HOW-TO GUIDE
How to Identify Ground Bounce Using an Oscilloscope
By the OhmSavvy Research Team ·
Safety: the probe ground is mains earth
- On a mains-powered bench scope, every probe's ground clip or spring is connected to mains earth. Connect grounds only to the circuit's ground. Touching one to a supply rail or a signal shorts that point to earth.
- Ground bounce work is normally done on low-voltage logic boards. If the board's ground might be tied to a non-isolated mains supply, do not connect a ground clip to it, and never float the scope by removing its earth.
What ground bounce is
Inside a chip, the output transistors connect to the board's ground through bond wires and package leads, and those have a small inductance. When an output switches from high to low, it discharges the load capacitance on its pin, and that current flows out through the ground pin. When many outputs switch at the same moment, such as a data bus or a bank of GPIO pins, the currents add up. The fast change in current across the ground inductance lifts the chip's internal ground above the board's ground for a few nanoseconds.
01
Outputs switch
Several fall high to low together
02
Current surge
Load charge flows to the ground pin
03
L di/dt
Voltage across package inductance
04
Ground lifts
Internal ground rises briefly
05
Quiet pins move
Low outputs spike; inputs mis-read
Every output and input on that chip is referenced to the internal ground, so the bounce appears on outputs that are supposed to be steady, and it shifts the thresholds of the chip's inputs. The result is a false edge on a neighboring line, often a clock, reset or chip select, that has nothing to do with the signals that caused it.
Estimate it before you measure: V = L di/dt
The formula
This is a simplified estimate. The real waveform is a short damped ringing pulse, not a flat step, and the inductance below is an illustrative figure: a few nanohenries is a commonly quoted order of magnitude for a leaded package's ground path, but it depends on the package and how many ground pins it has. The point is how the result scales.
| Case | L | Current change | Time | di/dt | V = L di/dt |
|---|---|---|---|---|---|
| One output switching | 5 nH | 50 mA | 2 ns | 25 mA/ns | 5 × 25 = 125 mV |
| Several outputs at once | 5 nH | 200 mA | 2 ns | 100 mA/ns | 5 × 100 = 500 mV |
| Same, slower edges | 5 nH | 200 mA | 5 ns | 40 mA/ns | 5 × 40 = 200 mV |
| Same, two ground pins | 2.5 nH | 200 mA | 2 ns | 100 mA/ns | 2.5 × 100 = 250 mV |
Half a volt on a 3.3 V system is a large share of the noise margin between a valid low and the input-low threshold, which is why ground bounce causes real faults.
1Pick a quiet output to watch
The best window into the chip's internal ground is an output on the same chip that is held low while its neighbors switch. Because a low output is connected to the internal ground through its output transistor, it follows the bounce. If you write the firmware, set a spare pin as an output driven low; on a bus, pick a bit that stays at 0 while others toggle. Probe it as close to the chip as you can, not at the far end of a long trace.
2Set up the probe with a short ground spring
This is the step that decides whether the measurement means anything. The spikes you are looking for last a few nanoseconds, and the long ground clip lead on a standard probe rings at a similar speed on its own. If you are unsure how the probe, ground lead and spring attach, start with how to connect an oscilloscope probe correctly.
- Probe on 10x, channel ratio set to 10x, and probe compensated.
- Short ground spring to a board ground point right beside the chip, ideally a ground pin or a ground via next to it.
- DC coupling, bandwidth limit off, and about 200 mV/div on the quiet channel.
It is also worth knowing your own baseline first. See how to measure noise floor so you know what the scope adds on its own.
3Trigger on the switching edge
Put a second probe on one of the outputs that is switching and use it as the trigger source, falling edge, level at mid-swing. Ground bounce is mostly tied to outputs falling from high to low, so start with the falling edge, then check the rising edge too. Use a fast timebase, around 10 ns/div, with the edge near the left of the screen. Normal trigger mode keeps the display from rolling between bursts. See how to trigger an oscilloscope if the trace will not settle.
4Read the spike on the quiet line
If ground bounce is present, the quiet line shows a sharp spike, usually followed by a little ringing, that starts at the moment the switching output falls. Measure its peak against the low level with voltage cursors.
Then run the confirming test: change how many outputs switch together. If the firmware can toggle one pin, then four, then eight at once, real ground bounce grows roughly in step with the number switching, as the worked example predicts. A spike that stays the same size whatever the count is coming from somewhere else.
