Economies of Scaling
Moore’s Law keeps the pressure on as semiconductor fabs and foundries continue to chase down higher quality chips at lower prices, a seemingly endless technological rat race that most of us consumers take for granted as the natural progression. In the world of semiconductors, scaling towards smaller chips (via smaller transistors and wires) has allowed for such substantial progress in performance. It has also brought technology closer and closer to what might be deemed ‘Moore’s Limit’, as the fundamental concept behind Moore’s Law is exponential in nature. The critical hurdles being unveiled as scaling continues to be the focal point is process variability in manufacturing and transitioning costs. The risks of variability seem to play by similar rules as Moore’s Law, offsetting the increase in performance with a decrease in consistency. Similarly, the costs attributed to upgrading manufacturing facilities include increasingly complex machinery to offset this. The economy of scaling favors the big players more and more as technology develops.
Starting simple, transitioning towards the cutting edge technologies has substantial advantages but at an enormous initial investment. Currently, the newest wafer sizes being pursued will measure in at 450 mm (about 18 inches) and transistors at 22 nm (IBM released their 22 nm Ivy Bridge just this year). According to GlobalFoundries, the estimates for a fab at 22 nm hovers around $6.7 billion, process R&D at $1.3 billion, and design costs at $150 million. According to VSLI, process and tool R&D for the development of 450 mm wafers is an absurd $32 billion. Upgrading these manufacturing plants tend to cost more than abandoning them for entirely new ones, making the transition yet more painful. These economies of scale will likely result in continued consolidation, as the SMEs in the business eyeball exit strategies. Transitioning is only the first challenge, as these smaller transistors also demand a more complex manufacturing environment. Equipment that keeps paces with the growing complexity necessary to produce these products is a driver of these high costs, mostly in the form of offsetting variation.
Variation is a constant. The issue with variability is that it tends to increase in consequence as a result of a decrease in size. Simply put, smaller things vary in larger ways. Delving into the variety of factors that may cause variation is exhaustive, but specific elements of process variation are particularly interesting in regards to scaling. As noted above, smaller transistors are the primary result of increased scaling in semiconductors. Where the problems occur in regards to process variation is that a smaller transistor will have fewer atoms relative to their smaller size. Fewer atoms means that any variance whatsoever as to the number of atoms present (particularly relevant to the dopant atoms added in during manufacturing) will represent a larger percentage of the transistor. As these dopant atoms are crucial for the necessary voltage threshold, the bell curve of the voltage threshold flattens out towards higher percentages at either end, wasting any that fall below zero or require too much. This means less predictability and lower percentage yield from the manufacturing process, and also provides just one example why the manufacturing process changes so drastically for each newly scaled product.
Aside from the increased performance derived from scaling, there are also advantages in regards to operational costs. The International Technology Roadmap for Semiconductors published an extensive analysis of the direction of the semiconductor business, which highlights some interesting notes regarding the transfer to 450 mm wafers. The projection is that moving to 450 mm wafers will result in approximately a 30% improvement in cost/cm2 and a 50% improvement in cycle time reduction. The last change in wafer size, from 200 mm to 300 mm, occurred 10 years ago and did exactly that. This means that, despite the large initial investment discussed earlier, having the capital to invest towards 450 mm will be crucial in remaining competitive in the long-term.
Economies of scaling reflect economies of scale. Companies with the leverage to invest in new technologies will derive better products and, in some cases, cheaper production in the long-term. Due to the increasing variability in production as scaling continues to abide by Moore’s Law, these transitioning costs will likely not decrease any time soon. As a result, consolidation within this industry seems likely to continue as companies struggle financially to keep pace. As always, you need money to make money.