Silicon is the material of our time. We live in the age of silicon; it is all around us in terms of electronic gadgets and computers. It is often said that silicon is the quintessential electronic material as its use is so ubiquitous.

It has not always been that way. Before the 1940s, transistors did not exist at all and in the early 1950s, transistors were made of germanium. However, germanium transistors are not very reliable, in particular they are difficult to package and process. This changed on May 10, 1954. On this date, Texas Instruments announced the first silicon transistor, which removed the inherent flaws of germanium-based transistors. Within ten years after the invention of the silicon transistor, a silicon chip containing over 2000 transistors was constructed. Today, the Pentium-4 processor made by Intel contains 42 million transistors. By the end of this decade, we will likely see a processor containing one billion transistors. This is an amazing and revolutionary progression of silicon technology. However, I want to focus on an accompanying revolution in our understanding of the theory of materials. In particular, silicon technology has had a profound impact here too!

One measure of scientific impact is to examine the technical literature. In the 1970s, approximately 30?000 papers were published with the word ‘silicon’ in the abstract. In the 1980s, this number increased to 84?000 papers. In the 1990s, the corresponding number is 136?000 (that averages one paper every 90 minutes!). In short, a quarter of a million papers have been written over the last 30 years that mention silicon. For a theorist interested in electronic materials research, this is a ‘treasure trove’ of information. Specifically, such information can be used to test and benchmark theory. This is an imperative activity. Consider the following situation. In a crystal of silicon one might have around 1023 electrons and nuclei. In principle, the application of the known laws of quantum mechanics would allow one to predict all physical and chemical properties of matter. But owing to the complexity of the problem, it is absolutely hopeless to extract physically meaningful results without some dramatic approximations. What approximations will work and how well?

An obvious answer to this question is to employ the silicon database as our benchmark. Such an approach has been attempted by a number of workers whose goal is to understand the electronic structure of solids. For example, the first realistic energy band diagrams for electronic materials were constructed using silicon data. One can approximate the true electronic interactions by a ‘one-electron’ potential that contains the average interactions of all the chemically active (valence) electrons within the system. In the 1960s, these potentials (called ‘pseudopotentials’) were fit to optical data, mostly based on silicon and related materials. This was a key test and the results were strikingly successful. One could accurately replicate the experimental results with just a few parameters. If this initial test had failed, it would have suggested that a one-electron description of matter was not possible. I believe that failure would have set back our current theoretical understanding of the electronic properties by 30 years. I would also claim that many current solid state spectroscopic methods would not have existed if these initial tests failed.

In more recent work, electronic potentials have been fixed not by experiment, but from ‘first principles’. These ‘first principles’ pseudopotentials often rely on approximate forms of density functional theories such as the local density approximation. Again, silicon played a crucial role in assessing the validity of this approach. Initial applications of first principles pseudopotentials included studies of silicon under pressure. This produced some rather remarkable predictions, e.g. that some high-pressure forms of silicon would be superconducting (this was later confirmed by experiment). Recently, the first quantum molecular dynamics were done on silicon, including theoretical studies of the electronic and structural properties of liquid and amorphous silicon. My list is by no means complete (I have not even touched on Nobel Prize winning work on scanning tunneling microscopy of the silicon surface or the role of electronic materials in the quantum Hall effect), but it is representative.

I would argue that without the silicon database, our knowledge of materials would be dramatically curtailed. The advances in silicon technology have played a crucial role in providing the benchmark material of merit for assessing the fundamental science of condensed matter.

Will silicon continue to play this role? I think so. Gordon Moore predicted in 1965 that the number of transistors per integrated circuit would double every 18 months. Moore’s Law has worked for the past three decades and many predict that it will hold for the remainder of this decade. As long as Moore’s Law holds, I see no reason why silicon won’t continue to be the benchmark of choice for theorists for at least the next decade

Read full text on ScienceDirect

DOI: 10.1016/S1369-7021(02)00664-8