By In 1965, Dr. Gordon Moore was asked to write an articledescribing the future of electronics for the 35th anniversary issueof 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 laboratoriesat Fairchild Semiconductor, and they were creating state-of-the-artICs with 60 components. In this seminal article, Cramming MoreComponents onto Integrated Circuits, (Electronics; volume 37, no.8; April 19, 1965), Moore postulated his now-famous law: The numberof transistors that can be integrated on a chip will growexponentially by a factor of two.

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

So What Is a Planar Transistor?
Digital computers rely on switches that represent two statesbinary1 or binary 0, on or off. OK, it really is more complicated thanthat in the real world. On and off are represented by voltagedifferentials, not absolutes, but for our purpose, on or off willwork just fine. The first electronic digital computers used vacuumtubes to function as binary switches. Then, in 1947, BellLaboratories invented the transistor, a device that was able tocontrol the flow of electricity through a solid material likesilicon; hence, the term solid state. Silicon is a substance knownas a semiconductor, which means, depending on its state, it eithermay act as a conductor or an insulator. A basic transistor consistsof an emitter, a base, and a collector. Current flows from theemitter to the collector through the base. When a specific voltage(usually negative) is applied to the base, free electrons arerepelled, changing the base from a conductor to an insulator. Thisestablishes a method for controlling binary signalsno voltage tothe base one state, negative voltage to the base another state. Onor off. 1 or 0.

Early transistors resembled little top hats or flat-toppedmountains and therefore were called mesa transistors. Mesatransistors were coolthey could be mounted on printed circuitboards or soldered onto an electronic chassis just like any othercomponent. But they still were three dimensional and bulky(particularly bulky when you consider how many are required tocreate any sort of useful digital computer).

Miniaturization required two technologicalbreakthroughstransistors were a scientific breakthroughthere is asignificant difference between a technological breakthrough and ascientific breakthrough. The 1956 Nobel prize in physics wasawarded to William Shockley, John Bardeen, and Walter Brattain fortheir research on semiconductors and their discovery of thetransistor effect. It wasnt until 2000 that Jack Kilby was awardedhalf of the physics prize for his part in the invention of theintegrated circuit.

And
The two technological breakthroughs that allowed the creation ofthe IC were the diffusion and photographic masking process and theplanar transistor. Diffusion is the process of adding impurities toa semiconductor to create regions with different conductionproperties (i.e., regions with excess negative or positive ions).This technique was coupled with a lithographic process that allowedpatterns or masks to be applied to an IC that creates the actualcircuitry. The planar transistor then was developed using theaforementioned technology. A planar device is a semiconductormanufactured using diffusion and mask- ing techniques on the waferitself. Thus, it essentially is a flat and very smalltransistor.

So, in 1965, when Moore made his predictions, they were based onthe technology then in use for creating integrated circuits. Thattechnology could create integrated circuits with 50 devices. Intelcurrently is shipping the first microprocessors manufactured on its90-nanometer process technologya technology that allows creation ofchips 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. Thefirst is, Are we approaching the limits of this technology? Thisquestion really just is challenging the validity of Moores law.More about that in a minute. Another more fundamental, morephilosophical question is, Why do we continually require suchmassive increases in computing power? In 1983, Microsoft Wordcontained about 27,000 lines of source code; today, Word has morethan one million lines of code. NT has something like 16 millionlines of code. Are we creating bloated, inefficient softwareproducts simply because we now have the power to run suchstuff?

I can remember writing code in an era (just a few years ago)when we would pride ourselves on finding the most efficient routineto 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 theknowledge that a year from now that code will run even bettersimply because it will be executed on a faster machine. That may bemore efficient in terms of dollars spent (programmer hours are,after all, rather costly), but it certainly isnt very elegant. Idont believe I am any more efficient composing this article on Word2003 (XP) than when I used to write using WordPerfect 5.1 on DOS.Back then (1995), we had expensive hardware, cheap operatingsystems, and slightly more expensive application software (based oninflation-adjusted dollars).

There is a dichotomy here. We are using much more sophisticatedand efficient equipment. General-purpose computers (PCs) areincredibly powerful. But we have created increasingly complexsoftware to run those machines. Perhaps it is time to start lookingat optimizing software for specific tasks and regain someefficiency. When we build enterprise systems, scalability is ofutmost concern. When we create desktop software, we throwefficiency out the door. Would you want a programmer who spent thelast 10 years working on overfeatured office software fine-tuningyour ERP?

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

It now is almost 40 years since the original predictions weremade. We no longer talk about minimum component cost or evendevices per square inch. We now describe ICs in terms of the sizeof a single transistor. Earlier I mentioned Intels 90-nanometerprocess technologywhich can create 50nm transistors. A nanometer isone billionth of a meter.
These transistors feature gate oxides that are only five atomiclayers thick (1.2 nm). You might notice some new terminology here.Forty years have passed; we cant keep talking about emitters andcollectors. The new technology for IC planar transistors uses thesource (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) willfeature six transistors with gates just 35nm long into an area of0.57 square micrometers. The thickness of these gates is measuredin atomic units (that is the size of atoms!). Surely we must beapproaching the limits of this technologyright?

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

The bottom line is Moores has been a self-fulfilling prophecybased on continually reducing the size of components. If yourecall, the original prediction didnt say we can double the numberof transistors on a chip every year. It was about complexity andminimum component cost. Maybe its time to give Moores law the boot.Make some bigger chips, find new materials, whatever. I know Moorehad no idea what a conundrum he was creating 40 years ago. Let itrest in peace.

I Come to Praise Moore, Not to Bury Him
We shouldnt mourn the demise of Moores law. It is, after all, anexponentialand all exponentials are doomed to end. I remember whenmy father offered me a penny today and two cents tomorrow and fourthe next day, etc. I think this was in response to my request for adollar. Being a shortsighted five-year-old, I took the dollar. Bigmistake. 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 simplemolecules up to about 50 nanometers for complicated biologicalmacromolecules such as proteins and enzymes. In comparison, a humanhair is 150,000 nm in diameter and represents the smallest featurean unaided human eye can see. A water molecule is about 0.3nanometers in diameter. In other words, a nanometer is almostinconceivably small!