Moore s Law

Moore's Law

Moore's Law is commonly reported as a doubling of transistor density every 18 months. But this is not something that Gordon Moore, co-founder of Intel, has ever said. It is a nice blending of his two predictions; in 1965, he predicted an annual doubling of transistor counts in the most cost-effective chip, and he revised it in 1975 to every 24 months. With a little hand waving, most reports attribute 18 months to Moore's Law, but there is quite a bit of variability.

The popular perception of Moore's Law is that computer chips are compounding in their complexity at a nearly constant per-unit cost. This is one of the many abstractions of Moore's Law, and it relates to the compounding of transistor density in two dimensions. Others relate to speed (the signals have less distance to travel) and to computational power (speed x density).

So as not to miss the long-term trend while sorting out the details, we will focus on the 100-year abstraction of Moore's Law. But we should digress for a moment to underscore the importance of continued progress in Moore's Law to a broad set of industries.

The Importance of Moore's Law

Moore's Law drives chips, communications, and computers and has become the primary driver in drug discovery and bioinformatics, medical imaging and diagnostics. Over time, the lab sciences become information sciences, modeled on a computer rather than trial-and-error experimentation.

The NASA Ames Research Center shut down its wind tunnels this year. Because Moore's Law provided enough computational power to model turbulence and airflow, there was no longer a need to test iterative physical design variations of aircraft in the wind tunnels, and the pace of innovative design exploration dramatically accelerated.

Pharmaceutical giant Eli Lilly processed 100 times fewer molecules this year than it did 15 years ago, but its annual productivity in drug discovery did not drop proportionately; it went up over the same period. "Fewer atoms and more bits" is Lilly's coda.

Accurate simulation demands computational power, and once a sufficient threshold has been crossed, simulation acts as an innovation accelerant compared with physical experimentation. Many more questions can be answered per day.

Recent accuracy thresholds have been crossed in diverse areas, such as modeling the weather (predicting a thunderstorm six hours in advance) and automobile collisions (a relief for the crash test dummies), and the thresholds have yet to be crossed for many areas, such as protein folding dynamics.

Unless you work for a chip company and focus on fab-yield optimization, you do not care about transistor counts. Integrated circuit customers do not buy transistors. Consumers of technology purchase computational speed and data storage density. When recast in these terms, Moore's Law is no longer a transistor-centered metric, and this abstraction allows for longer-term analysis.

The exponential curve of Moore's Law extends smoothly back in time for more than 100 years, long before the invention of the semiconductor. Through five paradigm shiftssuch as electromechanical calculators and vacuum tube computersthe computational power that $1,000 buys has doubled every two years. For the past 30 years, it has been doubling every year.

Each horizontal line on the logarithmic graph given in Figure 4-1 represents a 100 times improvement. A straight diagonal line would be an exponential, or geometrically compounding, curve of progress. Kurzweil plots a slightly upward curving linea double exponential.

Figure 4-1. One-hundred-year version of Moore's Law. Each dot is a computing machine. (Source: Ray Kurzweil.)

Each dot represents a human drama, although the humans did not realize that they were on a predictive curve. Each dot represents an attempt to build the best computer using the tools of the day. Of course, we use these computers to make better design software and algorithms (formulas) for controlling manufacturing processes. And so the progress continues.

One machine was used in the 1890 census; another one cracked the Nazi Enigma cipher in World War II; still another one predicted Eisenhower's win in the presidential election. And there is the Apple II, and the Cray 1, and just to make sure the curve had not petered out recently, I looked up the cheapest PC available for sale on Wal-Mart.com, and that is the gray dot that I have added to the upper-right corner of the graph.

And notice the relative immunity to economic cycles. The Great Depression and the world wars and various recessions do not introduce a meaningful delay in the progress of Moore's Law. Certainly, the adoption rates, revenue, profits, and inventory levels of the computer companies behind the various dots on the graph may go through wild oscillations, but the long-term trend emerges nevertheless.

Any one technology, such as the complementary metal oxide semiconductor (CMOS) transistor, follows an elongated S-curve of slow progress during initial development, upward progress during a rapid adoption phase, and then slower growth from market saturation over time. But a more generalized capability, such as computation, storage, or bandwidth, tends to follow a pure exponentialbridging a variety of technologies and their cascade of S-curves.

If history is any guide, Moore's Law will continue and will jump to a different substrate than CMOS silicon. It has done so five times in the past and will need to again in the future.

Problems with the Current Paradigm

Gordon Moore has chuckled at those in past decades who predicted the imminent demise of Moore's Law. But the traditional semiconductor chip is finally approaching some fundamental physical limits. Moore recently admitted that Moore's Law, in its current form, with CMOS silicon, will reach its limit in 2017.

One of the problems is that the chips are operating at very high temperatures. This provides the impetus for chip-cooling companies to provide a breakthrough solution for removing 100 watts per square centimeter. In the long term, the paradigm must change.

Another physical limit is the atomic limitthe indivisibility of atoms. Intel's current gate oxide is 1.2nm thick. Intel's 45nm process is expected to have a gate oxide that is only three atoms thick. It is hard to imagine many more doublings from there, even with further innovation in insulating materials. Intel has recently announced a breakthrough in a nanostructured gate oxide (high-k dielectric) and metal contact materials that should enable the 45nm node to come online in 2007. None of the industry participants has a CMOS road map for the next 50 years.

A major issue with thin gate oxides, and one that will also come to the fore with high-k dielectrics, is quantum mechanical tunneling. As the oxide becomes thinner, the gate current can approach and even exceed the channel current to the point that the transistor cannot be controlled by the gate.

Another problem is the escalating cost of a semiconductor fabrication plant (called a fab), a cost that is doubling every three years, a phenomenon dubbed Moore's Second Law. Human ingenuity keeps shrinking the CMOS transistor, but with increasingly expensive manufacturing facilitiescurrently $3 billion per fab.

A large component of fab cost is the lithography equipment that is used to pattern the wafers with successive submicron layers. Nanoimprint lithography can dramatically lower cost and leave room for further improvement from the field of molecular electronics.

We have been investing in a variety of companies that are working on the next paradigm shift to extend Moore's Law beyond 2017. One near-term extension to Moore's Law focuses on the cost side of the equation. Imagine rolls of wallpaper embedded with inexpensive transistors. One of our companies deposits traditional transistors at room temperature on plastic, a much cheaper bulk process than growing and cutting crystalline silicon ingots.