For as far back as most practicing engineers can remember, the speed and density of digital logic has just about doubled every three years. Between 1974 (the first year of introduction for the 8080) and 1997 the speed of operation for typical microprocessors has rocketed up from 200 KHz to 200 MHz. It's an industry-wide trend, and it's unstoppable: customers demand speed.
If the speed of operation has changed by that much, you would think that the role of the bypass capacitor would have changed dramatically, too. It hasn't. This lowly component is still used in the same basic way, for the same basic purpose, as it was twenty-three years ago. It's time for a change.
Why change is necessary
As logic speeds have spiraled upward, the frequency content of a typical digital signal has moved up into the gigahertz range. The performance of bypass components, which are expected to safeguard the power system against signal-induced fluctuations, must now be extended up into the same range. Yesterday's design rules, adequate for use at low speeds, are not a good match for today's screaming-fast logic.
The primary symptoms of an inadequate, old-fashioned bypass capacitor array are increased power supply noise, increased crosstalk among signal traces, and increased electro-magnetic radiation. Power supply bypassing is a serious matter that merits serious attention. Digital design shops that have not reviewed their bypass capacitor design rules in several years would do well to take a close look at the issue.
Electrical Performance Of Bypass Capacitors
The subject of bypass capacitor layout has ramifications for both electrical and mechanical design. The primary electrical design issue has to do with what is called the parasitic series inductance of the bypass component. The parasitic series inductance of a bypass component acts like a little inductor wired in (guess what) series with the component. At higher and higher frequencies, the impedance of this little parasitic inductance becomes larger and larger, until it dominates the performance of the component.
In the critical 100-1000 MHz band, the effectiveness of a typical bypass capacitor is determined almost entirely by its series inductance. This is the frequency band now being used increasingly by digital logic (see footnote). For good performance we want low series inductance. The series inductance of a bypass component is determined almost entirely by the layout of the capacitor's mounting pads and its associated vias. Every time we double the logic edge rates, we become twice as dependent on these layout details.
Series inductance is impacted unfavorably by any of the following:
- Long traces (> 0.01 inch) between the capacitor pads and vias
- Capacitor mounting arrangements that stand the capacitor body up away from the board,
- Capacitors with any sort of mounting configuration other than surface-mount pads, and
- Skinny via holes (less than 0.035 inch diameter).
On the other hand, series inductance is impacted favorably by:
- Surface-mount configurations using wide, squat mounting pads with vias jammed up next to the pads (no traces),
- Great big via holes, and
- Thin boards (less than 0.030 inch thick) that bring power and ground planes right up near the body of the capacitor.
Series inductance is a real and measurable effect. On a bare board (bypass capacitors only), an RF engineer can use a network analyzer to make a plot of the power-to-ground impedance versus frequency for the assembled bypass capacitor array. This is a good measurement to take. The impedance plot will clearly show a region around 1 MHz where the impedance, as a function of frequency, is going down. This is the region where the bypass capacitors are doing their best job.
The plot will then show another region around 100 MHz where the impedance, as a function of frequency, has hit bottom, turned around, and begun heading back up. That is the region dominated by series inductance (which is determined by layout).
At the highest frequencies (1000 MHz and above) the impedance tops out, turns around, and heads back down again. This is the region dominated by the natural capacitance between the power and ground planes (if you have them). This inter-plane effect is, for most practical purposes, an ideal capacitance.
If a finished product suffers from too much noise in the 100 to 1000 MHz region, the best, most effective way to reduce noise in that region is to improve the layout of the bypass capacitor pads and vias. Improving their layout reduces the effective inductance of the parts, leading to a direct reduction in power and ground noise. If you can't improve the layout, try these other ideas:
Add more bypass capacitors to the board (this costs money and takes up space for parts).
- Change to a more effective style of part. For example, look for the new side-ways mounted 0612-style (as opposed to 1206-style) surface-mounted capacitors, or the broad, flat packages offered by Circuit Components, Inc., or the new AVX brand LICA style capacitors. Some of these parts offer an effective inductance 20 times as effective as a single 1206-style part.
- Reduce the logic switching currents by reducing the loads on each gate, or reducing the number of simultaneously switching gates.
If none of the above approaches work, you can try using a higher-dielectric-constant material between the power and ground planes (Zycon is one brand name that comes to mind), or stacking the power and ground planes closer together. These last alternatives tend to influence the power-to-ground impedance mostly at the upper part of the range, from 500 MHz up.
Good bypass design plays a critical role in the control of power and ground noise, crosstalk, and electromagnetic radiation. If you are having problems with these issues in the 100 to 1000 MHz band, take a look at the effective series inductance of your bypass capacitors. The problem may lie there. If your designs are working well today, and bypassing is not a problem, don't get complacent. Progressive advances in the speed of digital logic will cause problems soon enough. Continual adjustment of bypass capacitor placement and routing rules is the only safe course of action.
The frequency content of a digital signal extends mostly from DC up to a frequency equal to one-half divided by the 10-90% rise or fall time of the signal. For example, a digital signal with 1-ns rise/fall time has a frequency content that extends up to 500 MHz. The frequency content is a function of the edge speed, not the clock rate. Forget about harmonic analysis. We are talking here about real, digital, non-repetitive signals.