Electronics Guide

Articles in This Category

Why Partition Across Multiple Boards

Designers choose multi-board architectures for reasons that extend well beyond raw board area. The decision balances technical capability against cost, time to market, and the realities of the product life cycle.

Scalability and Configurability

Separating functions onto distinct boards lets a single platform scale by populating more slots or swapping cards. A chassis can ship with a baseline configuration and grow as demand increases, and customers can mix line cards, compute modules, or I/O cards to match a specific application. This composability is impractical on a monolithic board, where every option must be designed in and paid for up front.

Manufacturability and Yield

Large, dense boards are expensive to fabricate and assemble, and a single defect can scrap an entire panel. Partitioning into smaller boards improves manufacturing yield, simplifies test, and allows the highest-layer-count or most exotic materials to be confined to the boards that actually require them. A high-speed switch card may use ultra-low-loss laminate, while a slower I/O card uses economical FR-4.

Serviceability and Upgrade

Field-replaceable modules let operators repair or upgrade a system without discarding the whole assembly. Hot-swappable cards support maintenance without taking the system offline, a requirement in carrier and data center equipment. Modularity also extends a product's life: a faster fabric card or a new interface module can be introduced while the chassis, backplane, and power infrastructure remain unchanged.

Heterogeneous Technology and Reuse

Different functions favor different process technologies, voltages, and thermal environments. Isolating an RF front end, a high-voltage power stage, or a sensitive analog measurement chain onto its own board contains noise and eases compliance. Proven modules can also be reused across product families, amortizing engineering effort.

Interconnection Architectures

The defining characteristic of a multi-board system is how its boards exchange signals and power. Several architectures dominate, often combined within one product.

Backplane and Midplane Systems

A backplane is a passive or active board into which daughtercards plug through high-density connectors, providing both data interconnection and power distribution. A midplane variant accepts cards from both front and rear, improving airflow and enabling separate service and I/O domains. Modern backplanes route dedicated differential pairs between communicating cards rather than shared multi-drop buses, eliminating stub reflections and cumulative loading. See Backplane Architecture for an in-depth treatment.

Mezzanine and Stacking

Mezzanine cards mount parallel to a carrier board through board-to-board connectors, adding function in a compact vertical stack. Established standards define the mechanical and electrical interface: the FPGA Mezzanine Card (ANSI/VITA 57.1) offers low-pin-count and high-pin-count variants on a 1.27 mm array connector, and FMC+ (ANSI/VITA 57.4) extends the transceiver count and reaches roughly 28 Gbps per lane. The older PCI Mezzanine Card (IEEE 1386.1) and its switched-fabric successor XMC (ANSI/VITA 42) remain common in embedded and rugged systems. Board-to-board stacking is space efficient, and the short interconnect keeps loss low, but it constrains the interface to the connector pin count and to a fixed stack height, which in turn fixes component clearance on both boards.

Cable and Flex Interconnects

Where boards cannot share a rigid connector plane, cables and flexible circuits carry signals between them. High-speed twinaxial cable assemblies increasingly bypass lossy backplane traces by routing differential pairs directly from card to card, while rigid-flex construction integrates flexible sections into an otherwise rigid stack. These options trade mechanical freedom against added connectors, controlled-impedance routing, and shielding requirements.

Optical and Wireless Links

For the highest data rates or the longest reaches, electrical interconnection gives way to optical links, and for mobile or rotating subsystems, to wireless connections. These media sidestep the loss and crosstalk limits of copper but introduce their own integration concerns, addressed under Distributed Systems.

Signal Integrity at Board Boundaries

Every transition between boards is a discontinuity in the signal path. A high-speed channel that crosses two boards may traverse a transmit package, a connector, a length of backplane or cable, a second connector, and a receive package, with vias at each board entry and exit. Managing these transitions is the central signal integrity task of multi-board design.

Impedance Continuity and Reflections

Connectors and vias rarely match the surrounding transmission-line impedance exactly, and each mismatch reflects energy back toward the source. Designers optimize connector footprints, back-drill unused via stubs, and tune launch geometry to minimize these reflections. The cumulative effect of several small discontinuities can close an eye that each individual feature would leave open.

