So, here we go again with the typical "what's the best ground layout?" question. Seems this question has different answers depending on the application, manufacturer, and decade.

From what I've read, the trend nowadays is to use a single ground plane for everything. Avoid star grounds, ground plane is where it's at. Dedicate a complete layer to ground and add plenty of vias, especially for decoupling capacitors. Do not split the ground plane

This article, however, recommends making a slit to split the ground plane around the precision analog sections.

Regarding Analog VDD, the same article seems to suggest the DVDD in certain ICs is not really a DVDD in the same sense a AVDD to DVDD in a MCU is. An ADC, for example, doesn't need to be high speed, its digital pins don't need to drive large capacitive loads, and don't drive high currents. They can also be slew rate limited by source termination resistors. All of this makes the DVDD line on an ADC much quieter than a MCU driving 50mA on each pin, all of them switching at very high speed.

So I'm assuming, for the particular case of an ADC, a ground plane containing both digital and analog grounds may be "good enough". Then again, the part I'm using, a HI7190, has an evaluation kit with a rather complex layout, consisting of 4 layers with split ground and power planes. But this layout dates from 1994, back from when split planes were all the rage.

This particular ADC uses about 1 to 3mA for each power rail (AVdd, DVdd, AVss), so I don't see a reason why it should need different power planes, as there is no high current draw, especially if using source termination on the SPI lines.

It will be, however, connected to a rather noisy ESP8266 microcontroller. The ESP, as most other RF chips, is known to draw a lot of current (peaks in excess of 200mA) from the supplies. In this case, should the ground plane be split? Or, since the ESP8266 micros come usually in a PCB module, it will be connected to the main board with just one GND point, which will effectively "split" the ground plane.

  • 3
    \$\begingroup\$ I perceive the design of Ground Planes as far less "controversial", "time-changing" and "opinionated" then you seem to do: You've got a circuit that deals with RF reception. OK, there's little question whether you can put that on the same ground as high-speed digital. You can't; there's pretty simple noise calculations. "All the rage" isn't an engineering reason; actual engineers apply the theory they've learned to design something that fulfills the requirement they understand employing the skills they've acquired. "All the rage" engineering is only done when people aren't actually competent. \$\endgroup\$ – Marcus Müller Jan 2 '18 at 14:34
  • \$\begingroup\$ @MarcusMüller see the recommended questions at the right of this page. In the answers sections you get contradictions. What's the right answer then? Stackexchange-style "the one with the most votes"?. Re: RF. I specifically didn't mention RF as RF is a completely different issue. To me, RF is black magic I can't even begin to understand. I admire people who design RF circuits. \$\endgroup\$ – hjf Jan 2 '18 at 14:47
  • 1
    \$\begingroup\$ Can you ask a specific question, without all of the grandiose editorial about your perception of "trends" and "rages"? This sounds more like voodoo than engineering. \$\endgroup\$ – Dave Tweed Jan 2 '18 at 15:18
  • \$\begingroup\$ @DaveTweed should I split, partition, "make a slit around" a "precision" device's ground, or should I just connect everything to a single, solid ground on a PCB layer? \$\endgroup\$ – hjf Jan 2 '18 at 15:32

It's wrong to 'split' a ground plane, just because you hear it's the right thing to do. Similarly, star grounds, ground planes, gridded grounds, are all wrong, if you're doing it 'because somebody said it's the current fashion'.

Inappropriate ground connections can get you into trouble in various ways.

With high speed logic, you want your transmission lines between ICs to have the proper impedance, which means an adjacent ground. This is easiest to achieve with a ground plane. If there's a few mV of noise in ground connections and logic swings of 100s of mV or more, that's no problem.

With accurate analogue circuit, the many micro-ohms resistance of a plane, if it has a current flowing through it, can develop many microvolts of error signal at the input to an opamp. Here is where you should think about where the current paths are, where the voltages are sensed from. Star grounding can help organise thoughts, but if care is used in the layout, a ground plane can be got to work. If it can't, a judicious split in the plane to force currents to take another route can rescue the layout.

Unfortunately, there's no alternative to understanding what's required, understanding where the currents are going to flow, and keeping the error signals out of where they shouldn't be.

| improve this answer | |
  • \$\begingroup\$ This is one of the things where I don't understand the "whole ground plane" vs "split ground plane" vs "partitioned plane". To me, it seems the "partitioned" approach makes the most sense. In this approach, current loops are confined in their own partitions. But this seems to go against recommendations from people who claim "there is no reason to do this, unless you have a very good reason". \$\endgroup\$ – hjf Jan 2 '18 at 14:45
  • 1
    \$\begingroup\$ Unfortunately, you have to understand it. There is no magic bullet. A single solid ground gets very few people into trouble. If you do get into trouble, then fighting your way out of it is a very illuminating experience. Or to put it another way, good decisions come from experience, experience comes from bad decisions. You have to put in the hours, the designs, before you get a feel for what's right. \$\endgroup\$ – Neil_UK Jan 2 '18 at 15:42
  • \$\begingroup\$ @Neil_UK is right about this, 95% of the time a solid plane works just fine, 5% of the time you are doing low frequency or precision single ended measurement and the solid plane is possibly the wrong choice (Actually, single ended measurements are usually the wrong choice...). All engineering is the art of the compromise there is seldom one right answer, and even once you have that feel for whats right, you still sometimes get a 'learning experience'. \$\endgroup\$ – Dan Mills Jan 3 '18 at 13:55

Not the answer you're looking for? Browse other questions tagged or ask your own question.