Rule out probe ringing and other look-alikes
Several things produce spikes at switching edges. Work through them before you change the board. The most common false alarm is the probe itself; the full method for separating probe ringing from circuit ringing is in how to measure overshoot and ringing, and general trace noise causes are in why an oscilloscope waveform looks noisy.
| Cause | What gives it away | Test |
|---|---|---|
| Ground bounce | Spikes on all quiet outputs of one chip, at others' edges | Grows with the number of outputs switching |
| Probe ground-lead ringing | Ringing whose frequency shifts when the lead moves | Swap to the ground spring; it shrinks or vanishes |
| Crosstalk | Spike only on lines routed next to the switching one | Compare a quiet pin far from the switching traces |
| Radiated pickup | Spike with the probe tip touching ground | Short the tip to the ground spring and look again |
| Supply ripple or noise | Slower, periodic, not tied to the logic edges | Trigger on the logic edge; ripple will not lock |
The tip-to-ground test is quick and telling: with the probe tip touching its own ground spring at the chip, any spike that remains is being picked up by the probe loop, not measured on the pin.
Fixes for ground bounce
Every fix reduces either L or di/dt. Re-measure after each change with the same probe setup, so you are comparing like with like.
- Decoupling close to the pins. A small capacitor between each supply pin and ground, as close as possible, supplies the switching current locally and shortens the loop it flows around.
- A solid ground plane and more ground vias. Every parallel path lowers the inductance, as the two-ground-pin row shows.
- Slower edges. Many microcontrollers have a GPIO slew-rate or drive-strength setting; small series resistors at the outputs are another common approach. Check the datasheet for what your part supports.
- Fewer simultaneous switches. Stagger outputs in firmware, or reduce the capacitive load each one drives.
Which scope you need for ground bounce
You need two channels, one on the quiet line and one on the switching output, and as much bandwidth as you can get. A scope's own rise time is roughly 0.35 ÷ bandwidth, about 3.5 ns at 100MHz and 7 ns at 50MHz, so spikes only a few nanoseconds wide are smoothed and shown smaller than they are. Treat the measured height as a lower limit; see what oscilloscope bandwidth means. A 1-channel handheld cannot do this job. Compare bench scopes in our best beginner oscilloscopes roundup, and for probing technique on logic lines in general, see how to measure a logic signal with an oscilloscope probe.
Bench scopes in our reviews
100MHz, 4 channels, 1 GSa/s; more bandwidth than the Rigol, for projects that will get more demanding.
100MHz, 2 channels, 1 GSa/s; an oscilloscope and waveform generator in one unit.
50MHz, 4 channels, 1 GSa/s; the default beginner recommendation for years.
As an Amazon Associate I earn from qualifying purchases. Prices and availability are shown on Amazon.
Frequently asked questions
What is the difference between ground bounce and crosstalk?
Crosstalk couples from one trace or wire to a neighbor through capacitance and inductance between them, so it depends on how close the two run. Ground bounce comes from the shared return path inside the chip, so it affects every output of that chip at once, including ones routed far from the switching lines. Probing two quiet outputs on opposite sides of the package helps separate them.
Does ground bounce also affect the power supply pin?
Yes. The same effect happens in the supply connection, where it is often called supply bounce or VCC droop. Ground bounce is usually linked to outputs switching from high to low, which pushes current out through the ground pin, while supply droop is linked to outputs switching from low to high. A quiet output held high shows the supply side.
Can ground bounce damage a chip?
It is mainly a signal integrity problem rather than a damage problem. The usual symptom is false triggering: a spike crosses a logic threshold on a clock, reset or chip select line and the receiver acts on it. Very large negative excursions can exceed an input's absolute maximum rating, which is listed in the datasheet, so large bounce is still worth fixing.
Will a 20MHz bandwidth limit help me see ground bounce?
No. The bandwidth limit filters out exactly the fast spikes you are looking for, so they look smaller or vanish. Leave it off for this measurement. It is useful for measuring ripple on a supply rail, where the fast spikes are not what you want to quantify, but not for finding bounce.
Why does ground bounce get worse with faster chips?
Faster output stages change their current in less time, so di/dt rises, and the voltage across the ground inductance rises with it. Moving to a newer, faster logic family on the same board can therefore create bounce problems that the older, slower parts never had, even though the supply voltage and the loads are the same.
