Intel has broken ground on its first dedicated facility for 450mm wafer production. The new foundry — D1X Module 2 — is scheduled to come online in 2015, with Intel committing to roughly $ 2 billion in bring-up costs through the end of the year. Foundry costs have continued to balloon in recent years; the 450mm facility will likely set records in this area due to the cost of the new manufacturing equipment.
New wafer sizes are essential to continuing to ramp the economies of scale that have driven semiconductor manufacturing to date, but the timing of these deployments has been controversial. While Intel committed to a 450mm ramp alongside TSMC, other major foundries have balked at the notion. This is partly a result of the nightmarish 200mm to 300mm transition, which kicked off during the height of the dot-com bust and required major shifts in how wafers were processed at the factory. Over the past 12 months, IBM, Samsung, and GlobalFoundries have begun to talk about a 450mm transition, but Intel will likely be the first company to deploy the hardware commercially. But why switch in the first place?
A calculated bet
The advantage of 450mm wafers is that while the equipment costs are formidable, the long-term production costs are advantageous. GlobalFoundries has done an extensive presentation on this — for a chip of a given size, a 450mm wafer can yield 3,400 dies while a 300mm wafer yields just 1,450.
This means a 450mm fab with a 40 to 45K wafer starts a month is equivalent to a 300mm fab with 100K wafer starts a month. Given the enormous cost of building new foundries, 450mm wafers are expected to save 20% to 25% in capital expenditures. 450mm also helps defray the increased cost of producing smaller process geometries on 300mm wafers, as shown below. While the x-axis isn’t labeled in this graph, the first green dot corresponds to the 10nm node.
But for Intel, the push is particularly acute. Intel has a huge number of fabs and it’s on the cutting edge of semiconductor technology, with strong pushes for both lower process nodes and larger wafers. Key to both of these endeavors, however, is the expectation that it can sell all the products that it manufacturers. In discussions regarding ARM products, commenters have often questioned whether Intel can maintain its historic margins while competing for smartphone and tablet products given that the ARM chips on which such devices rely often sell for $ 10 to $ 30.
The simultaneous push towards 14nm and 450mm wafers are part of how Intel intends to simultaneously preserve its margins while hitting competitive pricing. If a 450mm wafer packs 2.5x the processors of a 300mm wafer, the cost-per-processor is far lower if and only if Intel can ship every single chip. If it can’t, then the cost structure inverts. This is the essential nature of the “Atom Everywhere” plan. Intel needs — has always needed — widespread uptake of its Atom processors to justify the enormous build-outs the company has done at Fab 42 and now at Fab D1X.
Long-term, 450mm wafer production is going to ramp across the entire industry; GlobalFoundries believes the capability will approach 50% of the semiconductor market by 2028. That’s only slightly slower than the 300mm ramp, which Intel pioneered in 2000 with the P4, and hit the 50% mark around 2009. Even today, a substantial amount of manufacturing is done on sub-300mm wafers, and that will continue to be the case for more than a decade once the 450mm ramp starts — it costs too much to retrofit a fab for TSMC, UMC, or Samsung to do so all at once.
For Intel, the cost savings of building out 450mm could be as significant as the cost savings of being more than a process generation ahead of the overall market.