SCIENCE

Silicon Chips


Dr Jonathan Edwards, ST Microelectronics UK Ltd. on 27 July 2001

Microprocessors are small electronic components that are now found in almost every consumer product. They appear in cars to control the engine and other mechanical components, and also in communication systems from radios to satellite navigation systems. In the home, they manage burglar alarms and cookers, radios and music centres, terrestrial and satellite TVs. The devices are shrinking in size, but their capabilities are increasing each year. In colour televisions, less than 5% of the cost is in silicon; whereas in a modern digital camera it makes up about 35% of the cost.

If we look at the number of designs that are started for electronic products, the number per year is stable. There seems to be a natural rate of invention and ideas for new products. But in terms of the number of transistors (the basic electronic module within a chip), the number per product is growing at an annual compound rate (CAGR) of 23%. This translates into significant opportunities for more functions in smaller spaces.

In 1965, Gordon Moore of Intel Corp. stated his now famous Moore's Law that the number of transistors, and hence the capability, of each chip was doubling each 18 months. In fact the CAGR has been 59% over the last 20 years. In 1994, a typical chip had 2000 gates and the minimum width of components was 0.5 micron. In 2000, this had become 450 000 gates at 0.15 micron, and we are now starting to create chips at 0.13 micron. By 2004, the gate size is expected to be less than 0.10 micron.

Along with this increase in capability, there has followed a demand for a single supplier to provide all the components by which these chips can be utilised. The full range of design technology, software development for applications, and the know-how to put them all together comes from a few large companies as a complete package. They work closely with the end product designers, who then build the boxes and add the manual controls onto the system boards.

Over the years, design times for products have been shortening, and volumes per product have been increasing. When the first black and white televisions were introduced, it took 20 years before 1 million units had been made and sold. After the first VCRs appeared it took 8 years to sell a million. Now, it is expected to sell 1 million DVD players in less than one year.

All of these factors are putting on more and more pressure to increase the densities and improve the yield of production of these ubiquitous chips.

A `chip' is largely glass, into and onto which have been placed carefully contrived impurities and metal wires. At the base, is a substrate of semiconductor material on to which is laid a largely regular array of transistor and other standard circuits. From these a set of leads of tungsten are formed vertically to a layer where metal connections are made between the circuits. Adding tungsten leads is then repeated to make the necessary interconnections in up to 7 separate layers of metal.

The design process starts with defining a set of standard combinations of the basic units into a library of circuits, which is presented to a piece of software known as the design tool. The engineer then designs with these higher level library circuits and the tool automatically lays out the connections required between them. From these it can generate the masks required to lay down the metal connections at each stage of the manufacturing process, all automatically. If the current density during operation is too high, the metal migrates under the pressure of the flowing electrons, and this can create voids and chip failure. With the smaller dimensions now present, this has become a critical design issue.

One of the standard circuits is a memory cell. As the size of the semiconductor connections has reduced, to obtain the necessary capacitance, these memory modules have now become taller rather than wider. In addition, at each gate, the thickness of oxides at the contacts is now only an average 5.6 molecules thick, and each silicon dioxide molecule is about 0.312 nanometer diameter.

Andy Pepperdine