Insertion Loss and Equalization

Long inter-board channels accumulate substantial frequency-dependent loss from dielectric absorption, conductor skin effect, and copper surface roughness. Standard FR-4, with a dissipation factor near 0.02, becomes unsuitable for backplane-length channels much above a few gigahertz, so high-speed designs specify low-loss laminates in the range of roughly 0.005 to 0.010 and ultra-low-loss materials at 0.002 or below. Serializer/deserializer equalization then recovers what the channel degrades: transmit pre-emphasis or feed-forward equalization, continuous-time linear equalization at the receiver input, and decision-feedback equalization to cancel post-cursor reflections and loss tails. Together these techniques sustain 25 Gbps per lane with non-return-to-zero signaling and 56 and 112 Gbps per lane with four-level pulse-amplitude modulation (PAM-4) across realistic multi-board channels. The next generation is in standardization: IEEE 802.3dj specifies 200 Gbps per electrical lane, and PCI Express 7.0 raises the per-lane rate to 128 GT/s, both using PAM-4.

Crosstalk and Return Paths

Densely packed connector pins couple energy between adjacent channels, and a signal that changes reference planes at a board boundary needs a continuous return path to follow it. Differential signaling, ground pins interleaved among signal pins, and careful via and connector assignment keep coupling and return-path discontinuities within budget. Related coverage appears under Crosstalk and Coupling and Differential Signaling.

Clocking and Synchronization Across Boards

Distributing a timing reference across boards is harder than distributing one on a single board. The clock traverses the same connectors, cables, and vias as the data, accumulating jitter and skew at every discontinuity, and a receiver on one card must recover data launched by a transmitter on another whose timing may differ in both frequency and phase.

Clocking Strategies

Three approaches dominate. A common reference fanned out from the backplane holds every card to one frequency, at the cost of routing a low-jitter clock to every slot and matching its length. Embedded clocking, used by essentially all modern serial links, lets each receiver extract timing from the data stream through a clock-and-data-recovery loop, which removes the distributed clock entirely but imposes encoding overhead and a jitter-tolerance requirement on the receiver. Independent clocking gives each card its own oscillator and absorbs the residual frequency offset in the protocol layer with elastic buffers and periodic skip ordered sets, as PCI Express does in its separate-reference-clock modes. Related material appears under Reference Clock Architecture and Jitter and Timing.

Systems That Must Align Events

Some systems need more than working data links: phased-array radar, distributed data acquisition, and coordinated motion control must align events across physically separate boards. These designs add explicit low-jitter reference distribution and a time-transfer protocol such as IEEE 1588 Precision Time Protocol, and they treat skew as a system budget spanning connectors, cables, and board-to-board length differences rather than a property of any one board's routing.

Power Distribution Across Assemblies

Multi-board systems must deliver clean, stable power to every card while limiting the noise that one board injects into another. Distribution strategy shapes both efficiency and signal integrity.

Most modern systems use a distributed, or intermediate-bus, architecture in which a relatively high voltage feeds across the backplane and on-card regulators step it down at the point of load. Because resistive loss rises with the square of current, quadrupling the distribution voltage cuts that loss by a factor of sixteen for the same delivered power; the approach also isolates each card's switching noise behind its own regulator. Telecommunications equipment has long distributed −48 V DC, and data center hardware has migrated from 12 V to 48 V racks, the level adopted in the Open Compute Project's Open Rack V3. High-power computing racks are pushing further still, with ±400 V and 800 V DC distribution introduced to move hundreds of kilowatts without impractical busbar currents, though 48 V remains the volume standard. Power-distribution-network design then manages DC resistance to limit voltage drop, plane and decoupling capacitance for high-frequency response, and isolation between noisy and sensitive domains, as detailed under Power Distribution Networks.

Power sequencing adds a system-level dimension absent from single-board designs. Boards energize in a defined order so that interfaces are not driven before their supplies are valid, and hot-swap controllers limit inrush current as cards are inserted into a live backplane.

Thermal and Mechanical Considerations

Partitioning a system across boards distributes heat sources throughout a chassis and creates mechanical interfaces that must survive insertion, vibration, and thermal cycling.

Thermal Management

Card spacing, airflow direction, and the placement of high-power components determine whether every board stays within its temperature limits. Midplane designs admit air through cutouts to support front-to-back cooling, and conduction-cooled card formats move heat to the chassis rails for sealed or high-shock environments. The interconnect infrastructure itself dissipates power in resistive losses, which thickened copper or dedicated power bars help carry. See Thermal Effects on Signal Integrity for how temperature influences electrical behavior.

