Seeing Beyond Moore's Law

For decades, the semiconductor industry has successfully and single-mindedly pursued a set of traditional performance trajectories. The pursuit of Moore's Law — the assertion that the number of transistors on an area of silicon doubles every 18 months and consequently that the cost per function of ICs falls by half in about the same time — has delivered enormous value to the market and to the companies that have enabled this progress.

Today, as the semiconductor industry looks toward the future, there is much consternation that the pursuit of Moore's Law will become untenable by conventional methods. The International Technology Roadmap for Semiconductors (ITRS) plots a course that is clear for the next 10-15 years, after which the industry treads perilously close to physical device limitations that appear insurmountable.¹

But is shrinking linewidth likely to be as competitively crucial in the future as it was in the past? Our observations and research suggests not — that there are other performance parameters that will become increasingly more important. To explain why we believe the future will look different from the past, we must look at the semiconductor industry through the lenses of models — disruptive technologies and the drivers of modularity.

Clayton Christensen has described a phenomenon he has termed "disruptive technology" that can lead to highly successful incumbent firms being unseated from their leadership positions.² He has shown that interaction of two trajectories — the pace of performance improvement that innovating firms make available, and the rate of improvement that customers can utilize — triggers fundamental changes in the way companies need to compete.

1. Implications of the disruptive technologies model for product architecture and industry structure.

As depicted in Figure 1 , the pace of technological progress almost always exceeds the rate at which customers can utilize those improvements. As a result, the performance delivered by products can, over time, come to exceed the performance that customers in any given tier of the market can utilize. When this "performance overshoot" occurs on a historically valued measure used by firms to garner premium prices, customers cease to value further improvements on that measure. These customers will gladly accept increased performance but will be increasingly unwilling to pay a premium for it. Market forces then compel companies to find new ways to attract overserved customers, whose attention shifts to value provided along new trajectories of innovation. Examples of such alternative dimensions of competitive improvement include convenience of use, product variety and/or ease of customization, and speed to market.

After product functionality has overshot what customers in any given tier of the market can utilize, improvements in speed to market, and in the ability to conveniently customize the features and functions of the products to the needs of specific customers, become the trajectories of innovation that gain competitive traction and earn premium prices. To compete on these dimensions — speed, convenience and customization — technologically interdependent product architectures tend to give way to modular architectures, in which the interfaces among subsystems become well specified and standardized.³  This enables system designers and assemblers to respond quickly to changes in customer requirements by upgrading certain subsystems without redesigning everything, and to mix and match best-of-breed components to quickly and conveniently customize. In turn, the modular architecture enables focused, independent companies to thrive making only one component or subsystem of the product. When product functionality has become more than good enough, focused suppliers that interact within a modular architecture, and supply non-integrated firms that design and assemble end-use systems, enjoy strong advantages over integrated competitors in terms of cost, speed and flexibility. ⁴  Whereas the dominant business model on the left side of Figure 1 is vertical integration, on the right side the integrated firms tend to be displaced by a population of specialized, non-integrated firms. ⁵

Because shifts to new dimensions of innovation are triggered by overshooting, they occur at different times in different tiers of the market. The shift starts in the least demanding tiers of the market. In a multi-tiered market, therefore, we would expect that, in the most demanding applications, competition would revolve around functionality and reliability, even while competition for customers in less demanding tiers revolves around speed, convenience and customization. Truly price-driven competition — meaning that price premiums cannot be earned by improving in any dimension — occurs only in any market tier where improvements on the other dimensions have satiated what customers can absorb. Because the pace of technological progress outstrips the ability of customers to utilize it, however, these new dimensions of innovation and competition creep inexorably up-market over time. As they do, non-integrated companies whose focused offerings interface within modular architectures invade up-market as well.

2. The traditional farm layout of semiconductor fabs exemplifies the necessary decoupling of tools that produce results that are highly variable.

The rub is that it is easy for firms to become blind to shifts in the types of performance that are valued. This is because history has rewarded their current competencies and business models, and the most obvious opportunities to innovate within those boundaries are with unsatisfied customers in tiers of the market above them. This is where the innovations that the company has structured itself to deliver are rewarded by premium prices. The shift toward modular architectures and disruptive dimensions of competition always takes root among those customers who matter least to the established companies. It is therefore difficult to sense and easy to disbelieve that the dimensions of innovation that led to profitable success in the past may differ from those needed in the future.

