What Is a System-on-Module (SoM)?

A system on module (SoM) is a complete computing core, with the processor, memory, storage and power management built onto one small board, that plugs into a second board called a carrier. The carrier supplies the connectors, expansion slots, sensor interfaces and mechanical fit the product needs; the module supplies the compute. Separating the two lets a product keep its carrier design while the module underneath is upgraded, replaced or scaled.

The same idea goes by several names. PICMG calls it a computer-on-module (CoM), which is where the COM Express and COM-HPC standards get their names. NVIDIA calls its Jetson products modules, and the boards they plug into Jetson carrier boards. Raspberry Pi calls its version a compute module. In every case the principle is the same: the difficult, high-speed, short-lifecycle engineering lives on the module, and the application-specific engineering lives on the carrier.

This guide explains what goes on the module and what goes on the carrier, how a SoM compares with a system-on-chip and a single board computer, which module standards matter and how they differ, where NVIDIA Jetson fits, the advantages of a module-based design and when an SBC is simpler, and how to choose a module and turn it into a working system.

The vocabulary overlaps, so it helps to pin it down. System on module and computer on module mean the same thing; SoM is the generic term and CoM is the PICMG term used by the COM Express and COM-HPC standards. A system-on-chip (SoC) is the silicon itself, the single package that holds the CPU cores, graphics and I/O controllers, and is what sits at the centre of the module. A single board computer (SBC) is a complete computer on one board, with the processor, memory and all the I/O connectors together, and no module to remove.

There is a grey area worth knowing. Several rugged board makers sell a carrier with the module already fitted and call the result a COM-based SBC. It is used like an SBC, but because the module can be changed it keeps the lifecycle advantage of a modular design. Our COM Express carrier boards are sold both ways: bare, for customers fitting their own module, and preassembled with the Intel module, heat spreader and storage in place, while our COM Express SBCs come ready to run.

For more on the chip at the centre of every module, how a system-on-chip combines CPU, graphics and I/O on one die explains the trade-offs against a discrete design.

The split is consistent across every standard. The module carries the parts that are hardest to lay out and that change most often: the processor or system-on-chip, DRAM (usually soldered), boot storage such as eMMC or NVMe, the power sequencing the processor demands, and often the Wi-Fi or trusted platform module. High-speed signals such as PCIe, USB 3, display and Ethernet are routed to the module connector but not to any physical port.

The carrier board turns those signals into a product. It decides which of the available PCIe lanes become an M.2 slot, a MiniCard socket or a 10GbE controller; which USB lines reach a locking connector on the front panel; whether the display output is HDMI, DisplayPort or an embedded panel; and what the board looks like mechanically. It also carries the parts the module does not: the DC input stage and any wide-range or isolated power, serial ports, CAN bus, digital and analog I/O, camera connectors, ignition control and the mounting holes that match the enclosure.

That is why two products on the same module can look nothing alike. A Jetson Orin NX on a 3.5 inch industrial carrier with dual Gigabit Ethernet and CAN is a machine controller; the same module on a carrier with four GMSL camera inputs is a perception computer. The compute is identical; the carrier is the difference.

Electrically, the module connector is a contract. The standard fixes which pin carries which signal, so a carrier designed to it knows where to find each PCIe lane, USB port, display interface and power rail, whichever vendor made the module. The carrier supplies power, usually a single input such as 12 V or 5 V depending on the standard, and the module brings itself up from that.

In software, the module is described by a board support package (BSP): the boot loader, kernel, drivers and device tree or ACPI tables that tell the operating system what hardware is present. x86 modules use a standard UEFI BIOS and boot Windows or Linux much like a PC. Arm modules usually ship with a Linux BSP, often built with Yocto, and the carrier’s own devices are added to the device tree. When a carrier is designed, the BSP is extended to match it.

Open module standards are published by two bodies. PICMG maintains COM Express and COM-HPC, aimed mainly at x86 performance; SGET maintains SMARC, Qseven and OSM, aimed mainly at lower-power Arm and x86 designs. NVIDIA Jetson sits outside both as a proprietary family. The table compares them by connector, processor class and typical power, which is how most selections narrow down.

Power classes are broad because each standard covers a range of modules: a Type 6 carrier might take a low-power Atom module or a high-performance Core module. COM-HPC is not a replacement for COM Express, which continues alongside it; our comparison of COM-HPC and COM Express sets out where each one fits.

A computer on module (COM), also written computer-on-module, is the same thing as a system on module: the term comes from PICMG and SGET, the two bodies that publish the main open module standards. The table compares the four open standards most often met in industrial designs.

COM Express suits x86 performance with PCIe expansion, SMARC and Qseven suit low-power Arm designs that need a removable module, and OSM suits products built in volume where the module is placed like any other component. Our COM Express modules fit the carrier boards described above.

For a full comparison of sizes, connectors and carrier design across all five standards, see our SMARC vs Qseven vs OSM vs COM Express guide.

COM Express defines pinout types as well as sizes. The type fixes which interfaces appear on the connector, so a module and carrier must share a type to work together, while modules of different sizes but the same type can often share a carrier with suitable mounting points. Three types are current; older ones still appear on long-life products.

Our FAQ on COM Express module types covers the history and the less common types in more detail.

Yes. Every Jetson is a system on module: an NVIDIA Orin or Thor SoC with its memory, storage and power management on a small board, designed to plug into a carrier. What sets Jetson apart is that it follows NVIDIA’s own specification rather than an open standard, so modules come from one source and each generation defines its own connector and carrier requirements.

