LVDS Routing Rules for High Speed Differential Pairs
LVDS, or low voltage differential signalling, is a low swing differential technique that carries data at several hundred megabits per second over a printed pair or a balanced cable. Its small voltage swing and low drive current give it low noise and low power, and those same properties are what make the routing rules matter. A single ended signal tolerates a certain amount of abuse; a differential pair carrying a few hundred millivolts of swing does not.
Why LVDS Needs Its Own Rules
Two things distinguish an LVDS link from an ordinary digital trace. The first is that it is a transmission line, so the signal arrives as a wave and anything that changes the impedance along the way reflects part of the energy back to the source. The second is that the receiver recovers the data from the difference between the two conductors, which means that noise coupled equally onto both is rejected while noise that arrives on one conductor alone is not.
Those two properties set the agenda. The routing has to hold a constant impedance, which means a solid reference plane and a controlled geometry. The pair has to stay closely coupled so that external fields affect both conductors similarly, which is what gives the common mode rejection its value. And the two conductors have to remain the same electrical length, because a phase difference between them turns into a differential signal that the receiver sees as noise rather than as data.
Stack-Up and Shielding
A multilayer construction is the normal starting point. The layers immediately above and below the pair should be ground, so that the pair is shielded from other signals and has a reference plane close to it. On a board that is not densely packed, routing the LVDS pairs on their own layer and keeping other signals elsewhere is the cleanest arrangement, and it removes a class of coupling that no amount of spacing on a shared layer can fully control.
Controlled Impedance and Tight Coupling
Every differential standard specifies an impedance, and LVDS is no exception: a point to point link is normally routed as a hundred ohm differential pair, with the exact figure taken from the driver and receiver datasheets rather than assumed. The geometry that produces that impedance, the trace width and the gap between the two conductors, is calculated from the stack-up and entered into the constraint manager before routing starts, so that the router cannot produce a pair that looks correct and measures wrong.
Tight coupling is the second requirement, and it follows from the first principle of a differential link. When the two conductors are close together and carry equal and opposite currents, their magnetic fields largely cancel and the pair radiates far less than two independent traces would. Coupling also narrows the impedance and improves the pair’s immunity to a field arriving from outside. The practical rule is that the two conductors stay parallel and at a constant gap for the whole run, with no via or foreign signal inserted between them.
Length Matching and Phase
Because the receiver takes the difference between the two conductors, a skew between them appears directly as a loss of margin. The two traces are therefore matched in length to a tolerance that depends on the data rate, and the matching is best done by adjusting the shorter trace with small serpentine detours near the point where the mismatch occurs rather than by adding length in one lump at the end. The techniques are the same ones used to match any high speed bus, and are set out in serpentine routing and length matching.
The matching has to survive the whole path, not just the routed section. The package lead length inside the driver and receiver, the pin delay of the device, and any connector or cable in the path all contribute, and a pair that is matched on the board but asymmetric in its connector will still arrive with skew.
Spacing Between Pairs and Other Signals
Coupling between two differential pairs is not cancelled the way coupling inside a pair is, so adjacent pairs have to be separated. A working rule is that the gap between pairs should be at least three to five times the gap within a pair, and where the routing is dense a line of ground vias between them gives better isolation than distance alone, following the same reasoning as the 3W rule for crosstalk.
The same logic applies to single ended signals. LVDS pairs and TTL or other single ended nets are best kept on different layers. Where they must share a layer, the separation should again be at least three to five times the intra pair spacing, because a single ended aggressor couples into the pair unequally and the imbalance converts directly into differential noise.
Plane Splits and Layer Changes
A differential pair should never cross a split in its reference plane. The argument that the two conductors are each other’s return path is not sufficient, because the reference plane is still what defines the impedance of the pair, and a pair that runs over a void has a different impedance over that section than over the rest of the run. The result is a reflection at each end of the discontinuity, which is exactly what the controlled impedance work was meant to prevent.
Where a pair has to change layers, the return path has to change with it. Adding ground vias beside the signal vias gives the return current a short path between the two reference planes rather than forcing it to find its own way around, and keeping the two conductors symmetric through the transition keeps their lengths equal. A layer change that moves one conductor further than the other introduces skew even when the routed lengths match on each layer. Whether the pair runs as a microstrip on the surface or as a stripline between planes also changes its impedance and its loss, and the trade-offs are set out in microstrip and stripline routing.
Termination and Unused Pins
The termination resistor sits at the receiver, and its placement is part of the routing. It should be as close to the receiving pins as the layout allows, with the connection between the resistor and the pins kept short, because any length beyond the resistor is an unmatched stub that reflects. Its value is chosen to match the differential impedance of the pair, commonly in the range of about 90 to 130 ohm, with 100 ohm the usual figure for a point to point link, and its tolerance matters because it sets the differential voltage the receiver actually sees.
Unused pins are part of the same discipline. Unused receiver inputs are left open, unused outputs are left open, and unused driver inputs and enable pins are tied to a supply rail or to ground rather than allowed to float, because a floating input can sit at an indeterminate level and draw current or oscillate.
FAQ
What impedance is an LVDS pair routed to? A point to point link is normally routed as a 100 ohm differential pair, with the figure taken from the driver and receiver datasheets. The trace width and gap that produce it are calculated from the stack-up and entered as a routing constraint.
Why must a pair be tightly coupled? Because closely spaced conductors carrying equal and opposite currents cancel most of their external magnetic field, which reduces radiated emissions and improves immunity to an external field.
Can a differential pair cross a plane split? No. The pair still needs its reference plane to define its impedance, and crossing a split creates an impedance discontinuity and a reflection.
How close should the termination resistor be to the receiver? As close as the layout allows, with a short connection. Any trace beyond the resistor is an unmatched stub that reflects energy back into the link.
