SMARC vs Qseven vs OSM vs COM Express: Computer-on-Module Standards Compared

SMARC is one of five open standards for computer-on-modules: small boards that carry a processor, memory and storage, and plug or solder into a carrier board designed for the product. The others are OSM, Qseven, COM Express and COM-HPC. Each fixes a footprint, a connector and a pinout, so a module from one supplier can replace a module from another on the same carrier, and each suits a different class of product, from a battery-powered handheld to an edge server.

Two bodies look after them. SGET (Standardization Group for Embedded Technologies) publishes OSM, SMARC and Qseven, which are aimed mainly at Arm and low-power processors. PICMG publishes COM Express and COM-HPC, which are aimed mainly at Intel and AMD x86 processors with more power and more PCIe. Choosing between them early matters, because the standard sets the carrier board, the thermal design and much of the software work for the life of the product.

This guide is the reference view: why the module and carrier split exists, a master comparison table, each standard in depth, carrier board design, a choosing guide by application, migration paths, lifecycle and a checklist. For the general idea of a module, start with our explainer on the system-on-module. For more on the two PICMG standards, see COM Express module types and our comparison of COM-HPC and COM Express.

A modern processor is hard to design around. Memory routing, power sequencing and high-speed serial links need many PCB layers, simulation and a long validation cycle, and the processor itself may only be on sale for a decade or so. A computer-on-module puts that difficult part on a small, finished board from a module maker: the system on chip, soldered memory, boot storage such as eMMC and the power circuits it needs. The product team designs a carrier board instead, a simpler board that turns the module’s signals into the connectors, power input, expansion and mechanical fit the product needs.

The split brings three benefits that matter in industrial and embedded systems. The carrier can stay the same while the module is upgraded or replaced at end of life. One carrier can take a range of modules, from a low-power processor to a high-performance one. And because the standards are open, a carrier is not tied to one supplier. The cost is an extra interface between two boards, some height and a second board to buy, which is why a single board computer is often simpler for one fixed configuration.

A short history

Computer-on-modules began with proprietary formats and early open ones such as ETX around 2000. PICMG ratified COM Express in 2005 and COM-HPC in 2021. SGET was founded in 2012 and approved SMARC 1.0 that year, took over Qseven from its original consortium in 2013, and published its first OSM specification in 2020. NVIDIA Jetson modules work on the same principle but are a proprietary family, with a connector and pinout that NVIDIA defines for each generation.

The table is the quick reference. It shows the maximum each specification defines; what a particular module brings out depends on its processor, so always check the module datasheet.

Read across a row and the pattern is clear. The SGET standards trade bandwidth for size and power, which suits Arm processors with integrated graphics, camera pipelines and NPUs. The PICMG standards trade size for PCIe lanes, networking and power, which suits x86 processors and add-in accelerators. Qseven sits between them as the older SGET format that covers both architectures.

OSM (Open Standard Module) is the newest SGET standard and the only one with no connector. The module is a small board with a grid of contacts underneath, placed on the carrier by a pick-and-place machine and reflow soldered like a large chip. SGET describes the contacts as a ball grid array and several module makers describe the format as LGA; either way the module becomes a permanent part of the carrier. The specification was written for small, low-cost, high-volume products and covers microcontroller-class, Arm and x86 processors.

Across the sizes, OSM also defines up to five Ethernet interfaces, up to two CAN buses, four UARTs, UFS flash and up to 39 GPIO, plus an 18-pin communication area for antenna or fieldbus signals on every size. OSM 1.2, released in November 2024, stays compatible with modules built to 1.1. It adds single-channel LVDS on Size-S, a second camera interface on Size-M and a second I2S audio interface on Size-0.

Size-L is now big enough for an AI-capable processor. Advantech’s AOM-2721, announced in 2024, puts a Qualcomm QCS6490 with a 6.9 W TDP and 12 TOPS of AI performance on a 45 × 45 mm OSM module, with PCIe Gen 3, LPDDR5 memory, UFS storage, USB 3.2 and DSI, DisplayPort and eDP outputs. Advantech lists support for Windows IoT Enterprise LTSC on it, which shows how far Arm modules have moved from Linux-only designs.

Where OSM fits

Soldering removes the connector cost and height, makes the joint robust under shock and vibration, and suits automated assembly at volume. The trade-offs are just as real. A module cannot be swapped on the bench or in the field without rework, hidden joints usually need X-ray inspection, and moving to another supplier’s module means a manufacturing change and requalification. OSM suits products built in volume where the module is chosen once for each product version: handhelds, smart sensors, IoT gateways and compact HMIs.

SMARC (Smart Mobility Architecture) was approved by SGET in 2012 for low-power processors from the tablet and mobile world, and it has become one of the most common module standards for Arm. A SMARC module is a board with 314 gold edge fingers that slide into a low-profile, right-angle MXM 3.0 connector on the carrier at 0.5 mm pitch. Two sizes are defined: 82 × 50 mm for compact, low-power designs and 82 × 80 mm for processors that need more room and cooling. SGET puts typical module power under 6 W, which is why many SMARC products run without a fan.

