Should Analog and Digital Grounds Be Split?

A converter datasheet shows two ground pins, AGND and DGND. A reasonable engineer reads those names, looks at a board that has analog circuitry on one side and a microcontroller on the other, and concludes that the board should have two ground planes joined somewhere sensible.

That conclusion is wrong often enough, and expensively enough, that it is worth taking apart carefully. The names on the pins are not instructions about your PCB.

Quick answer. Usually not. AGND and DGND are the converter's two internal grounds, brought out separately because the package leads cannot carry digital current without disturbing the analog section. They describe what happens inside the chip, not how to divide your board. For most converters the manufacturer's own guidance is to join the two pins and connect them to one continuous ground, then control where digital current flows by placement rather than by cutting copper.

What the two pins actually are

Inside a converter there is analog circuitry and digital circuitry on one die. Both need a ground reference. Ideally they would share one, but the bond wires and package leads have enough inductance that digital switching current flowing through a shared lead would develop a voltage across it, and that voltage would appear directly as an error at the analog section.

So the manufacturer brings out two pins. The digital ground current gets its own path to the outside world, and the analog reference stays clean.

Analog Devices puts the naming problem bluntly in its converter grounding material: calling the pin DGND is a category mistake waiting to happen. The pin is named for the current that flows through it inside the device, not for the system ground it should connect to. Their guidance is that both pins normally go to the same place, and that place is the analog ground.

That is the opposite of what the names suggest, which is why this keeps catching people out.

Why splitting the plane usually backfires

Suppose you take the names at face value and cut the plane into an analog region and a digital region, joined at one point near the converter.

Everything that crosses that boundary now has a problem. A digital trace running from the converter to the microcontroller sits over the digital side, crosses the moat, and continues over the analog side. Its return current cannot follow it across the gap, so the return runs along the split to the single joining point, crosses there, and comes back.

The loop you have created is not slightly larger. It is as large as the moat is long, and it encircles the analog section on its way around.

Henry Ott, whose work on mixed-signal layout is one of the standard references, treats the split-plane habit as one of the field's more damaging misconceptions, and recommends partitioning instead: one continuous plane, with the board divided into analog and digital areas by where you place things and where you route, not by where you cut copper. Texas Instruments reaches the same conclusion in its analog applications material, and adds the honest caveat: partitioning only works if you then have the discipline not to route digital signals through the analog area.

Partitioning, concretely

Partitioning is an arrangement discipline. It costs nothing in copper and everything in planning, which is why it has to happen at placement rather than after routing.

• Decide a line across the board. Analog on one side, digital on the other.
• Place the converter on that line, with its analog pins facing the analog

side and its digital pins facing the digital side.

• Keep the plane continuous underneath, across the whole board.
• Route analog signals only over the analog region and digital signals only

over the digital region.

• Keep the digital supply and its decoupling return loop entirely on the

digital side.

The return currents then separate themselves. Digital return current flows under digital traces, which are all on the digital side, so it never enters the analog region. You get the separation the split was supposed to give you, without the reference discontinuity that the split creates.

The converter orientation is the part people skip. A converter placed on the partition line but rotated so its digital bus faces the analog section guarantees digital current across the analog region no matter how carefully you route.

Where the sources genuinely disagree

It would be convenient to end there. The honest position is that two respectable traditions give different answers, and knowing which you are reading matters.

The partitioning position, associated with Ott and echoed by TI's applications material, is: one plane, divide by placement, never cut.

The converter-centred position, which appears in many datasheets and in some manufacturer application notes, is: separate analog and digital planes, joined underneath the converter package, with the converter bridging them.

These are not as far apart as they look. Both agree that digital return current must be kept out of the analog reference. Both agree the two converter pins are tied together. They differ on whether to enforce the separation with a gap in the copper or with discipline in placement and routing.

Two things decide which applies to you:

How much digital current the device draws. Analog Devices distinguishes ordinary converters from large devices, such as processors with integrated codecs, that pull hundreds of milliamps of transient current through their digital ground. For those, they do recommend connecting the digital ground pin to the system digital ground, because the device is internally designed for that case. For a converter drawing a few tens of milliamps, they recommend the standard scheme instead.

What the manufacturer's own layout guidance says for that part. A datasheet and its evaluation board layout are statements about one specific die, package and internal architecture. When the manufacturer's layout guidance for your part is specific, follow it, and follow it completely rather than taking the plane split and leaving out the placement that went with it. The advice Analog Devices warns against is the generic line "connect AGND to analog ground and DGND to digital ground" when it appears without any device-specific reasoning behind it.

If the datasheet asks for something and you have a reason to deviate, the burden of proof is on you, and the way to discharge it is to trace the return currents rather than to argue from principle.

What to do when you have to decide

A practical order, for a board with one or two converters and a microcontroller:

1. Start from one continuous ground plane. This is the default and it is right

more often than any other default.

1. Partition by placement. Analog in one region, digital in another, converter

on the boundary and correctly oriented.

1. Read the converter's layout section properly, including the evaluation

board. Do what it says for that part.

1. If you end up with a split because the manufacturer asked for one, make

sure no signal crosses it anywhere, and that the join is where they said.

1. Check your work by following the digital return current and confirming it

never has a reason to enter the analog region.

Step 5 is the one that actually validates the design. Steps 1 to 4 are how you get to something worth checking.

What does not help

• Splitting the plane and then routing the SPI bus across the split.
• A ferrite bead or a zero-ohm link bridging two ground regions carrying

high-frequency return current. It is a deliberate inductance in a return path.

• Separate planes for analog and digital on a two-layer board, where there was

not enough continuous reference to begin with.

• Copying the plane split from a reference design without also copying the

placement, the stackup and the part.

What comes next

Lesson 5 takes on the other grounding topology that gets applied outside its range: star grounding, where it genuinely solves the problem it was invented for, and what it does to a fast digital board.

Sources and further reading

• James Bryant, Analog Devices: Grounding Analog-Digital Converters — why AGND and DGND exist, why the names mislead, the standard scheme and the large-DSP exception.
• Analog Devices MT-031: Grounding Data Converters and Solving the Mystery of AGND and DGND — the same material in tutorial form, with layout figures.
• Texas Instruments Analog Applications Journal: mixed-signal grounding and partitioning (SLYT512) — single plane with partitioning, and the routing discipline it requires.
• Autodesk summary of Henry Ott's mixed-signal partitioning guidance — the partitioning method and why splitting enlarges return loops.