[chris (OP)]

Computer chips contain millions of transistors, but what are transistors, and how does a transistor work?

Chris

[Geezer]

We could probably generate several books trying to provide a comprehensive answer to that question. However, I'll take a shot at a very brief description.

There are many different types of transistor, but rather than get all tied up in the different technologies, it's probably best to focus on a couple of the mechanisms that allow transistors to be used in electronic circuits. A transistor can be thought of as a "controllable resistor", which is actually the origin of the name. Many transistor types have three terminals. If you imagine that there is a resistor between two of those terminals, in a common type of transistor, the value of the resistor can be adjusted by applying a current to the third terminal. The greater the current applied, the less the resistance becomes.

I'm making some rather gross simplifications, but that's a pretty good model to keep in mind for the moment.

The variable resistor can be used to control the flow of current through it, and that current can be much greater than the control current that's applied at the third terminal. Therefore, a small current can be used to control a much greater current flow. In other words, this type of transistor is a current amplifier. This feature can be used in all kinds of analogue amplifying devices. In audio applications, multiple transistor stages will be used to amplify very small signals into signals capable of driving a lot of power into loudspeakers.

In digital equipment (computers etc.) transistors are used as switches rather than analogue amplifiers. In this situation, the control input will only have two steady states, so the resistor will either have a large resistance or a small resistance. Multiple transistors are connected into fast switching amplifiers, and some special transistors may have multiple control inputs so that logic "gates" can be constructed. The multiple inputs might be arranged so that, for example, if any one of them is applying current, the controlled resistor will have a low resistance. This might be referred to as an "NOR" function. By designating high and low currents to represent binary values. In this case, the "N" means that the output or "OR" gate is inverted. So, if either input is "on" (usually represented by "1") the output is "off" (usually represented by a "0"). If both inputs are "off" (0) the output will "on" (1).

It is possible to build any digital system of arbitrary complexity entirely out two input NOR gates. In practice however, computers employ many different types of specialized gates to optimize speed and minimize silicon area and power consumption. Those employed in memory devices are very highly specialized.

Again, experts in the field will note that I have made many simplifications here!

[LeeE]

The only thing I'd add to Geezer's very good explanation is that the 'controllable resistor' also incorporates a one-way valve, so the current can only flow in one direction.

[graham.d]

Lee, that's not really true of MOS devices. The source - drain current flows in either direction. But we are talking about simple modelling here.

[LeeE]

Oh! - I didn't know that - I still tend to think in terms of Germanium PNP or NPN transistors.

[Geezer]

I owe all my knowledge to my crystal set  []

BTW, I really only answered the application part of the question. I never got into the how do they work part. Would someone like to take a shot at it? If I do it, I'll have to blow the dust off my books first!

[graham.d]

Your explanation was pretty good, Geezer. Describing how a transistor works (bipolar or MOS) can be done at lots of levels and going down to quantum mechanics is more than what is needed I expect, and more than I can do without getting the books out either :-) What you said about using the voltage on a gate to "switch" a resistor between a low value conducting state and a very high value, non-conducting state and then making logic gates with these structures is quite right.

The transistors in computer chips are MOS type transistors and are in two forms, PMOS and NMOS. A PMOS device will be switched on when its "gate" voltage is lower than either of the other terminals (source or drain) by a certain value (threshold voltage, Vtp) and an NMOS device is turned on when its gate voltage is higher than either the source or drain voltage by a certain value (threshold voltage, Vtn). The source and drain terminals on a MOS transistor on an integrated circuit are usually interchangeable; it is conventional to refer to the source as the lower voltage in an NMOS device or the higher voltage of a PMOS device, but each device is really symmetrical in construction (though not always so in specific more sophisticated structures).

