magic like DSA) is not a technical roadblock to 7nm, but it may be an economic one – triple or even quadruple patterning will put the cost through the roof (not just more masks but also yield drop), and make design extremely difficult due to very restrictive design rules – the gates can be a regular array which can be patterned, but the metal layers can’t be.”
So shrinking won’t reduce die cost and the other route to reducing die cost – increasing wafer size – is now problematical.
It wasn’t always like this. In mid-2012, Intel paid $3 billion for a 15% stake in ASML and said it would put another $1 billion into the 450mm and EUV R&D projects, Samsung bought a 3% stake in ASML for €503 million with a commitment to put up €276m for ASML’s R&D programmes, and TSMC paid €838 million for a 5% stake in ASML and another €276 million for the R&D projects. Intel is also funding a 450mm development effort by Nikon.
Now, it seems, the 450mm effort has dried up and the EUV effort has stalled on the light source issue.