Semiconductors are manufactured by taking an extremely clean semiconductor material (commonly silicon) and introducing impurities to produce transistor and other electrical circuits on the surface of the material. In effect, the processes very similar to the etching that produces a PCB, however the scale of the surface features are about four orders of magnitude smaller. The smallest structure that can be produced on a semiconductor defines the amount of circuitry that can be introduced. Practical limitations mean that there is a rapid law of decreasing returns when the silicon `die' (the square containing each circuit) exceeds a certain size, about 1 inch square. Above this size, inevitable impurities drastically reduce the yield. So the smaller the circuitry can be made the more can be squeezed in, or the smaller the overall die can be made. When I entered the industry in 1987, 3 micron structures were about state of the art. Today 0.18 or 0.15 micron is leading edge and a circuit that required 1 square inch in 1987 will today require perhaps 0.01 square inch to manufacture - a factor of 100:1 improvement. Other process improvements have realised even greater density increases.
This size reduction has roughly equivalent effects on the cost, power consumption and maximum possible speed of the circuit. What was not possible 10 years ago is commonplace today... the growth in the computing power of the personal computer is a good example of the effect this has on devices that use a lot of semiconductors.
The future becomes a little more clouded as we consider what will happen over the next ten years. A number of barriers are now being hit. These include a rapid drop in the breakdown voltage of the semiconductor itself (modern chips have to be run at 1.8V or lower and this drops with each new generation) and the speed limitation of the aluminium interconnect. Also important is the fact that the structures are becoming smaller than the wavelength of the light beams being used to etch them and even quantum effects are being encountered at the size that the circuits are now operating. Every previous barrier has been overcome and I believe the semiconductor industry will continue to reduce the cost and increase the performance of the next generations of parts at the same rate.
Designing chips used to be a case of hooking up transistors and testing. (Texas Instruments famous 741 op amp had 47 and was designed this way.) This rapidly gave way to `gate' level design (for digital circuits), where pre-defined NAND, AND, and OR structures, each consisting of several transistors were hooked together. The next level was to recreate TTL blocks on chips (counters, multiplexers, flip flops) and hook them together into the required circuit.
Today a designer might be looking at a chipp with perhaps 25 million transistors as a leading edge design. To create the circuit he has to use large pre-defined blocks. These would consist of memories, processors and DSP cores, which can be quickly and easily placed on the design. Hooking them together and proving the circuit works correctly has become the key challenge....
This sort of design has moved rapidly away from schematics. Today, advanced designers work with high level description languages... similar to computer programs these define the circuit operation and are then `compiled' to produce the actual transistor configuration. Circuits are exhaustively simulated before actual manufacture - even so, removing bugs from silicon designs is a major headache.
Another major trend is for application specific chips (ASIC)'s - these are effectively blank chips whose configuration is defined by the customer. The fastest growth area in semiconductors today is for programmable ASIC's, otherwise known as field programmable ASIC's (FPGAs). These are chips that can be re-configured by the user even after the chip has been placed on the PCB, this gives much needed design flexibility.
Types of chips
The first semiconductors were analog;- op amps and comparators - made by connecting a few transistors together. Today there are many families of products, this is a brief summary:
Semiconductors used to store the ones and zeros of computing (bits) with the capacity of a single chip reaching 16Million bits. There are several main types:
DRAM - Dynamic RAM has the highest density but has to be continually `refreshed'. Effectively the chip only has a short term memory and the computer has to remind it every few milliseconds as to what is stored. Used as the high density memory in computers and other equipment.
SRAM - Static RAM holds its contents as long as the power is on. Lower density and higher speeds mean it is usually used as a`cache' for a larger DRAM or in power critical applications.
Flash - Using the tunnelling effect of electrons, high voltages can drive them through an insulating barrier to deposit charge on an insulated conductor. At normal voltages, this becomes nonvolatile memory, the charge remains even when the power is off. Used to store the `code' required by a computer on power up. Other applications include the images from your digital camera.
The need for analog circuits hasn't gone away and today there is a wide selection of amps, A/D and D/A converters, multiplexers, reset controllers and so on. Analog Devices, Linear Technology, Maxim, Dallas and Elantec are leading manufacturers all with excellent catalogues available on their web sites.
From the 4 bit micros controlling your washing machine to the monster 64 bit CPU at the heart of a modern PC, micros are now found nearly everywhere. Best estimates are that some 7 BILLION have already installed worldwide. Modern manufacturers include the famous Intel, their competitors AmD and at the other end of the spectrum Motorola and Microchip.
Hooking micros together is the logic `glue' - traditionally built of TTL blocks (TI, Philips, National) but increasingly made from programmable chips (Xilinx and Altera are the leaders) and for large volumes, ASICs.
In the last fifteen years, third party companies have started to provide wafer fabs (chip factories) as a sub contract service. This means any team of designers can produce a chip, have it made and sell it on the open market for a relatively low start up cost. Many have done so, usually targeting a specific market segment. The graphics chips in modern PC's are good examples of specific circuits that are targeted by small specialist companies.
Where's it all going?
The chip business is volatile in the extreme, however when the bumps and peaks are smoothed out, it has achieved a 15% annual growth rate every year for the last 30 years. At that rate of increase it is set to become one of the biggest, if not the biggest industry in the world in the next decade. Cars and tourism are vying for the top slot. At the same time the unit price of the average chip has dropped by perhaps 20% / year. The next major revolution looks set to be driven by the internet - getting everything `on line' so your home can be controlled through your PC, security cameras, process monitors and your mobile telephone - every type of information gathering device will be accessible through a computer. This will require a new generation of cheap `Internet protocol' chips to connect your washing machine to the net at a very low cost. It may sound futuristic, but the products are already being developed.
Chips get everywhere - in the past fifteen years PC's and mobile `phones have been the driving force behind much of the semiconductor industry's growth. The Internet may provide more stimulus than either of them. One things for sure, you will use perhaps 20 semiconductors today and many, many more in the coming decade.