At one point, it seemed everything was lining up very well for 450mm: a consortium of the proponents (G450C), a post-450mm delivery of EUV (along with significant investment in ASML), and some agreed-upon industry deadlines, and many 450mm-specific industry groups springing up in support.
That seems to be changing…
What’s changed? First, there’s some evidence that G450C doesn’t seem to really represent the Big Four view on 450mm, particularly Samsung; next, ASML is pausing development on 450mm; additionally, there are signs that the die-size sweet spot for yield is with smaller die for FinFETs; and, finally, Intel lost out on Apple at 14nm. These all relate to 450mm in some way and all translate into detractors, I believe.
First, there’s the rumors that Samsung, with significant revenue from memory, has stated privately that they are not willing to risk losing any market share to 300mm-based memory suppliers during a 450mm transition. Any additional manufacturing cost, resulting in higher prices or lower margins, puts Samsung at a serious disadvantage against 300mm competitors. Regardless of the long-term cost implications of a transition to 450mm, there will be some initial period of time where 450mm wafer cost will be relatively high and give a wafer cost advantage to 300mm fabs. Unless there is a breakthrough in technology that only exists on the “other side” of 450mm, any memory provider making a move to 450mm is going to be at a disadvantage for some period of time.
Which brings us to the next concern. There are rumors that ASML is putting 450mm development on hold. With 450mm potentially intersecting at 7nm already (in 2018), delaying 450mm EUV lithography adds confusion. If both 300mm and 450mm have quad-patterning litho available at 7nm, 450mm loses an advantage it would have had until EUV makes its debut. Likewise, if EUV is available for both 300mm and 450mm wafers, 450mm’s EUV litho advantage may be completely gone. Hence, the danger of switching to 450mm wafers (Samsung) while there’s not a “450mm-only” technology advantage available to them.
Die Size Trends
Brian Black from AMD at the 3D conference last week in San Francisco stated that their best yields are happening with smaller die for FinFET. This fits well with the 3D paradigm and high-bandwidth logic/memory stacks. In this case, you start chipping away some of the advantage that 450mm has over 300mm—that is, in packing efficiency for large die size.
Samsung’s Apple Win
Word has it that Samsung has won a portion of Apple’s mobile processor business for 14nm FinFETs. Samsung and TSMC will split the manufacturing. One can bet that if Samsung delivers well, they will have a significant piece of the business.
Intel is going after the foundry market. That’s no secret. Their future growth likely depends on this level of diversification, even if the foundry margins are not as good as their traditional, processor markets. Likewise, their success in FinFETs has been remarkable. However, the success has been in their bread-and-butter product space, that is, their own processor designs. They haven’t yet won the major (i.e. Apple) customers in the tablet and smart-phone markets they need to in order to ramp-up their foundry business big-time. Right now, Samsung is the trusted source for IC delivery in the smart-phone world. Asking Apple to trust Global or Intel with their livelihood, when Samsung is just waiting for Apple to stumble and beat them with their own Android-directed processors, may be asking too much until Intel’s foundry model is proven out.
With their memory dominance and foundry wins, Samsung seems to be in the driver’s seat and is likely delaying any 450mm implementation. Remember, that in any relationship of near-equals, whomever needs the relationship the least is the one in charge. I’d say that is true for 450mm and Samsung—they could make or break the effort, and they don’t appear to be “all-in.” Compound that with ASML’s pause on 450mm development and the eroding die-size advantage for 450mm. Finally, in talking with suppliers, there are some who have lowered their expectations that 450mm may ever happen. Perhaps this is wishful thinking (for some), or sour grapes for others who have invested into a 2017/18 date. Either way, the tide seems to be turning, until some other announcement or development says otherwise.
This all adds up to a caution sign for 450mm. In no case is there a blinking red-light saying “stop.” However, there are warning signs that perhaps the supply-chain needs to proceed carefully. Carefully means to continue to evaluate the conditions and the associated technology roadmaps to see how things are lining up, and make corrective decisions as needed. Personally, I see a 450mm transition as the last-available industry disruption where a number of fab productivity measures that have been sitting on the sidelines could get inserted into manufacturing. So, I have been hoping there would be better news. Let’s see what the new year brings.