Mechanical Robustness

Backplanes and carriers must resist flexure during card insertion, support the assembled weight, and absorb connector mating forces. Those forces accumulate with contact count: ANSI/VITA 46 caps the mating force of a VPX connector at 0.75 N per contact, so a fully populated slot with several hundred contacts still requires several hundred newtons to seat, which is why large cards use injector or ejector handles rather than hand pressure alone. Card guides, alignment pins, and connector keying ensure correct insertion and prevent mechanical damage or hazardous mis-mating, while strain relief protects solder joints and contacts from ongoing stress. Vibration and thermal cycling then act on those same interfaces over the product life, making contact retention and fretting corrosion long-term reliability concerns rather than assembly-time ones.

Standards and Interoperability

Standards-based architectures let cards from different vendors operate together, simplifying sourcing and reducing development risk through proven reference designs. Several families dominate multi-board electronics:

• PICMG: CompactPCI, AdvancedTCA (PICMG 3.0), and MicroTCA define form factors and electrical interfaces for telecommunications and embedded computing. AdvancedTCA organizes its backplane into Zone 1 for power and shelf management, Zone 2 for the base and fabric data interfaces, and Zone 3 for user-defined rear I/O, typically to a rear transition module. Each fabric channel carries eight differential pairs, four in each direction, arranged in dual-star or full-mesh topologies; the companion PICMG 3.1 specification has grown from 1 and 10 Gigabit Ethernet to 40GBASE-KR4 and 100GBASE-KR4 signaling.
• VITA: VME, VPX (ANSI/VITA 46), and OpenVPX (ANSI/VITA 65) serve rugged military, aerospace, and industrial applications. OpenVPX defines interoperable backplane and module profiles, slot types (including payload, peripheral, switch, storage, and bridge), and interconnect topologies built on the VPX connector.
• OCP: The Open Compute Project publishes open hardware specifications for hyperscale data centers. Open Rack defines the rack envelope and its power busbar, the OCP Accelerator Module and its universal baseboard standardize multi-accelerator carriers, and the Data Center Modular Hardware System family standardizes how server functions are partitioned across a host processor module and its peripheral boards.
• Serial link specifications: IEEE 802.3 defines the backplane and copper-cable Ethernet physical layers, the -KR and -KR4 variants, that multi-board fabrics run over, and PCI-SIG defines PCI Express. These documents, rather than the mechanical form factor, set the loss, crosstalk, and jitter budgets a card-to-card channel must meet, and they are what a compliance test actually measures.

Standardized interfaces also discipline modular design: when module boundaries align with a published specification, integration becomes a matter of compliance rather than custom negotiation. The trade-offs of that discipline are explored under Modular Architecture.

Modeling and Verifying Multi-Board Channels

A multi-board channel cannot be validated one board at a time. The transmitter package, the launch via, the source board's traces, its connector, the backplane or cable, the destination connector and vias, and the receiver package combine into a single electrical path, and the interactions between those elements, particularly reflections resonating between successive discontinuities, appear in no individual model.

The usual practice is to characterize each element as a set of scattering parameters, taken from three-dimensional field solvers for passive structures and from measurement for connectors and cable assemblies, then cascade them into an end-to-end channel model. That cascaded model must be checked for passivity, causality, and reciprocity before it is trusted, because a physically inconsistent block yields results that look plausible and are wrong. Behavioral transmitter and receiver models, usually in IBIS-AMI form, then drive statistical link analysis, which predicts bit error rates far below what a time-domain simulation of a few million bits could resolve. See Modeling and Simulation and Statistical Signal Integrity.

Measurement closes the loop. Vector network analyzer sweeps of a fabricated channel, with test fixtures de-embedded, confirm or refute the model, and time-domain reflectometry locates the discontinuity responsible when the two disagree. The results are expressed as a link budget that allocates loss, crosstalk, jitter, and noise among the boards, connectors, and cables, so that each team owns a defined share of the total. Those methods are developed under Link Budget Analysis and Measurement and Test.

Bringing the System Together

A multi-board system is only as reliable as its integration. Once individual boards are validated, the assembled system must be verified as a whole: signal integrity across real connectors and channels, power sequencing in the correct order, thermal behavior under load, and the electromagnetic compatibility of the complete chassis. Emissions in particular are a system property rather than a board property: cables between boards act as efficient antennas for common-mode currents, and enclosure seams and ventilation apertures determine how much of that energy escapes, which is why grounding architecture must be settled before the boards are laid out. Field experience and maintenance strategy then feed back into the next design iteration. This system-level work is the subject of Multi-Board System Integration.

Multi-board design is ultimately an exercise in managing interfaces. The boards that perform a system's functions are well understood in isolation; the engineering value lies in connecting them so that signals arrive intact, power stays clean, heat escapes, and the assembly endures its service environment. The four topics listed above develop each of these dimensions in depth.

Related Topics