100 MHz vs 1 GHz: High Speed Interfaces on the PCB

An interface running at a hundred megahertz and one running at a gigahertz are separated by a factor of ten in clock rate, and by much more than that in layout difficulty. The difference is not simply that high speed interfaces need tighter rules. It is that the dominant loss and distortion mechanisms change, the margin available for error shrinks, and the assumptions that make a slower design work stop being valid.

What Actually Changes Between the Two

The first change is that the electrical length of the route grows relative to the signal. A hundred megahertz link may complete several round trips within one bit period on a small board, which means reflections settle before the receiver samples. At a gigahertz the same physical route is a significant fraction of a bit period, and every reflection arrives inside the window that matters.

The second change is loss. Dielectric loss rises with frequency, and conductor loss rises with the square root of frequency, so a channel that is essentially transparent at a hundred megahertz can consume a large part of the signal amplitude at a gigahertz. The third change is that everything that was negligible becomes measurable: via stubs, connector transitions, package parasitics and copper roughness.

Return Path and Reference Plane Discipline

Both designs need a continuous reference, but only the faster one is punished severely for losing it. A split in the plane under a hundred megahertz trace adds a small amount of inductance to the return path and usually goes unnoticed. Under a gigahertz signal it becomes a radiating loop, and the resulting crosstalk and emissions appear as unexplained failures in testing.

Layer transitions follow the same logic. A return via beside every signal via is good practice at any speed and mandatory at the higher one, because the inductance of the long return path becomes a significant fraction of the total path at the shorter wavelength. Our notes on microstrip and stripline routing describe how each structure behaves and why the return path is easier to control on an inner layer.

Impedance Control and Its Tolerance

The faster interface needs a controlled impedance, and it also needs that impedance to be consistent along the whole route. A hundred megahertz link may tolerate a step change at a connector without consequence, while the same step at a gigahertz creates a reflection that lands inside the sampling window. Consistency, not just the nominal value, is what the design has to deliver.

The achievable tolerance is set by the fabricator and by the stackup, which is why the impedance requirement has to be agreed before layout rather than checked afterwards. Our notes on impedance tolerance explain how the specification, the coupon and the finished board relate to one another.

Via Stubs, Transitions and Layer Count

A via stub that is electrically short behaves as a small capacitance and is harmless. As frequency rises, the same stub approaches a resonant length and produces a notch in the channel response. The faster design therefore either uses thinner boards, avoids long stubs by placing the routing layers adjacent to each other, or specifies back drilling.

The layer stackup often follows from this constraint rather than from routing density. If the fast signals can be routed on layers close to the surface, the stubs stay short and the design becomes simpler. Pushing them to an inner layer to gain routing space may cost more in signal integrity than it gains in convenience.

Material Selection Starts to Matter

At a hundred megahertz, standard FR-4 is adequate for almost any digital link. At a gigahertz the dielectric loss of the laminate becomes a significant part of the channel budget, and the variation of dielectric constant with frequency and temperature begins to affect impedance and delay. The material choice moves from a cost consideration to an electrical one.

Where the design sits between the two, the decision should be made from the loss budget rather than from habit. Our discussion of PCB dielectric constant describes how the laminate properties enter the calculation, and why a material that is adequate at one frequency may not be at another.

Connectors, Cables and the Channel Beyond the Board

The faster link also forces attention outside the board. A connector that was adequate at a hundred megahertz may have a stub or a pin geometry that creates a resonance in the gigahertz band, and a cable that worked before may now contribute a significant part of the total loss. The channel budget has to include everything between the two chips, not only the copper on the board.

This is why high speed designs are often evaluated with measured channel data rather than with calculation alone. Simulation is useful for ranking options, but the connector and cable models have to be credible for the result to mean anything.

Design Effort Follows the Constraint

The practical consequence is that effort should be allocated by constraint rather than applied uniformly. On a mixed design, the hundred megahertz interfaces can be routed with ordinary discipline while the gigahertz links receive controlled impedance, matched lengths, short stubs and continuous references. Treating every net as critical wastes time; treating the fast ones casually wastes a prototype.

That allocation is a floorplanning decision, and our notes on high speed PCB layout floorplanning describe how to identify which nets belong in which category before routing begins, rather than discovering it during validation.

Test and measurement also change character. A hundred megahertz link can be verified with an ordinary oscilloscope and a passive probe, while a gigahertz link needs a high bandwidth instrument, a low inductance probe tip and an understanding of what the fixture itself contributes to the measurement. Engineers who have only worked at the lower speed are often surprised by how much of the measured degradation comes from the way the measurement was made rather than from the board. Building a short reference structure on a spare panel, measuring it and comparing the result with simulation is the quickest way to establish how much the setup can be trusted.

FAQ

Is a 100 MHz interface ever a high speed problem? It can be, if the edge rate is fast and the route is long. The relevant comparison is the rise time against the propagation delay, not the clock frequency. A slow clock on a long backplane can behave like a transmission line, while a faster clock on a short local route may not.

Do I need a special laminate for a 1 GHz link? Not automatically, but the loss budget has to be calculated rather than assumed. Standard FR-4 handles short gigahertz links comfortably; longer routes, tighter margins or higher frequencies push the design toward lower loss materials, and the decision should follow the numbers.

How much does a via stub matter at 1 GHz? It depends on the length of the stub and the dielectric constant of the laminate. As a guide, keep the stub shorter than roughly one tenth of the wavelength in the dielectric at the highest frequency of interest. Beyond that, back drilling or a different layer assignment is worth the cost. The measurement that reveals the problem is usually a channel response plot rather than a single impedance number.