In the pursuit of Moore's Law improvements in feature geometries, manufacturers have been making 60% more transistors available to circuit designers per area of silicon, compared with what was available a year earlier. In contrast, the ratio at which designers are able to utilize transistors in circuits of any given tier of complexity has only been increasing 20% per year. In other words, the unthinkable is happening. Because feature sizes are not fine enough to meet the needs of designers of the most complex circuits, Moore's Law is losing its relevance in less demanding tiers. In its stead, new rules of competition are taking its place.

In the microprocessor and semiconductor equipment industries, computing performance and price/performance, fueled by the industry's collective attention to Moore's Law, have been the measures of improvement valued in all tiers of the market. In the future, however, and beginning in the least demanding tiers (such as hand-held devices), speed to market, and the ability to conveniently custom tailor the features and functions of microprocessors to the needs of specific classes of customers, will be the types of innovations that matter. Continued adherence to the Moore's Law trajectory might be necessary to remain competitive, but will unlikely be sufficient to ensure success. The leading companies in these industries are vulnerable to the same trap of disruption that caused many of history's best-managed companies to fail — a trap that is grounded in the belief that continued improvement along the historically valued trajectory of innovation will be sufficient for continued market dominance.

3. In the modified hybrid layout, advances in process understanding and tool characterization allow the limited recoupling of tools.

In every other industry where this transition in the basis of competition has occurred, acceleration in product design cycle time and product lifecycle time has been accompanied by acceleration in manufacturing throughput times. Therefore, it is safe to assume that, as competitive pressures to be faster intensify, somebody will figure out how to move away from high work-in-progress (WIP) wafer-lot approaches common to today's fabs, and produce semiconductor products in a low-WIP, single-wafer, connected flow format. In the value-added chain of the future, the throughput time of fabs will shift from more than adequate to not good enough. Our model implies that the overall architecture of wafer flow as a result will become interdependent and optimized for speed — rather than the modular, bay-centric, WIP-intensive structure that was appropriate in the past. As a consequence, the equipment in fabs with single-wafer flows will need to be configurable to each fab's proprietary architecture. Foundries already are dictating the design rules to which IC designers must conform. We believe that, in the future, they will similarly dictate to equipment makers what is and is not acceptable.

4. In the advanced modified hybrid model, some tool-to-tool wafer cassette moves occur, enabled by further improvements in reliability and variability.

The relentless pursuit of Moore's Law historically has not allowed the semiconductor manufacturing process to mature and stabilize, particularly semiconductor manufacturing equipment. Equipment manufacturers have been driven by their best customers to continuously push the technological frontier to achieve smaller geometries. In this race to satisfy the demands of their best customers, they have designed and built tools to "wring out" the highest performance possible. In this era, high performance means smaller geometries, not tools that are highly reliable.

The tools themselves were highly interdependent, meaning that one component or subsystem of the tool could not be changed without affecting the overall performance of the system (e.g. low reliability, high variability, etc.). In addition, the output of the tool — the results on the wafer — could not be predicted with high degrees of certainty. This perpetuated the need for the segregation of equipment into process-specific bays.

5. As results on the wafer become more reliable and tool performance improvements continue, single-wafer line flow could become possible.

Advances in process understanding and tool characterization allow the limited recoupling of tools depicted in the modified hybrid layout (Fig. 3). Further improvements along the performance trajectories of reliability and variability facilitate the evolution of fab architecture to an advanced modified hybrid (Fig. 4) in which some tool-to-tool wafer cassette moves occur. As standard interfaces emerge (e.g. reliable results on the wafer) and tool performance improvements continue, single-wafer line flow could become possible (Fig. 5 ). Each evolutionary step results in a marked increase in factory efficiency.

As is reasonable given their history, microprocessor design, fabrication and equipment manufacturing firms have been laser-locked on maintaining the compute-performance trajectory established by Moore's Law. Attention currently is paid to shrinking linewidths, building smaller transistors and fabricating larger wafers. Sleep is lost over whether or not this is possible, and how much it will cost.

Will devices hit a physical limit? Probably. But this may be the right answer to the wrong question. The important question is, as technological progress surpasses what users can utilize, how do the dynamics of competition begin to change?

If history is any guide, customers in any given tier of the market still will expect the levels of performance to which they have become accustomed. Delivering these levels of performance will be necessary but not sufficient for competitive success. The growth markets of the future will not be in today's monolithic one-size-fits-all artifacts. Fundamentally different business models, in the form of operating processes and cost structures, will dominate the different ends of the market.

The evolution toward this future will not be driven by or grounded in the choices of managers in today's industry-leading companies. Competition in the relevant tiers of the market will force these new trajectories of improvement to become critical. The only question is which companies will have developed the capabilities and organizational structures required to thrive in these markets.