A bypass-capacitor dialogue peels back the layers, Part 2: The theory of ground relativity

Editor's introduction: Bypass capacitors, grounding, and decoupling are relatively low-visibility, low-glamour issues and generally not the subject of many feature stories, but are vital to a successful, reliable, error-free design. Our multipart series on the subject was the most popular articles by far we presented over the past 12 months.

Authors David Ritter and Tamara Schmitz of Intersil engaged in a further dialogue on the subject. Here's Part 2 of the conversation. (For Part 1 , click here). Dave and Tamara believe in the value of arguing, the value of education, and getting to the heart of the problem without ego; in short, ripping apart a problem for the sake of knowledge. “Listen in” and learn:

Tamara : But let's be specific. What do you mean by a “good ground” or “good ground plane”?

David : Well, ground is supposed to be 0 V.

Tamara : But is it ever really 0 V?

David : No, of course not. There is always some impedance, always some current causing a voltage drop.

Tamara : So ground at one point is never the same as another point.

David : Sometimes we work on isolated problems where we can assume a local ground will be relatively consistent. On the other hand, some applications are in a high-RF environment, near a transmitter or microwave oven, for example. These folks have huge signals coupling into their grounds.

Tamara : So how do we construct a “good ground”? Should our readers just use a ground plane?

David : Sometimes the answer is yes.

Tamara : But often times, there is enough current in the ground plane to cause significant voltage drops between one spot and another.

David : So the question is: how do you ground individual circuits in a system to optimize performance?

Tamara : It depends on the type of circuit.

David : Yes, and you may use multiple grounding schemes within one system.

Tamara : Still, all grounds need to hook back to the same place eventually.

David : Yes. But do we take the individual grounds and connect them directly to one place?

Tamara : We could, and that would be called a star ground (a very popular and–when used correctly–successful way of grounding a system.)

David : I've used that technique for small circuits, but we need to look at the bigger picture, too.

Tamara : When you get to large-area circuits, the problems are tougher. You can't have a useful bypass capacitor connected over half an inch away from a part. All of that trace inductance will degrade the performance of the capacitor.

David : I like to view ground as a local phenomenon. Follow the supply and input currents through the small local loops around a single chip (for instance) and keep those loops as small and tight as possible. The grounds from the local circuits then connect to a larger ground system, which we design by following the larger scale currents.

Tamara : Can you give me an example?

David : Sure. (Example 1) We were building a two-input video oscilloscope (called a 'Waveform Monitor'). Here's a simplified schematic of the front end (Figure 1 ):

Figure 1: Two-channel selectable front-end circuit schematic
(Click on image to enlarge)

Tamara : That's a couple of video amplifiers feeding into a 2:1 mux, with another buffer on the output, right?

David ::Exactly. We built a board that looked like this (Figure 2 ):

Figure 2: Two-channel selectable front-end layout
(Click on image to enlarge)

David : This is a four-layer board, although two of the layers are hardly used (light blue and navy blue). Red is the top layer and the last layer is a ground plane.

Tamara : Looks like a very straightforward design and clean layout.

David : Turned out that there was too much coupling between the inputs. (Even when Input B was turned off, it showed an attenuated version of Input A.)

Tamara : How big was the signal on Input B?

David : Anything larger than -90 dB was unacceptable in our video applications. We were measuring a signal about -55 dB.

Tamara : I'd like to take a better look at the fringe currents that caused the coupling (Figure 3 ):

Figure 3: Front-end layout showing fringe currents
(Click on image to enlarge)

Tamara : I see. The fringes of the current flow from Input A are causing a signal in Input B.

David : That's what we guessed. Anytime the fringe current paths overlapped, we got some crosstalk. To test this theory, we put a few cuts in the board, Figure 4 (green lines are cuts):

Figure 4: Two-channel selectable front-end layout with ground slices (in green)
(Click on image to enlarge)

David : Amazingly, the coupling went away–it actually dropped below the noise level.

Tamara : So why did that work?

David : Follow the currents. There is always a loop for current, there are always fringe currents inside the ground plane, and voltage drops associated with those currents. The cuts prevent the currents from mingling so they could no longer couple directly from Input A to Input B (or the reverse).

Tamara : So this was an example of a ground plane gone wrong. The ground plane with cuts in it actually performed better.

David : Yeah, most people believe that providing a low-impedance connection to ground (like a plane) is enough. Sometimes that is true, but it's too one sided. If you need really high isolation, you need to follow and control the currents around the entire loop.

Tamara : (Example 2) If you do have a single chip, though, then a ground plane is a great idea. Let's use one of the input amps from the layout you've already shown. Figure 5 shows how I would hook up the 0.1 µF bypass capacitor to the ground plane.

Figure 5: Operational amplifier layout with vias to ground plane
(Click on image to enlarge)

David : The via to ground adds a small inductance in series with the bypass connector, as does the trace. The ground connection of the chip (if it has one, this op amp doesn't) often has a via connecting it to ground, so the loop has twice the via inductance.

Tamara : The two resistors on the left (input termination and gain resistor) also connect to ground. It would be great to give this chip a good local ground for all 4 connections.

David : We often improve the situation by adding a trace below the chip (still on the top layer). This provides a direct and tight loop for the current from the power supply to return to ground. In this case, though, there is no space below the chip to connect the left pair of ground vias to the right pair, so I would lower the impedance of the ground by adding pads on layer one. (Figure 6 )

Figure 6: Operational amplifier layout with vias to ground plane and top-layer pads
(Click on image to enlarge)

Tamara : What about trace thickness? Thick traces give low-impedance connections. It seems that you would want to keep all of the power and ground traces as wide as possible. Would you ever want to use a thin trace?

David : Sure. We are trying to keep the impedances around the local loop very small, but we would actually like to have a higher impedance back to the supply. A thin trace actually adds some decoupling from the main supply lines, helping to isolate one chip from the rest of the system.

Tamara : Show me on Figure 2.

David : The high-frequency path is right around the chip. The light blue traces are the thick, low-impedance power lines. The thinner traces connect each chip to those power lines. High-frequency signals will remain local instead of traveling back into the supply.

Tamara : Wouldn't you normally use a ferrite bead to decouple high frequencies from the rest of the supply?

David : Sure, if you have the board space and budget for one. Otherwise, using a thin trace is a crude, but effective alternative.

Tamara : One last question before we wrap up: People ask me: do you believe in multi-point or star grounds?; it's like it is a religious question. Do they ask you the same thing?

David : Often. But I'm not sure how to answer because I use a mixture of techniques. You need to design the ground system using all of you ingenuity. It's not a matter of belief.

Tamara : I agree. Like we've discussed here, it's about following the currents!

(to be continued. . . . )