The Art, Science and Politics of Electronic Components

Electronic components are the vocabulary of electronic design. In the same way that English vocabulary constrains the ways that we can express ourselves, the the available components determine how designs are expressed. Our language also sets limits on what thoughts we can express at all. Similarly, the available components constrain what things we can design. I think that electronic components provide an interesting case study in the interaction between physical possibility, physical laws (or models), technology and society.

### Idealized Physics:

The simplest components have two connections (or terminals) and are known as passive components. The simplest physical model of passive components is the linear model. A truly linear two-terminal component would have only one property, its impedance, which is the relationship between the current through the component and the voltage across the component. For example, if we apply a fixed voltage (V) to a component, a certain current (I) will flow.

The relationship between fixed voltage across two terminals and the corresponding current flow is described by Ohm's law:

E = I * R
R is the component characteristic known as resistance, which is an impedance that does not depend on the frequency of the applied voltage. If the impedance varies with frequency, the impedance is said to be capacitive or inductive. According to the ideal linear model, a two-terminal component can be entirely described by its resistance, capacitance and inductance.

### Commerce (and Applied Physics):

An electronic component catalog has large sections for resistors and capacitors, and a smaller section for inductors. You will find that there are many many different kinds of resistors, capacitors and inductors. There might be 200 different 10 ohm resistors, which according to the linear model are all identical (have the same value.) Why so many different components? Some of the variety is explained by similar components from different manufacturers, but there are also more fundamental reasons.

All real components deviate from the ideal in every possible way. Passive components are designed to be the best practical approximation of an ideal resistance, capacitance or inductance. For example, a resistor is designed to provide a fixed value of resistance with very little capacitance or inductance.

There are three main classes of non-ideality:

parasitic impedances
All components have frequency-dependent impedances, so all resistors have some capacitance and inductance. These undesired impedances are called parasitic impedances. In many cases, the components are good enough that designers can get away with using the ideal model and ignoring the effect of parasitics. Even when the parasitics are important, you can still use a linear model by pretending that the resistor is a combination of an ideal resistor, ideal capacitor and ideal inductor.
limited linearity
Real components are somewhat nonlinear. Their impedance varies depending on how much voltage you apply. (Some are designed to be highly nonlinear, but that's another story.) If you apply way too much voltage, the impedance will be irreversibly changed, with much production of smoke. Usually nonlineary can be ignored as long as you stay well away from the smoke-producing operating region.
inaccuracy and instability
The value of all real components is not precisely what it is supposed to be when the component is manufactured, and this inaccurate value keeps changing as the component ages and the temperature changes. Humidity and mechanical stress or vibration will also affect the value. Often ordinary components are good enough, but sometimes the accuracy and stability of a precision component is needed.

### Specifications: the Socially Constructed Component:

One of the most interesting aspects of components is their specifications: the manufacturer's numerical guarantee of how ideal the component is (spec sheet example). A 100 ohm 1% resistor is guaranteed to have an actual resistance of between 99 and 101 ohms at the time of manufacture. Additional specifications say that after exposure to heat, humidity, aging, etc., the value may shift by another 2.5% (96.5 to 103.5 ohms) Other specifications are absolute maximum ratings. If you continuously apply more than 7 volts (at 70 degrees C) then you exceed the power rating, and your warranty is void.

Manufacturers don't really guarantee that all of their components will meet all of their specifications all of the time, but if a particular component doesn't meet specifications, then it is said to have failed -- it's busted.

A key thing to realize here is that the specification is just a more general model, and not a complete description. Of course, at some microscopic level, each individual component is distinct, so the manufacturer must resort to statistical approximations. The only exact model of a thing is the thing itself, and this is a useless model because it has no generality. But what is more interesting from the designer's perspective is what the manufacturer chooses to specify.

