Backplane Design: Long Traces, Many Slots and Heavy Power

A backplane is the board that carries the connectors for every card in a system. It usually has no active components of its own, it is physically large, and it is the most demanding board in the chassis in terms of layer count, thickness and manufacturing tolerance. It also has a failure mode that matters more than on any other board: when a backplane fails, the whole system stops, and replacing it means stripping the chassis.

What a Backplane Is

The term covers several related arrangements, and it is worth separating them because the design rules differ.

A passive backplane carries connectors and routing only, with no active devices. All the processing power lives on the plug-in cards, which keeps the backplane simple, replaceable and independent of the technology in the cards.

A motherboard carries the connectors as well, but it also carries the processor and the main supporting circuits. It is a different product with a different repair strategy.

A midplane places the connectors in the middle of the chassis, so cards plug in from both sides. It halves the stub lengths for some topologies and gives a more compact mechanical arrangement, at the cost of a more complex assembly.

In all three cases, the backplane work is the same in principle: connect a set of cards to each other and to a power source, with enough signal integrity to carry the fastest interface in the system, in a mechanical format that survives being plugged and unplugged.

Why It Is Difficult

Four properties make a backplane harder than a normal board, and they interact.

It is large. Trace lengths are measured in tens of centimetres, which means loss and pattern dependent jitter accumulate long before the signal reaches the far card.

It is thick. A backplane can be 3 mm or more, so a through via leaves a stub that is significant at multi gigabit data rates, and the aspect ratio of the holes challenges the plating process.

It has many layers. Twenty or more layers is common, because the connectors are dense and the power distribution is heavy, and every layer adds to the cost and to the lamination complexity.

It has repeatable interfaces. The connectors are pressed or plugged into place, and the mechanical tolerance of the holes is far tighter than on an ordinary board, because a press fit connection depends on the hole diameter being right.

Connectors and Mechanical Interface

The connectors define the backplane more than any electrical consideration, because the mechanical interface is what the whole chassis is built around.

Press fit connectors are the standard for a large backplane. Each contact is a compliant pin that is pressed into a plated through hole without solder, which removes an entire soldering operation and allows a connector to be replaced. The connection depends on the hole diameter being held to a tolerance of a few hundredths of a millimetre, and on the plating being uniform through the barrel. Our article on press fit covers the hole tolerance and the tooling that goes with it.

Because the connectors are pressed in, the backplane has to be stiff enough not to flex during insertion. That usually means a thicker board, and sometimes a stiffener or a metal frame, since a board that bows will not accept a card cleanly and can damage the contacts.

The mechanical design also sets the routing. Card positions, keying, guide rails, the insertion force of each connector and the space between slots are all fixed by the chassis, and the backplane layout works inside those constraints. Where the system needs hot swap capability, the connector has to provide staged pin lengths so that ground and power connect before the signals, which is a connector selection rather than a layout decision but one that changes the pin assignment in the schematic.

Long Traces and Stubs

Two electrical problems dominate, and they pull in opposite directions.

Loss is proportional to trace length, and a backplane route is long by definition. At multi gigabit rates the attenuation of a long FR-4 trace is the limiting factor in the link budget, and the answers are a lower loss material, a wider trace, a thicker dielectric, or a shorter path through a better topology. The material choice usually has the largest effect, which is why backplanes are among the few boards where a premium laminate is routine rather than exceptional.

The stub is the second problem. A signal enters a card connector, travels through the board to the corresponding connector on another card, and the portion of the via below the connected layer is unused. On a 3 mm board that stub is several millimetres long, and at high frequency it behaves as a resonant transmission line that removes energy from the signal at a specific frequency. Removing it is the purpose of back drilling, which is almost universal on high speed backplanes.

Blind and buried vias solve the same problem by construction, and they achieve better electrical performance, at the cost of sequential lamination. On a very large board the trade is usually decided by the total number of lamination cycles against the loss budget for the fastest interface in the system.

Power Distribution

A backplane feeds every card in the chassis, so the current it carries is the sum of what all of them draw, and the voltage drop it can tolerate is much smaller than the voltage itself.

The first design decision is the copper weight and the layer allocation for the power and ground planes. A system drawing several hundred amps at a low voltage needs thick copper and often more than one layer per rail, which is one of the main drivers of the layer count. Where the current is very high, bus bars bolted to the backplane supplement the copper, and a high current design of this kind needs the current path modelled rather than estimated.

The second decision is the distribution topology. A single feed at one edge and a long run to the far end produces a voltage gradient across the board, so the cards in the middle see a different supply from the cards at the end. Feeding the rail from both ends halves the gradient, and a distributed plane with multiple feeds is better still.

