By In 1965, Dr. Gordon Moore was asked to write an article describing the future of electronics for the 35th anniversary issue of Electronics magazine. At that time, integrated circuits (IC) were four years old, planar transistors were only six years old, Moore was the director of the research and development laboratories at Fairchild Semiconductor, and they were creating state-of-the-art ICs with 60 components. In this seminal article, Cramming More Components onto Integrated Circuits, (Electronics; volume 37, no. 8; April 19, 1965), Moore postulated his now-famous law: The number of transistors that can be integrated on a chip will grow exponentially by a factor of two.

But thats not really what he said, and he didnt postulate anything. What he did say was the complexity for minimum component costs of ICs had increased at a rate of log 2 per year, and over the short term (10 years), this rate could be expected to continue. He further predicted that by 1975 the number of components per integrated circuit for minimum cost will be 65,000. Moore observed the cost per component is inversely proportional to the number of componentsto a point. As components are added, a point is reached where costs per component begin to rise. That point is the minimum component cost and is determined by the state of the technology at the time. In 1965, it was 50 components per circuit.

So What Is a Planar Transistor?
Digital computers rely on switches that represent two statesbinary 1 or binary 0, on or off. OK, it really is more complicated than that in the real world. On and off are represented by voltage differentials, not absolutes, but for our purpose, on or off will work just fine. The first electronic digital computers used vacuum tubes to function as binary switches. Then, in 1947, Bell Laboratories invented the transistor, a device that was able to control the flow of electricity through a solid material like silicon; hence, the term solid state. Silicon is a substance known as a semiconductor, which means, depending on its state, it either may act as a conductor or an insulator. A basic transistor consists of an emitter, a base, and a collector. Current flows from the emitter to the collector through the base. When a specific voltage (usually negative) is applied to the base, free electrons are repelled, changing the base from a conductor to an insulator. This establishes a method for controlling binary signalsno voltage to the base one state, negative voltage to the base another state. On or off. 1 or 0.

Early transistors resembled little top hats or flat-topped mountains and therefore were called mesa transistors. Mesa transistors were coolthey could be mounted on printed circuit boards or soldered onto an electronic chassis just like any other component. But they still were three dimensional and bulky (particularly bulky when you consider how many are required to create any sort of useful digital computer).

Miniaturization required two technological breakthroughstransistors were a scientific breakthroughthere is a significant difference between a technological breakthrough and a scientific breakthrough. The 1956 Nobel prize in physics was awarded to William Shockley, John Bardeen, and Walter Brattain for their research on semiconductors and their discovery of the transistor effect. It wasnt until 2000 that Jack Kilby was awarded half of the physics prize for his part in the invention of the integrated circuit.

And
The two technological breakthroughs that allowed the creation of the IC were the diffusion and photographic masking process and the planar transistor. Diffusion is the process of adding impurities to a semiconductor to create regions with different conduction properties (i.e., regions with excess negative or positive ions). This technique was coupled with a lithographic process that allowed patterns or masks to be applied to an IC that creates the actual circuitry. The planar transistor then was developed using the aforementioned technology. A planar device is a semiconductor manufactured using diffusion and mask- ing techniques on the wafer itself. Thus, it essentially is a flat and very small transistor.

So, in 1965, when Moore made his predictions, they were based on the technology then in use for creating integrated circuits. That technology could create integrated circuits with 50 devices. Intel currently is shipping the first microprocessors manufactured on its 90-nanometer process technologya technology that allows creation of chips with a half-billion transistors.

Stop the Madness
We can create a device with 5×108 transistors on a 200mm wafer. That is incredible, and it immediately raises two questions. The first is, Are we approaching the limits of this technology? This question really just is challenging the validity of Moores law. More about that in a minute. Another more fundamental, more philosophical question is, Why do we continually require such massive increases in computing power? In 1983, Microsoft Word contained about 27,000 lines of source code; today, Word has more than one million lines of code. NT has something like 16 million lines of code. Are we creating bloated, inefficient software products simply because we now have the power to run such stuff?

I can remember writing code in an era (just a few years ago) when we would pride ourselves on finding the most efficient routine to perform a particular task. We often would drop to machine (well, assembly) code if it was more efficient. Those days are gone. Now, we get a project out the door as quickly as possible with the knowledge that a year from now that code will run even better simply because it will be executed on a faster machine. That may be more efficient in terms of dollars spent (programmer hours are, after all, rather costly), but it certainly isnt very elegant. I dont believe I am any more efficient composing this article on Word 2003 (XP) than when I used to write using WordPerfect 5.1 on DOS. Back then (1995), we had expensive hardware, cheap operating systems, and slightly more expensive application software (based on inflation-adjusted dollars).