### Missing Specifications

Often important characteristics are not specified. For example, the amount of power a resistor can safety handle is normally specified as a continuous rating. In our sample data sheet the resistor can soak up 1/2 watt all day long when the temperature is 70 degrees C or less. The physics tell us that when we absorb power in the resistor it is going to get hot, and the more power the hotter it gets. At some point the resistor burns up.

But what if the resistor will have to absorb 100 watts, but only for one millionth of a second? Our physical understanding says that the resistor can only heat up so fast, and the main problem is the final temperature. If the resistor has enough mass, the temperature will hardly change at all with a short pulse. If this pulse happens once a second, the average power is only 100 millionths of a watt, or 0.0001 watts. This is 1/5000 of the continuous power rating, so linear thinking would say this is extremely safe.

But linearity is limited in the real world... In fact, figure 2 in our datasheet tells us that one millionth of a second is o.k., but four millionths is a no-no. The shape of the curve in the datasheet has to do with specific details of the resistor materials and construction. The graph was probably constructed by experiment rather than theoretical analysis: zap it and see if it smokes.

For this discussion, the main point is that this table is useful, and that most resistor data sheets don't have this table. It is up to the manufacturer what they want to specify. Without a pulse power rating, you would have use a huge expensive 100 watt resistor or you would have to trust your physical intuition and experience that a much smaller resistor is o.k.

One of the oldest resistor designs is the carbon composition resistor. Although it is noisy, has poor accuracy, stability, linearity, and is bulky for a given continuous power rating, engineers found that it was much better at withstanding high-power zaps than many newer designs. Engineers came to trust the performance of carbon composition resistors, and have resisted manufacturer's attempts to phase them out. The interesting point is that these resistors were never rated for pulse power, and still are not.

### Negotiations

Of course, component manufacturers do try to give their customers what they want. One of they ways that manufacturers have responded to the pulse power issue is to introduce new components designed for pulse operation that have a completely new specification: energy. If the rating is 10 watt-seconds, then any pulse is o.k. as long as the energy is less than that: 10 watts for one second or 10,000 watts for 0.001 seconds.

This is a nice new model. It's probably just as much of an oversimplification as the constant power model, but people will buy it because it's easy to work with. They don't care if the component can actually stand higher pulse energies at particular pulse lengths.

### Typical vs. Worst-Case

One important issue in specification is typical vs. worst-case specifications. Especially with more complex active components such as integrated circuits (chips), manufacturers will give seperate typical and worst-case specifications for values. The worst-case spec is what we've been discussing up to now: if violated, the component is considered to have failed. A typical spec is a "normal" value that indicates what the component is capable of.

Why are there two different specs? The two main reasons are yield and testing costs. After manufacture, electronic compoponents are tested to see if they meet their specs. The fraction that passes the test is the yield Although 90% of the components made may meet the typical spec, there are many different specs, and the actual component values vary independently. It is likely any given component will fail at least one of the typical specs.

### The Cost of Testing

For some components, the typical specs are far far better than the worst-case specs. One example of this is leakage current specs for inputs to CMOS integrated circuits. For example, the CD4066 switch has a worst case leakage current of 0.1 millionths of an amp. That number is small enough that many people are able to design to that spec. But the true leakage current is far far smaller.

I've seen typical specs ranging from 0.001 millonths of an amp to 0.000001 millonths of an amp, from 100x to 10,000x better than the guaranteed spec. Furthermore, when I've tested these parts, the actual performance is significantly better even than these typical specs, in some cases 1000x better than the "typical" spec.

What is going on here? Basically we have a component that is uselessly close to ideal. The leakage current is so close to zero that detecting it requires specialized lab instruments, careful technique and a fair amount of time. So the manufacturer only tests that it is close enough to zero for typical use.

Sometimes engineers do need the true component performance, or at least a level of performance much higher than the tested worst-case performance. Then they can either pay a lot more for testing up to the needed spec, test it themselves, or just cross their fingers and use the component without any special testing.

Last update 25 June 2004

My homepage | Art links | email