DC Voltage Drop: Why the Regulator Is Fine and the Chip Is Not

The regulator output measures perfectly and the chip is still undervoltage. This is one of the most common supply faults on a new board, and it is almost never the regulator. The voltage the die actually receives is the source voltage minus every drop along the delivery path and every drop along the return path, and those two paths are made of copper, solder joints, connectors and vias rather than of ideal wire.

The useful mental shift is to stop treating the supply as a node and start treating it as a path. Once it is a path, each segment has a resistance, each source of copper resistance has a current through it, and the sum of the products is the error you are chasing. That sum is the dc voltage drop.

Where the Voltage Actually Goes

At DC and at low frequency, the contributions come from a short list. The connector contact resistance, the on resistance of any protection device in series, the copper of the traces and pours, and the resistance of the via structures that move the current between layers. Each of these is small in isolation and entirely reasonable in a spreadsheet, and together they can consume a meaningful part of the margin.

The load current scales the whole list. A path that drops twenty millivolts at half an amp drops a hundred at two and a half amps, and the load current on a modern board is rarely the number that appears in the datasheet summary. Even with the feedback point placed correctly, the remote load can be outside tolerance while the regulator believes it is doing its job, which is why the power distribution path has to be budgeted as a path rather than as a node.

A Measurement Reference That Hides the Fault

Probe the supply pin with the scope tip and clip the ground lead near the regulator, and you have measured the drop of the supply path against a ground that is not the chip local ground. Part of the error is now baked into the reference and will not appear on the screen.

The correct habit is to measure both the source and the load, each against the ground at its own location, and then subtract. For high current or high precision rails, a differential measurement or a Kelvin sense arrangement separates the path carrying current from the path used to sense voltage; which one is appropriate depends on the instrument and on the safety limits of the system.

Five Places the Drop Concentrates

Connectors and fuses come first. Contact resistance and spring contacts contribute directly, and they change with temperature as the contact warms. Copper necks come second: a wide pour that narrows sharply at a pad or an avoidance is a concentrated resistor, and it is usually invisible in a full board view. Via arrays come third, where the vertical path is limited to a small number of barrels. The common ground return comes fourth, where a heavy load shares a segment of return copper with a sensitive load and lifts that local reference.

The fifth is the one that surprises people: the segment everyone assumed was large. Copper area does not guarantee low resistance if the current has to pass through a narrow section to reach it, and slots or a plane split can force a long detour that the area calculation never accounted for.

Why Warm Boards Fail and Cold Boards Pass

The resistance of copper, of connectors and of several protection devices rises with temperature. A board that meets its limit three minutes after power up can fail after thirty, and the mechanism is a small loop: more drop means the load works differently, which raises the current or the dissipation, which raises the temperature further.

This is why a single measurement taken at cold start proves nothing. Repeat the measurement at a settled thermal state and at the worst case current, and confirm that the hottest region of the board matches the segment with the largest drop. When the hot spot and the worst drop coincide, the model is right and the fix is obvious.

Finding the Drop Segment by Segment

Fix the load current and the operating mode so that the baseline is repeatable. Record the regulator side and the chip pin voltage at the same time, and confirm the difference changes with current in the way the resistance estimate predicts. Then walk the supply path and measure across each element: connector, protection device, copper neck, via group. A test method that records each segment separately is worth more than a single reading at the regulator. Repeat the exercise along the ground return, because surveying only the positive rail is how the other half of the drop stays hidden. Finally, repeat it on a warm board and correlate the numbers with the thermal image and with the manufacturing tolerance of the parts involved.

Where the boards are made in volume, the tolerance matters as much as the nominal value. Contact resistance varies between connector batches, copper thickness varies within the process window, and a design that passes with the best case parts is not a design that passes. Building the drop budget with the worst case combination, and verifying it on assembled boards, is what turns a marginal rail into a specified one. That budget should be part of the design review and part of the acceptance criteria recorded for the assembly stage, not something reconstructed after a field return. Copper weight and its tolerance come from the fabrication window, and the contact resistance of the connector comes from the component supply chain, so both ends of the estimate belong in the same document.

Three Assumptions Worth Retiring

A large power plane is not automatically a low resistance path, because necks, splits and via groups still create local bottlenecks. Adding bulk capacitance does not remove a steady state drop, since capacitance supplies transient current and the direct current resistance remains. And thickening only the positive rail misses half the loop, because the return path carries the same current through the same kind of copper.

FAQ

Can a regulator with remote sense still fail this test? Yes. Remote sense corrects the drop it can see, and it cannot compensate for a path element outside the sense loop.

Is a milliohm of connector resistance significant? At ten amps it is ten millivolts, which may be most of the margin on a core rail.

Should the measurement be done with a multimeter or a scope? Either, provided both ends are measured against the correct local ground and the current is known.

Why does the fault disappear when the probe is moved? Because the measurement reference changed. That in itself is evidence that the return path is contributing.

Summary

When the regulator output is correct and the die is still starved, stop replacing the regulator. Walk the current path with a meter or a differential probe, find the segment with the largest drop, and remember that the ground return is a segment too. Converting the supply from a point into a path, and checking it under load and at temperature, is what makes a DC distribution problem a measurement instead of a guess.