By all appearances, 450mm, an industry disruption, is now on track. In recent announcements, Intel effectively decoupled EUVL from 10nm, which allows the 2018, 450mm target date to stand intact regardless of any EUV slip. If EUV is delayed to 7nm, there’s still a 450mm solution in place for lithography with quadruple patterning. Intel’s announcement is one more piece to prop up the argument that 450mm is very real, with no other mitigating roadblocks.
Prior articles on 450mm.com discussed key elements needed for 450mm: true industry collaboration (G450C), chipmaker leadership (i.e. real investments), and government support. Add in a recent TSMC schedule slip, which seems to align all the major players’ 450mm target date (2018), and a backup lithography plan, and 450mm seems to be coming together. Industry supply-chain economics aside, with 450mm imminent, the most useful discussion is how to make the R&D and implementation most cost effective for the industry as a whole.
Shrinking as We Grow: Throwing Boulders into a Pond
Thinking a few years ahead, what are some of the impacts we might see in 2018, with high volume manufacturing on 450mm wafers? “Impact” is the appropriate word in this case. Think of a small pond, where one throws items into the water and sees the resultant ripple effect on the ecosystem.
For 200mm fabs (and below), the addition of a single fab, with so many other chipmakers and fabs, was akin to tossing a small pebble into the pond. The ripple effect was certainly visible: equipment, jobs, production, often on a local-level; but nothing too dramatic. 300mm fabs, all running leading-edge technology, with higher-throughput tools (as compared to 200mm) and automated handling equipment, have enough transistor capacity and efficiency to boost this pond analogy to one of tossing a rock into the water. Some waves get made—noticeably so. Especially now, with fewer leading-edge semiconductor manufacturers, each 300mm fab announcement or upgrade is significant news and has a pronounced impact on the local, regional, and global marketplace.
Fast forward to 2018. Only a few chipmakers have the funds to move forward with a 450mm fab. Each 450mm wafer will contain up to 2.5 times the number of die as the equivalent technology, 300mm wafer, would have. The advanced fab will cost at least $10 Billion US with a significant portion of this going toward capital equipment. For the pond analogy, it’s as if each 450mm fab coming online is equivalent to throwing a boulder into this same pond. The economic waves created are big and powerful. The impact is felt on local, state, national, and global levels. One of the issues is that there is effectively a minimum capacity a chipmaker is able to purchase to remain efficient. A likely fab size of 45,000, 450mm wafer starts per month, is a huge “quanta” of capacity. Because chipmakers cannot collaborate on division of markets or product timing, the size of each fab and their chipmaker’s competitive behavior may lead to some serious instability for the industry supply chain when 450mm is first introduced.
With this mental image, let’s think about the semiconductor ecosystem we might expect in 2018.
Logic: Foundries are becoming King
With the continued pace of consolidation throughout the industry, the foundry model has nearly become the defacto standard for leading-edge logic (vs. memory) manufacturing, with a few exceptions. Intel is one of the primary exceptions to this rule, most certainly within the 450mm camp (though they do have a small foundry operation). The foundries are competing fiercely for market share, and GlobalFoundries’ recent announcement on 14nm and Samsung’s rapidly growing foundry business (much dedicated to Apple) set up a remarkable three-way partner (G450C) and competitor battle with TSMC. All three have deep financial pockets.
Further market complexity comes from the fact that ARM and Intel are now directly competing for business, on most levels. For example, while ARM licensees are trying to enter the server market, historically Intel’s captive turf, Intel is aggressively going after the Ultrabook, smart phone, and tablet market, migrating the Atom and next-generation core processors into what has been historically ARM’s territory. In effect, with foundries being the primary fab for many ARM (and now AMD) products, Samsung, TSMC, and GlobalFoundries are, on some level, all pitted against Intel.
Thinking through possible perturbations of this four-way competitive landscape and what could transpire over the next five to six years is dizzying. As an example, AMD is no longer hamstrung by ownership of a fab and the economic barrier to 450mm wafers. They have full access to the same technology node on 450mm as their foundry partners. This access comes without the inherent R&D and capital investment risk that Intel must bear. No wonder Intel is hedging all bets with key investments to mitigate any technology or wafer-size risk. One wonders if TSMC, who has been historically a fast-follower in technology, will truly lead by going “all-in” as the first 450mm inductee. Do they, then, provide cost leadership for AMD, ARM-licensees, and other customers; or do early-adopter pains, as well as nuances in 450mm implementation (such as adverse impacts on cycle time, which primarily impacts high-mix foundries) put them at a cost disadvantage to Intel?
A Disruptive Convergence
Funding required for each major player in G450C, when 450mm comes online, with potentially exclusive access to leading-edge technology, the lower cost and large supply may force foundries to sign up for a “last-man-standing” battle. Keeping or increasing market share while sacrificing gross margins (not unlike memory providers did in the past) becomes the name of the game, as companies know lost market share often never returns.
As an illustration, in an analysis performed with the SEMI/MIT cash flow model, assuming there is a gradual ramp-up to full 450mm adoption, less than two 450mm fabs are required to meet market demand in 2018. Of course, there are a number of other assumptions, and inherent issues which go along with this. The first issue raised is that it’s difficult for the industry to purchase “fractional fabs,” for example, 1.7 fabs (the projected requirement in 2018). Beyond this, there’s another significant problem…we have four heavy-hitters all participating in the 450mm consortium. If 450mm truly provides cost advantage, while at the same time blocks access to leading-edge technology (no longer available on 300mm), there’s not likely any agreement that a sound company would sign that allows G450C partners to take turns adding industry capacity in the form of a 450mm fab. As mentioned previously, any such agreement would be anti-competitive and cross anti-trust boundaries. With these conditions, it is likely that four fabs (one for each consortium partner) will be built in a short period of time. If the G450C, Phase Two agreement holds true, these would all be built in New York. In any case, this is more than double the 450mm fabs required to meet projected market demand in 2018. Taking on the previous analogy, this is equivalent to throwing four large boulders into a small pond, nearly simultaneously. The effects will wash up and down the supply chain and may take quite some time to settle.
In future articles, we’ll present some results from the economic model and discuss possibilities under different 450mm build-out scenarios.