The Dutch maker of the lithography equipment used to etch the world's chips has been going cap in hand to its largest customers.
ASML Holding wants its clients to fork out some co-development funds for the creation of next-generation 450mm silicon wafer-baking technology and the lithographic gear to etch ever-smaller transistors upon slices of silicon.
Intel kicked in $4.1bn nearly a month ago and this morning Taiwan Semiconductor Manufacturing Corp, the foundry of choice for many processor and graphics chips among those who no longer have their own fabs, handed over €1.1bn, or about $1.38bn.
That's about a third of what Intel is paying to ASML for its 15 per cent stake in the equipment maker, and the question is whether or not the 5 per cent stake in ASML that TSMC is getting is proportional to the expected equipment sales and chip volumes both companies will make. The other obvious question is whether Samsung Electronics, which was also asked to put money into the hat back in July, will come through and buy a 5 per cent stake. ASML said it wanted to sell a 25 per cent stake in itself to raise funds to do future lithography and wafer baking research.
In a statement, TSMC said that it would acquire a 5 per cent stake in ASML for €838m (just a little over $1bn) and that it has also committed €276m ($342m) to various research and development programs relating to extreme ultraviolet lithography and 450mm silicon wafer production, slicing, and etching over the next five years.
"One of the biggest challenges facing IC scaling today is how to effectively control the escalating wafer manufacturing cost," explained Shang-yi Chiang, one of TSMC's chief operating officers. "We are confident that the additional funding for ASML's research and development programs will help secure and accelerate EUV development activities, in parallel with the necessary focus on improved performance of existing optical lithography tools and speed up the deployment of new technologies for 450-millimetre wafers. This effort will help the industry control wafer cost, and therefore protect the economic viability of Moore's Law."
ASML and TSCM have partnered before and developed the deep ultraviolet dry and immersion lithography techniques currently used to etch 28nm chips at TSMC and 22nm chips at Intel; this machinery uses an ultraviolet beam with a wavelength of 193nm to do the etching.
With extreme ultraviolet (EUV) lithography, the wavelength will be shrunk down to a much tighter 13.5nm while the wafer size will be boosted by 50 per cent from the current 300mm. ASML shipped the first prototype EUV machines to customers last year, but the technology has a long way to go before it is ready for production. If the bugs can be worked out of EUV technology, it should be able to get chip feature resolution down to 10nm or so, but after that, ASML will have to cut the wavelength in half.
Japanese camera and lithography equipment maker Nikon is also working on EUV gear. For now, however, vendors are expected to get by with the 193nm deep UV gear by double and sometimes triple patterning the chip masks, in effect distorting the chip image on purpose in multiple layers that results in a crisper image on the chip than you would expect.
Getting larger wafer sizes is at least as important as getting smaller transistor sizes, in as much as the jump from 200mm to 300mm wafers cuts per-die costs by somewhere between 30 and 40 per cent and the move to 450mm will do about the same.
In some ways, this might be the easier problem to crack. Last September, IBM, GlobalFoundries, Intel, TSMC, and Samsung committed to spending $4.4bn, and the New York state pledged $400m in tax breaks and investments, to work on 22nm and 14nm etching and on stretching wafers to 450mm. Most of that money is IBM and GlobalFoundries investing in their fabs, with only $400m of it going to set up the Global 450 consortium, run by Intel, to work on prototype 450mm wafer-baking gear.
By the way, a 450mm fab will cost about $10bn a pop, so it is still not clear who will have the dough – other than Intel – to build such a beast. People worry about Moore's Law hitting a technical wall, but it looks far more likely that it will hit an economic one first. Even if you can, technically, make smaller transistor gates on a larger die, you might not be able to afford to do it unless you have a monopoly on a chunk of the chip biz.
Your move, Samsung. ®