I started to write a description of how the MOS devices were constructed and how they worked in simple terms but it could easily be pages and pages. I would suggest looking the subject up on the web (with diagrams) rather than try to do that here. In not so layman terms, most digital chips are variations of a process whereby the basic wafer is doped to be p-- (weakly doped p-type silicon). Local ion implantaion with higher doped regions (p-) is used to define the areas where the NMOS devices will be formed. Deep ion implantation to form wells of n- which defines areas for the PMOS transistors. A PMOS works the same way as an NMOS transistor but everything (doping and voltages) is the opposite polarity. The NMOS source and drain are highly doped (quite low impedance) implanted n+ areas. The gate is an area of conductor (polysilicon) over very thin oxide that is between the source and drain. A positive voltage on the gate will repel the p (holes) carriers from the surface and a sufficient voltage (the threshold voltage) will cause a thin layer to "invert". This is where it can only really be understood with Quantum Mechanics. The surface behaves differently from the bulk because the atoms at the surface have "dangling bonds". The inversion causes the surface to become n-type by accumulation of electrons from the nearby source/drain regions, allowing conduction between these regions. Reducing the gate voltage reverses the situation and the device becomes non-conductive again. Virtually no current flows from the gate as the control is via the field over the gate oxide. With very modern very fine geometry device the gate will conduct a bit through tunneling and this is a nuisance. The device behaves differently in distinct regions of operation: Off state, sub-threshold conduction state, triode conduction region and saturated conduction region. Simplistically, for digital devices, the first state is off, the second state is avoided for any significant time, and the 3rd and 4th states are on.

Best I can do without making this a book.

Fabrication of the wafers involves lots of processing steps (well over 100 nowadays) and typically the photolithograpy will use 30 to 40 masks. Wafers will often be fabricated with two types of NMOS/PMOS transistors to allow operation at speed with thin gates, fine geometries and low voltage and also with thicker gate, larger geometries and thicker gates to alow interfacing at higher voltages. To allow efficient connectivity between the logic gates, 6 or 7 levels of metal may be used with vias between one level and the next. There are lots of variations on this theme though.

[techmind]

A large proportion of modern chips are built upon MOS transistors (Metal Oxide Semiconductor), comprising a metal 'gate', oxide 'gate-insulator' and semiconductor (doped silicon) 'channel'. N-MOS = N-doped semiconductor, P-MOS = P-doped semiconductor, and CMOS = complementary-MOS ie a mixture of N-MOS and P-MOS transistors on one device (chip).

The transistor controls the flow of current through its channel, which has a contact at boths ends. The flow of current is controlled by the voltage-potential on the gate (relative to the voltage on the channel). The doping (N- or P-) affects whether you need a gate-voltage positive or negative relative to the channel for conduction to occur.

For a simple MOS transistor (of which you would find a few million on a modern TFT (thin-film transistor) LCD monitor or television) you can think of it as being a bit like a sandwich or cream-tea:

First you have the substrate (like the glass of an LCD): this is like the bread, or scone.
Then you have the (silicon) semiconductor channel: this is like the butter
You have metal electrical contacts at both ends of the channel (e.g. at the left and right sides of the slice of bread)
On top of the butter, you have the jam. This represents the oxide insulator.
On top of the jam, you have the cream. This represents the metallic gate of the transistor which is its 'control' input.

For an LCD display, the dimensions of the transistors are about 1-2um for the smallest feature-size. For silicon chip microprocessors, we're down to a few 10's of nanometers.


Transistor circuits and chips are manufactured by depositing each layer (semiconductor, metal, oxide, etc) over the entire substrate, then covering that layer with a photoresist chemical, exposing the photoresist to ultaviolet light through a 'mask' which defines the pattern you want in that layer. The photoresist is then chemically removed where it's been exposed (or vice versa if positive/negative reist), then etching chemicals are used to etch away the unwanted layer through the holes in the photoresist mask. When that's complete, the remaining photoresist is removed. The next layer is then applied to the device, more photoresist applied, exposed through a mask, etched away... etc etc.
(This is not particularly environmentally friendly!)

You need 5-6 layers (and masks) as a bare minimum to do anything useful, allowing for the metal interconnects as well as the active transistors. I can well believe that you might use 30-40 masks for a really complicated IC.

[Geezer]

Thanks! The only thing I'd add is that MOS transistors belong to a class of devices known as Field Effect Transistors. There are other types of transistors, but MOS devices seem to dominate the the field at the moment.

[Geezer]

One point I'd like to make is that transistors do not actually "amplify" signals; however, they do allow a smaller signal to modulate/control a larger signal.

Interesting point DD. I'm not sure I really appreciate the distinction though.

(I will resist the temptation to ask you to amplify your point  [])