There is a dichotomy here. We are using much more sophisticated and efficient equipment. General-purpose computers (PCs) are incredibly powerful. But we have created increasingly complex software to run those machines. Perhaps it is time to start looking at optimizing software for specific tasks and regain some efficiency. When we build enterprise systems, scalability is of utmost concern. When we create desktop software, we throw efficiency out the door. Would you want a programmer who spent the last 10 years working on overfeatured office software fine-tuning your ERP?

What About Moores Law?
Well, in the first place, it isnt a law. It is an observation made based on a few metrics and a few years experience with IC design. (A colleague at Caltech, Carver Mead, is credited with actually calling the observations Moores Law.) It wasnt until the mid-70s, when Moores other prediction about 65,000-component ICs came to pass, that Moores law started to gain respect. By that time, technology had started to slow down a bit, so Moore altered his prediction to postulate a 24-month doubling cycle. In the late 80s, it was revised once again to an 18-month cycle. I suppose that makes this more of an empirical law than an a priori one.

It now is almost 40 years since the original predictions were made. We no longer talk about minimum component cost or even devices per square inch. We now describe ICs in terms of the size of a single transistor. Earlier I mentioned Intels 90-nanometer process technologywhich can create 50nm transistors. A nanometer is one billionth of a meter.
These transistors feature gate oxides that are only five atomic layers thick (1.2 nm). You might notice some new terminology here. Forty years have passed; we cant keep talking about emitters and collectors. The new technology for IC planar transistors uses the source (emitter), the drain (collector), and the gate (base). Current flows from the source through the gate to the drain. Transistors are measured in terms of the size (linear) of the gate. The next iteration of Intel magic (65-nanometer technology) will feature six transistors with gates just 35nm long into an area of 0.57 square micrometers. The thickness of these gates is measured in atomic units (that is the size of atoms!). Surely we must be approaching the limits of this technologyright?

Right! Maybe
According to the latest research, we may, in fact, be approaching the limits of CMOS (Complementary Metal Oxide Semiconductor) technology. A recent paper, Limits to Binary Logic Switch ScalingA Gedanken Model, was written by four Intel researchers and published in the November 2003 Proceedings of the IEEE. To fully understand this paper, I recommend you brush up on your understanding of quantum mechanics and the Heisenberg uncertainty principle in addition to advanced physics. If you dont have time for that, I can sum it up in a few sentences. It appears it probably will be impossible to create working transistors smaller than the 22-nm process technologywhich translates to a nine-nanometer gate length. Most scientists agree we will be able to manufacture 22-nm chips in about 15 years. Beyond that, all bets are off. That places the terminus of Moores law around 2018give or take a few years. The next step would be 16-nm process, which would result in a gate length of five nanometers. At that point, the source and drain are so close it will become impossible to predict electron location (quantum physics is, after all, about probabilities). Spontaneous transmission or tunneling through the gate is likely to occur. The Heisenberg uncertainty principle comes into effect and postulates we will have no way of knowing whether an individual switch will be on or off. In computer science, not knowing is unacceptable. Thus, we appear finally to have defined the limits of Moores Law. There are, of course, other considerationspower leakage, heat build-up, stray electrons. Any of the aforementioned probably is sufficient to bring this continued doubling to a halt.

The bottom line is Moores has been a self-fulfilling prophecy based on continually reducing the size of components. If you recall, the original prediction didnt say we can double the number of transistors on a chip every year. It was about complexity and minimum component cost. Maybe its time to give Moores law the boot. Make some bigger chips, find new materials, whatever. I know Moore had no idea what a conundrum he was creating 40 years ago. Let it rest in peace.

I Come to Praise Moore, Not to Bury Him
We shouldnt mourn the demise of Moores law. It is, after all, an exponentialand all exponentials are doomed to end. I remember when my father offered me a penny today and two cents tomorrow and four the next day, etc. I think this was in response to my request for a dollar. Being a shortsighted five-year-old, I took the dollar. Big mistake. I could have been the richest kid in kindergarten.

How small is a nanometer?
The size of molecules ranges from about 0.1 nanometer for simple molecules up to about 50 nanometers for complicated biological macromolecules such as proteins and enzymes. In comparison, a human hair is 150,000 nm in diameter and represents the smallest feature an unaided human eye can see. A water molecule is about 0.3 nanometers in diameter. In other words, a nanometer is almost inconceivably small!

Special Reports /

Born 1965, Died 2018Moores Law: Yes, it served technology well. May it now rest in peace.

Related Stories

Recommended For You