The third is decoupling, which is different on a backplane from a normal board. There are few active devices and a great deal of capacitance already present in the planes, so the requirement is dominated by the bulk capacitance needed to supply transient load steps from the cards and by the impedance of the plane itself at the frequencies the cards draw. The connections from the cards to the plane, through the connector pins, are part of that impedance and are often the limiting element.

Crosstalk and Topology

Two routing questions have a large effect on the achievable data rate.

Adjacent slots share the same edge of the backplane, so connector pins that carry fast signals sit close to pins from a neighbouring card. Coupling between them is the limiting factor on a densely populated backplane, and the remedies are spacing between high speed pins in the connector, ground pins assigned between signal pins, and routing on inner layers with reference planes on both sides.

The second question is the bus topology. A multi drop bus that connects several cards to the same set of traces creates stubs at every connector it passes, which limits the achievable rate and makes the bus unsuitable above a modest frequency. Point to point connections from a switch card to each slot avoid the stubs and dominate modern high speed designs, and the switch card becomes the single point that defines the system data rate. A daisy chain topology falls between the two and is used where the protocol tolerates it.

Whichever is used, the routing on the backplane is where the topology becomes real, because a point to point design that is routed with taps has become a multi drop bus with extra layers. The topology should be fixed before the layer assignment is made.

Manufacturing a Backplane

Everything about a backplane is at the edge of the process, which is why the supplier list is short and the lead time is long.

Panel utilisation. A large board may fit only a few times on a panel, so a small change to the outline can have a large effect on price. This is the same economics as any board, amplified by the size.

Thickness and aspect ratio. A thick board with small holes stretches the plating process, and the aspect ratio limit is what constrains how small the vias can be. Where the design needs small vias in a thick board, the answer is usually a sequential lamination rather than a tighter drilling process.

Drilling accuracy. The number of holes on a backplane is large, and the registration tolerance has to hold across the whole panel. Drill wander accumulates over a large area, which is why the annular ring allowance on a backplane is usually more generous than on a small board.

Press fit hole tolerance. The hole diameter for a compliant pin is specified to a few hundredths of a millimetre, and the plating thickness adds to it. That is a capability question for the specific fabricator, and it should be confirmed against the connector manufacturer requirement before the drawing is written.

Flatness. A large thick board has to stay flat enough for connectors to align with a card cage and for cards to be inserted. Lamination symmetry and copper balance matter more here than anywhere else.

Verification

Testing a backplane follows the same pattern as a normal board, scaled to the consequences of a failure.

Electrical test on every board verifies continuity and isolation across a large number of nets, and it is essential because a backplane cannot be reworked easily once connectors are pressed in. Impedance coupons measure the controlled impedance layers, and a TDR measurement verifies the characteristic impedance and reveals the stub resonances that a coupon alone will not show. Microsectioning confirms the plating thickness and the press fit hole diameter on the first article, and the press fit process itself is qualified by measuring the insertion force on a sample of holes before the production connectors are installed.

Two checks are specific to a backplane. A flatness measurement of the finished board, since the connector alignment depends on it. And a signal integrity measurement on a representative route, including the connector and the card interfaces, because the backplane is only one part of a channel whose performance is measured end to end.

Design Rules

• Fix the bus topology before assigning layers, so the routing does not turn a point to point design into a multi drop bus.
• Choose the laminate for the loss budget of the fastest interface, and confirm it with the fabricator early.
• Remove via stubs by back drilling or by blind and buried vias where the data rate requires it.
• Allocate enough copper and enough layers to the power and ground distribution, and feed the rails from more than one edge.
• Maintain a consistent reference plane under every high speed route, and stitch the references at every layer change.
• Space the press fit holes according to the connector specification, and confirm the finished hole tolerance with the fabricator.
• Budget the annular ring for the drilling registration across a large panel rather than for the tolerance of a small one.
• Reserve space for stiffening, mounting and card guides that the mechanical design requires.

FAQ

• How many layers does a backplane need? Twenty or more is common in a high performance system, driven mainly by the number of controlled impedance routes and by the power distribution.
• Do backplanes have components? A passive backplane does not. A midplane or an active backplane may carry power conversion, management or a switch device.
• Why are press fit connectors used instead of soldered ones? They remove a soldering operation, allow replacement, and are more reliable than a soldered joint on a thick board with a high aspect ratio hole.
• Can back drilling be omitted? Only where the stub is short compared with the rise time of the fastest signal. On a thick board at high data rates it is essential.

Summary

A backplane connects the cards of a system through a mechanical interface that was fixed long before the layout began. Its design is dominated by long lossy traces, via stubs in a thick board, tight press fit hole tolerances and a power distribution that carries the current of every card at once.

The rules that matter are the ones that follow from those constraints: choose the material for the loss budget, remove the stubs, allocate copper for the current, keep a continuous reference, and fix the topology before the stackup. Because the board is difficult to manufacture and difficult to replace, the capability discussion with the fabricator belongs at the start of the project rather than at the quotation stage.