12/23/2007
On Vacation/Transistor Anniversary
Here’s wishing all of you a Merry Christmas, Happy Holiday or whatever the appropriate greeting may be. Old Bortrum is taking a two-week break and a new column will hopefully be posted on or before January 9 of the New Year. In the interim, I will leave last week’s reprise of an earlier column posted below, presuming that many readers, like myself, have been too busy to keep up with their reading. Let’s hope 2008 brings more good news than bad.
Column posted 12/19/2007:
Last week our StocksandNews Editor Brian Trumbore called my attention to a tidbit in the December Smithsonian magazine noting the 60th anniversary of the invention of the transistor on December 23, 1947. Brian suggested that I mark the anniversary by recycling an earlier column related to the transistor. In view of the fact that I’m still in a care giving mode following my wife’s back surgeries and have not begun to address our Xmas cards or do anything at all related to Xmas, I accept Brian’s suggestion with alacrity. Accordingly, what follows is a column from July 13, 1999 on Moore’s Law, the first meeting I participated in at Bell Labs and what followed from the transistor of most concern for us computer users. Some of the material on Frosch and Derick is the same as that in a more recent column (10/24/2007) but places their work more firmly in the context of the science important to the silicon chip. I’ve deleted one paragraph on another subject.
7/13/1999 Moore’s Law
Troubling to many is the possible breakdown of Moore''s Law roughly a decade from now. This has been a subject of discussion for many years in the technical media but now is increasingly being mentioned in the financial press such as the Wall Street Journal. For those unfamiliar with Moore''s Law, back in 1965 Gordon Moore, a co-founder of Intel, proposed that the computing power of the veritable silicon chip would roughly double every year or so (later modified to 18 months). What he was saying was that the transistors on the chip would decrease in size over the years so that the number of them on the chip would keep increasing over time. The fact that you can now buy high power computers for such low prices attests to the fact that, amazingly, Moore''s Law has held for more than three decades and should continue to be true for at least a fourth. What makes it so amazing is that the prediction was made when there were only about 50 transistors on a chip and now the Pentium II chip, for example, has over 7 million!
When I first joined Bell Labs in 1952, I sat in on meetings attended by William Shockley and Walter Brattain, two of the co-inventors of the transistor, Ian Ross (later president of Bell Labs), George Dacey (some years ago, as a stockholder I cast my ballot for him as a director of W. R. Grace), John Hornbeck (later president of Bellcomm and then Sandia Laboratories), and others who would contribute heavily to the silicon revolution spurred by the transistor. As a young chemist who wasn''t expert in anything, I sat in awe of these truly brilliant scientists discussing an idea for what to me was an incomprehensible device called a junction field effect transistor (FET). In 1953, Dacey and Ross made such an FET using germanium. However, the germanium device performed poorly due to the presence of so-called "surface states" which loused up the FET''s performance.
To understand one contributor to these surface states, let''s talk about the crystal structure of germanium, which is the same as that of diamond. In the diamond structure carbon, as with silicon and germanium, has 4 electrons that are relatively "loose" or as we chemists call them, "valence" electrons. Each of these 4 electrons wants to get together with another electron to form a chemical bond. It''s rather like each carbon atom has 4 arms stretched out trying to shake hands with the electrons from 4 other carbon atoms. These "handshakes" form very strong bonds as we might judge from the well-known hardness of diamond. So, we have each carbon shaking hands with 4 other carbons around it, EXCEPT AT THE SURFACE. Here, some of the carbon atoms'' arms are reaching out, dangling in space with nothing to shake hands with. These are known in the trade, appropriately, as "dangling bonds" and tend to interfere with the desired flow of electron in the FET. The bothersome "surface states" are partly due to these dangling bonds.