SMARC 2.1, 2.1.1 and 2.2

SMARC 2.0, published in 2016, set today’s pinout. SMARC 2.1 added PCIe Gen 3, extended the USB 3 signal definitions and formalised MIPI CSI-2 camera lanes on the edge connector. The 2.1.1 update covered SerDes options for Ethernet and USB-C with USB 3.2 and DisplayPort alternate mode. SMARC 2.2, released in August 2024, is a maintenance release: it brings PCIe up to Gen 4, adds SoundWire audio as an alternative function, and corrects power domains, pin filling order and mechanical tolerances. It is intended to stay compatible with 2.1 designs.

A full SMARC 2.x pinout carries four PCIe lanes, two Gigabit Ethernet ports, USB 3 and USB 2, two CAN bus interfaces, UARTs, I2S and HDA audio, and display outputs that include HDMI or DisplayPort, LVDS, eDP and MIPI DSI. Native camera inputs over MIPI CSI-2 are a common reason vision, medical and robotics designs choose SMARC over a small x86 module. Modules are available with NXP i.MX, Qualcomm, MediaTek, TI and Rockchip processors, and with low-power Intel Atom parts.

Qseven dates from 2008 and was one of the first small-module standards to cover both x86 and Arm. Its modules are 70 × 70 mm, or 70 × 40 mm for the µQseven variant, with 230 edge fingers in an MXM connector at 0.5 mm pitch. The specification runs the module from a single 5 V supply and caps module power at 12 W, which keeps it in the same thermal class as SMARC.

SGET took the standard over from the original Qseven consortium in 2013, and revision 2.1, published in February 2016, is still current. Qseven 2.x carries PCIe, Gigabit Ethernet, USB 3.0, SATA, LVDS and a digital display interface. Modules are still made and supported, mainly for long-running products, but most new small-module designs now start on SMARC or OSM, and module makers typically point Qseven customers to SMARC for the next product generation.

COM Express is the PICMG standard behind most x86 computer-on-modules, and our COM Express range is built on it. It was ratified in 2005; revision 3.1, released in 2022, is current, and revision 3.2 is in development. Two things describe a module: its size and its pinout type. Sizes run from Mini at 84 × 55 mm through Compact at 95 × 95 mm and Basic at 125 × 95 mm to Extended at 155 × 110 mm. Compact, Basic and Extended modules use two 220-pin connectors; Mini uses one.

Revision 3.1 brought PCIe Gen 4 and USB4 to Types 6, 7 and 10, so recent modules can bring out the faster interfaces of current processors. Modules normally take 12 V from the carrier, and Mini modules accept a wide 4.75–20 V input, which suits battery and vehicle power. Type 2 and the other early pinouts, with legacy PCI and parallel ATA, survive in long-running products but are not used for new designs.

COM Express has one of the widest supplier bases of any module standard, with Intel Atom, Core, Core Ultra and Xeon and AMD Ryzen Embedded modules from many makers, and a large choice of off-the-shelf carriers. That is why it remains the default for x86 industrial, medical, transport and defence products, including stackable designs built to PC/104 form factors.

COM-HPC is PICMG’s high-performance standard, ratified in 2021. It uses two 400-pin connectors rated for signalling up to 32 Gb/s, enough for PCIe 5.0, and defines three groups of sizes. Client modules (95 × 120, 120 × 120 and 160 × 120 mm) carry desktop and mobile processors with display outputs, up to 49 PCIe lanes, USB4 and up to four SO-DIMMs or soldered memory. Server modules (160 × 160 and 200 × 160 mm) carry high-core-count processors, up to 65 PCIe lanes, up to eight DIMMs and 10 or 25 GbE, usually without graphics. COM-HPC Mini, added in revision 1.2, is a 95 × 70 mm module with one 400-pin connector, soldered memory and a 15 mm overall height, aimed at compact rugged systems.

Power is the other difference. COM-HPC modules take 12 V by default, with an optional 8–20 V input on Client and Mini. PICMG quotes input limits of about 358 W for Server, 251 W for Client and 107 W for Mini modules at the low end of their input ranges, before connector derating. Most designs run well below those limits, but they leave room for processors that COM Express cannot power or cool. COM-HPC adds to COM Express rather than replacing it; our guide to PCIe bandwidth explains why lane count and generation matter at this end of the range.

The carrier is where most of the engineering effort in a module-based product goes, and the standard sets its starting point. Each standards body publishes a design guide alongside the specification, and module makers publish reference carrier designs. Starting from one of those is usually the fastest route to a working board.

Signal integrity and layer count

PCIe, USB 3, USB4, DisplayPort and 10 GbE all need controlled-impedance, length-matched differential pairs from the module connector to the port. The faster the interface, the tighter the loss budget, so a COM-HPC carrier running PCIe 5.0 needs better board material and shorter routes, and sometimes retimers, compared with a SMARC carrier running PCIe Gen 3. As a rough guide, small OSM and SMARC carriers can often be built on fewer layers than COM Express and COM-HPC carriers, but the interfaces actually used decide the stack-up. Our explainer on PCIe covers lanes and links.

