# Why are capacitors sold with imbalanced tolerances?

Short version: Some capacitors (and presumably some other components) are sold with imbalanced/asymmetrical tolerances. Why?

Explanation:

Many ceramic capacitors are marked with, for example, +80% -20% tolerance or something similarly imbalanced.

For example, let's say that I have a capacitor with the (admittedly contrived) value of 17pF and a tolerance of +80%, -20%.

(Please ignore abuse of significant figures.)

• Maximum value: 17pF * (1 + 80%) = 17pF * 1.8 ≈ 30.6pF
• Minimum value: 17pF * (1 - 20%) = 17pF * 0.8 ≈ 13.6pF
• Mean value: (30.6pF + 13.6pF) / 2 ≈ 22.1pF
• Tolerance above mean: (30.6pF - 22.1pF)/22.1pF ≈ +38.5%
• Tolerance below mean: (13.6pF - 22.1pF)/22.1pF ≈ -38.5%

It would be fair to say that this supposedly "17pF" capacitor is virtually identical to a 22pF capacitor with ±40% tolerance.

By a similar process, a 10000pF +80% -20% capacitor (a real value from a catalog, not contrived) should be the same as a 13000pF around ±40%.

So, if I say I want a component of a given value, why am I being sold something that's quite a bit more likely to overshoot than undershoot this value? Is this imbalance useful to anyone?

• Your statistics are only correct if the distribution is symmetrical around the mean. My guess is that the value they advertise is the mean, but the distribution is asymmetric hence the asymmetric tolerances. I do not know why capacitors would be this way, so I will leave it up to someone who does to answer. – Kellenjb Jan 17 '12 at 17:26
• Can you point us to an example part? In circuits where capacitors are used for DC filtering, bigger is better. When you spec the capacitor, you care mostly about the minimum capacitance. – markrages Jan 17 '12 at 17:48
• I've never seen that kind of asymmetric spec on pF-range capacitors that would be used in a tuned circuit. – markrages Jan 17 '12 at 17:48
• I think that can be a conservative value (better an exceeding value than an insufficient value) as markrages say; or could be depending by the fabrication process, in which some parameter scale exponentially instead of linearly. But the latter is unlikely. – clabacchio Jan 17 '12 at 17:54

## 3 Answers

Unlike resistors, whose price is essentially independent of resistance except at extreme values which represent less than 0.01% of the market, most types of capacitors have a cost which is tied strongly to capacitance--it costs more to make a large cap than a small one. Further, capacitors are often used in circumstances were a cap which is larger than specified might work better than the specified one, up to a certain limit, but the bigger cap might not be worth a higher price.

Suppose a designer determines that a device needs a minimum of 8uF to work correctly in a particular situation, but anything up to 20uF would work just as well. Some manufacturers can produce devices within +/-20% of their target; other manufacturers are capable of +/-33% of their target. If published tolerances were symmetrical, one would have to specify that the part could be either a 10uF+/-20% or a 12uF+/-33%--a bit awkward. If, however, manufacturers by convention use -20% for the lower tolerance and adjust the upper tolerance as needed, then it's possible to directly compare and substitute parts with different tolerances without affecting circuit operation.

You mostly see the -20% +80% tolerance for ceramics with Y in their name. These ceramics have a good energy density, but are "sloppy" in that the final capacitance varies with temperature, applied voltage, and have significant manufacturing variations.

These types of capacitors makes sense in high volume applications since they are a bit cheaper to manufacture than those with other ceramics and tighter tolerances. Their main use is for bypass caps and secondary filtering on power supplies. In these applications the circuit may rely on some minimum capacitance, but lots more causes no trouble. Manufacturers know this and therefore spec these capacitors more for their guaranteed minimum value as apposed to the most likely center value.

Unless you have a high volume application where the small extra savings for the Y type ceramics makes a difference, I would just stay away from them.

For tolerance, the rating is the allowed variation from the nominal value. As Supercat points out, this is typically more useful if it doesn't vary so much on the negative side, since for many applications (e.g. bulk capacitance) you usually don't mind if the capacitance increases significantly, but significant reduction could cause problems.

To contrast the tolerance to temp coefficient, note that a EIA tolerance rating of Z is -20%, +80%. This is the opposite asymmetry to a V temp coefficient rating of +22%, -82%.

For temp coefficient:

I think the figure -20%, +80% given for tolerance means the maximum change of capacitance in the rated temperature range.
If we look at a typical Y5V dielectric (from one of Vishays datasheets) you can see the curve is not symmetrical around 25 dec C (which is usually where the marked value comes from) This would be rated something like +20,-70.

Here's another graph for an aluminium electrolytic, with a different (but still asymmetrical curve - Probably rated +10, -20):

It seems the tolerance codes are whatever the actual tolerance tested will "fit into" (i.e. the maximum allowable change) so for example the +20, -70 would probably be given the V code (+22, -82) as it is guaranteed to be inside this rating (hence Y5V)

• While it's worthwhile to be aware of the asymmetry you mention, I would think that the asymmetry of customer requirements is a larger factor. If a circuit would work with a resistor anywhere from 1K to 100K, a typical engineer might drop in a 10K, since it would cost the same as a 1.2K or a 86K. On the other hand, if a circuit needs a cap between 1uF and 100uF, and if a 10uF cap would cost more than a 1.5uF cap, the engineer would more likely use the 1.5uF cap. – supercat Jan 17 '12 at 21:25
• @supercat - yes I agree, since the "tolerance" is not the same as "temp coefficient", which for some reason I thought the question was asking about. The answer to that one is pretty much as you say, the variation from nominal tolerance is typically more useful if it's above the nominal value than below for non critical uses. I will edit to clarify this. – Oli Glaser Jan 17 '12 at 23:00