Serendipity often plays a role in science and technology. Two of my good friends and colleagues at Bell Labs were Carl Frosch and "Linc" Derick. Carl was older and somewhat of a father figure to me and was a chain smoker who always had two or three cigarettes during lunch. Linc was a tough-as-nails former fighter pilot who flew cover for Patton in World War II and was a meticulous experimentalist. Carl and Linc had been trying to diffuse impurities into silicon by placing silicon wafers in a furnace with the vapor of the particular impurity, say phosphorus, flowing over the hot silicon. But, they kept having problems with pitting of the silicon and nonuniform penetration of the impurities. (Good device performance depends on control and uniformity of impurities in the silicon.) They tried using all kinds of gases to improve the situation. One day in 1955, Derick was running an experiment using hydrogen. The regulator on the hydrogen tank wasn''t working properly and the pressure kept falling. At one point, Derick turned the valve to increase the pressure. However, when he returned later, the pressure had dropped to zero. Expecting the worst, he took out the silicon wafers and, instead of being pitted, he found them beautifully smooth and colored, "looking like jewels", he told me. Frosch immediately recognized that what had happened was that the hydrogen had "flashed back", reacting with the oxygen in the air to form water, actually steam. The hot steam reacted with the silicon to form very thin, smooth layers of silicon dioxide (the sand on a beach is silicon dioxide). The pretty colors were interference colors just like the rainbow effect you get with very thin oil films on water.
Frosch and Derick quickly tried adding water vapor deliberately to the hydrogen and got the same "jewels". They also showed that the oxide film could mask, or block the diffusion of certain impurities into the silicon. Not only that, but within a few weeks Jules Andrus and Walter Bond showed that, using something called photolithography, windows could be etched in the oxide layers. By using these windows, impurities could be diffused into the silicon only in the selected areas. Thus, within the space of a few weeks, the whole process was created for oxide masking and photolithography to control the precise geometry of impurities in silicon. This fundamental, but not too widely appreciated work laid the foundation for the manufacture of the integrated circuits on the silicon chips as we know them today.
"What does this have to do with the FET?" you ask. Well, in 1959 John Atalla at Bell Labs showed that this oxide layer of Frosch and Derick could be incorporated into a very well behaved FET that became known as the metal-oxide- semiconductor FET or MOSFET. OK, you say this is just another complicated device, why should we care? It happens that the year before, a chap at Texas Instruments, Jack Kilby had invented the integrated circuit, which incorporated several devices on a single chunk of silicon. (Robert Noyce, another co- founder of Intel, then at Fairchild, had a similar idea around the same time and is acknowledged as making an important contribution.) Eventually, the MOSFET was married to the integrated circuit and the MOSFET became the key computer memory device. Jean Hoerni, also at Fairchild, showed that leaving the masking oxide in place after fabrication of the device led to more stable performance, serving to "passivate" the surface.
What about Moore''s Law? The reason Moore''s Law has held all these years is that refining of the masking, etching and diffusion techniques has permitted a continuous reduction in the size of the MOSFETs so that today''s chips contain those millions of FETs. But what are the limits? Today, the thickness of the oxide layer is only about 15 silicon atoms thick. By 2010, this thickness is expected to be closer to 5 silicon atoms thick. Without going into details here, this insulating silicon oxide layer has served to "contain" the electrons in their proper place when they flow through the MOSFET. Unfortunately, there is a phenomenon in quantum mechanics known as "tunneling". This tunneling is sort of like the following situation. You come to the end of a road at the bottom of a hill and you want to get to the other side. Normally, you would have to push or drive your car to the top of the hill and hopefully coast down. If you were an electron, however, there is this weird thing in quantum mechanics that says there is a definite chance that you can tunnel through the hill and be on the other side. This is the problem for Moore''s Law. The electrons can tunnel through the silicon oxide layer when it gets too thin and when that happens it''s a whole new ballgame. Needless to say, many clever people are already working feverishly on ideas to keep Moore''s Law intact well into the 21st century!
Are there any loose ends we''ve left hanging? You''re right, what about those dangling bonds? Well, it turns out that the answer lies again with Frosch and Derick''s oxide. Those outstretched arms of the surface silicon atoms find in the oxygen plenty of electrons to shake hands with and, with the oxide formation, the dangling bonds disappear. Problem solved!
Neither Frosch nor Derick are with us anymore and just recently I had to retire one of the first red LED digital clocks, which Linc made for me on one of my service anniversaries at Bell. If the editor could tell me how to incorporate pictures in this column, I would finish with a picture that Carl Frosch used when he gave the first paper on the oxide masking process. It shows a masked and etched silicon chip that, in contrast to today''s crisply executed integrated circuits, spells out in very ragged fashion "THE END".
Allen F. Bortrum
|