Orin Nano and Orin NX share a 260-pin SO-DIMM-style edge connector and a common carrier design, so a carrier built for Orin NX also takes Orin Nano. Jetson AGX Orin and AGX Thor use a board-to-board connector and a higher power budget. A Thor design therefore needs a Jetson Thor carrier board rather than an Orin one.

Buyers accept the single source for the AI performance and software stack Jetson brings, and NVIDIA publishes lifecycle dates for each module. Our guides explain which Jetson module to choose and the difference between a Jetson module and a developer kit, and our range of NVIDIA Jetson carrier boards covers every current generation.

Most SMARC, Qseven and OSM modules carry Arm application processors, and the choice of SoC drives the choice of module. NXP i.MX processors are common in long-life industrial and medical products, with published longevity programmes; the NXP i.MX 8M Plus adds an NPU for light inference. Rockchip parts such as the RK3588 offer high multimedia performance at low power, and Qualcomm Dragonwing processors bring integrated AI engines and wireless.

Arm modules usually run Linux or Android rather than Windows, through the vendor’s BSP, and pin muxing means the same connector pin can serve different functions on different modules, so carriers are checked against each module’s documentation as well as the standard. Our guides to Arm processors for embedded and industrial use and x86 vs Arm cover the processor side.

The strongest reason is lifecycle. Embedded processors are typically available for seven to fifteen years, but the products built on them often have to be supported for longer, and the need for more performance usually arrives well before the end of that window. With a module-based design the carrier, the enclosure, the cabling and the certification evidence stay as they are; the module is swapped for a newer one on the same connector and the software is revalidated. Rugged board makers have kept products in production for well over a decade this way.

The second reason is scalability. Most standards allow a range of modules on one carrier, from a low-power part to a high-performance one, so a single product design can be offered at several performance points. A Type 6 carrier might ship with an Intel Atom module for a basic configuration and a Core or Xeon module for the top one; a Jetson carrier designed for Orin NX also takes Orin Nano.

The third is time and risk. Laying out DDR5 memory and PCIe Gen 4 to a modern processor is exacting, multilayer work with a long validation cycle. Buying that as a finished, tested module lets the product team spend its engineering time on the carrier, which is comparatively simple, and lets a custom carrier be turned around in months rather than the year or more a full board design can take.

Finally there is supply. Because the module is a standard part with several sources, a carrier designed to COM Express or SMARC is not tied to one vendor’s roadmap. Jetson is the exception: it is single-source, and buyers accept that for what it brings.

A module is not automatically the right answer. If the product has one fixed configuration, a standard form factor such as Mini-ITX or 3.5 inch already fits the enclosure, and the lifecycle requirement is covered by the board maker’s own long-life program, then a single board computer is simpler to buy, simpler to cool and usually lower cost at the volumes most industrial products run at. There is no connector between the processor and the I/O, which also removes one source of vibration-related faults.

The trade-off is what happens when the processor or the I/O has to change. On an SBC that is a new board. Teams that expect either to happen inside the product’s life, or that need an I/O set no standard board offers, are the ones a module-based design pays back. Our guide to what a single board computer is covers SBC form factors in detail.

A carrier is where the product’s character is set, and where most of the integration work happens. Whether it is bought off the shelf, adapted or designed from scratch, the same items need deciding.

Each of those choices ends up as copper on a printed circuit board, and they are easier to get right with the enclosure, cabling and certification in view from the start. A carrier sized for one spare expansion socket is often what turns a two-year product into a ten-year one.

Things Embedded supplies Jetson and COM Express carrier boards as working systems: we recommend the module and carrier to start from, fit and image them, validate thermals and I/O in the target enclosure, and support the result through its life. Where no standard carrier fits the requirement, we design a custom one; a fanless GPU computer built on a custom Jetson carrier board is a recent example, and a COM Express SBC running deep learning on Intel Iris Xe graphics shows the x86 side.

Validation is where modular designs earn their reputation. We test the complete system across the operating temperature range it will see, with the storage, wireless and expansion cards fitted, and record the configuration so the same build can be repeated for the life of the product.

Environment. Where does it run and how is it powered? Wide-temperature operation, conformal coating, shock and vibration, and a vehicle or MIL-STD power input all belong on the carrier, and the module has to be rated to match. Latching connectors and a heat spreader that mates to the enclosure matter more in a vehicle than in a cabinet.

Performance. Is the workload general-purpose x86, or AI inference? An Intel Core or Xeon module on a Type 6 carrier suits control, data acquisition and machine vision with conventional software; a Jetson module and carrier suit neural-network inference on camera and sensor data, and an Arm module with an NPU or an AI accelerator on the carrier suits lighter inference at low power.

Expansion. Which interfaces does the product need on day one, and which might it need later? PC/104 and PCIe/104 stacking, MiniCard and M.2 sockets, camera inputs and isolated I/O are carrier decisions.

Lifecycle. How long must it be supportable, and who supports it? Ask for the module and carrier availability commitments in writing, confirm that a newer module on the same standard is on the roadmap, and plan the software so a module swap is a revalidation rather than a rewrite.

For Jetson, NVIDIA publishes end-of-life dates for each module, which our FAQ on how long Jetson is supported summarizes. For Arm and x86 modules, the module maker’s longevity program is the figure to ask for.

These boards show the module and carrier split in practice. We fit the module, heat spreader, memory and storage, configure the BIOS, image the operating system and test the assembled board or system before it ships, or supply bare carriers to customers fitting their own modules.

Stackable carriers like the Gemini follow the PC/104 family of form factors, which lets I/O, power and network boards stack on the carrier in rugged systems.