Power input

Every standard expects the carrier to supply a defined input and handle the power sequencing signals. SMARC modules run from a low-voltage input of about 3 to 5.25 V and Qseven from 5 V, so a carrier that takes a 12 or 24 V industrial supply needs its own conversion stage. COM Express and COM-HPC take 12 V by default, with wider-range options on COM Express Mini and COM-HPC Client and Mini. Wide-range, isolated or vehicle power inputs, ignition sensing and hold-up for a safe shutdown are all carrier functions.

Thermal design

COM Express and COM-HPC define a heat spreader interface on top of the module, so heat can be conducted into an enclosure wall or a heatsink without a fan; that is how sealed, rugged COM Express computers are cooled. SMARC and Qseven modules usually have a vendor heat spreader too, while OSM modules shed much of their heat into the carrier copper and need a heatsink or thermal pad on top at higher power. Size the thermal path for the module’s sustained power at the top of the industrial temperature range, not its typical figure, or the processor will throttle.

Board support package and software

x86 modules on COM Express and COM-HPC boot through UEFI firmware and run standard Windows and Linux images, so a module change is mostly a revalidation. Arm modules on OSM, SMARC and Qseven depend on a board support package for each module: the bootloader, kernel and device tree that describe the processor and the carrier. Most are delivered as Yocto Project layers for embedded Linux, sometimes with Android or a real-time operating system. The standard pinout makes the hardware interchangeable, but the device tree and drivers still have to be adapted for each module and carrier pairing.

Mechanics and environment

Edge-connector modules need standoffs and fixings so they cannot work loose under vibration, and COM Express and COM-HPC modules bolt down at defined holes. Soldered OSM modules avoid the connector entirely. Conformal coating, wide-temperature components and the connector choice on the carrier then decide how well the whole assembly suits a rugged computer. Our COM Express carrier boards show the approach: the module mounts on the underside with its heat spreader facing the enclosure.

Does the product need x86 software and PCIe bandwidth, or low power, small size and camera inputs? That question narrows most choices to PICMG or SGET before anything else. Size, I/O and production volume then settle the standard. The table gives typical starting points, not rules.

x86 or Arm

The processor architecture often decides the standard rather than the other way round. x86 brings mainstream Windows and Linux support and familiar tools. Arm brings lower power, integrated camera pipelines and an NPU on many parts, with the operating system built from a BSP. Our comparison of x86 and Arm sets out the trade-offs, and our guide to Arm processors in embedded systems covers the main families.

Volume and service

Volume pushes towards OSM, where removing the connector and a second assembly step pays back over thousands of boards. Field service pushes the other way: if a technician must be able to replace the module, or the product will be upgraded in service, choose a connector-based standard. Low volumes with a fixed I/O set often need no module at all, and a standard single board computer does the job.

No standard lasts for ever, but the standards bodies plan transitions, and some moves are much easier than others. Moves within a standard usually keep the carrier; moves between standards always need a new one.

Roadmaps

PICMG has COM Express revision 3.2 in development and a working group on COM-HPC 1.4. SGET maintains SMARC and OSM through revisions that keep earlier designs compatible. Neither body has announced an end to COM Express or SMARC, and module makers continue to launch modules on both. Treat roadmap features as plans until a specification is ratified.

Lifecycle is usually the reason to choose a module in the first place. Embedded processors are typically sold for seven to fifteen years, and an industrial product often has to be supported for longer. On a module-based design, the end of a processor means qualifying a new module on the same carrier, not a new board.

Three habits make that work. Ask for the module maker’s availability commitment in writing and track product change notices. Qualify a second module, ideally from another maker on the same standard, before you need it. And keep the software portable: a recorded BIOS configuration, or a Yocto layer kept separate from the application, makes a module change a revalidation rather than a rewrite. Our guide to planning a module end of life shows the same process for NVIDIA Jetson, where no second source exists.

Pin-compatible is not the same as identical. Two modules on the same standard can differ in which optional interfaces they bring out, their power draw, their heat spreader height and their temperature rating, so check each module against the carrier’s requirements rather than relying on the standard alone.

Our catalogue covers COM Express today: DFI modules in Mini and Compact sizes, a DFI Type 6 carrier, and Diamond Systems rugged carriers and COM Express SBCs that ship with the module and heat spreader fitted. We do not list SMARC, Qseven, OSM or COM-HPC modules at present; where a design needs one, we can advise on the standard and supply it as part of a build. We fit the module, memory, storage and heat spreader, image the operating system and test the assembled computer before it ships.

The COM332-B(R.A) puts a Type 6 module on a standard microATX motherboard form factor, which can be a practical way to start software work before a custom carrier exists. Our COM Express SBC case study shows a Type 6 design in a reverse vending machine, and a custom Jetson carrier board shows the carrier design process for a proprietary module.