What Is a System-on-Module — and Why Do AI Robots Run on One?

A system-on-module, or SoM, is a small circuit board — often not much bigger than a credit card — that packs a processor, memory, storage, and networking chips into a single, pre-built unit. Instead of designing a computing system from individual chips, an engineer buys a SoM and plugs it into a larger “carrier board” built for their specific product. It’s why a growing number of robots, smart cameras, and factory machines are built not around a raw chip, but around a ready-made module.

What a SoM actually is

A system-on-module integrates the core of a computer — a processor, RAM, flash storage, power management, and often Wi-Fi or Ethernet — onto one small board. That distinguishes it from a bare system-on-chip (SoC), which is just the processor die itself and still needs an engineering team to wire up memory, power circuitry, and interfaces around it. It also distinguishes a SoM from a full single-board computer like a Raspberry Pi: an SBC is a finished product with its own case-ready connectors, while a SoM is a component meant to be built into something else.

How it fits together

The SoM plugs into a carrier board (sometimes called a baseboard) through a dense connector, much like a laptop’s replaceable RAM stick plugs into a slot. The carrier board supplies the ports, sensors, and physical layout specific to the finished product — a camera housing, a robot arm controller, an industrial gateway — while the SoM supplies the actual computing. A manufacturer can swap in a faster or cheaper SoM later without redesigning the whole product, and it can prototype with an off-the-shelf module instead of laying out a processor system from scratch, which can take a specialized team a year or more to get right.

Why robotics and edge AI depend on them

Robots and smart cameras need to process camera and sensor data on the device itself, in real time, rather than sending everything to the cloud — a category often called edge AI. That takes a genuine computing system: a processor plus enough memory and I/O to handle multiple camera feeds, run a vision model, and control motors, all within a power and size budget a robot can actually carry. Building that from individual chips is a multi-year undertaking most robotics startups can’t afford, so many buy a SoM instead.

AMD’s Kria line is a widely used example. Its K26 module wraps a Zynq UltraScale+ adaptive system-on-chip — combining fixed processor cores with reconfigurable logic — into a module rated for up to 1.4 trillion AI operations per second, with support for up to 15 camera inputs and native support for ROS 2, the standard robotics software framework. AMD sells it both as a bare module for mass production and as a starter kit — the KV260 Vision AI kit, priced at roughly $250–$285 depending on the distributor as of July 2026 — so a small team can prototype before committing to volume manufacturing. Nvidia’s competing Jetson line works the same way, which is part of why AI-hardware makers keep launching new SoMs: whoever’s module robotics developers standardize on gets designed into years of future products.

Beyond robots

SoMs aren’t unique to robotics — the same logic drives industrial gateways, medical scanners, point-of-sale terminals, and networking gear, anywhere a company needs real computing power but doesn’t want to become a chip-design shop. Vendors like NXP, Toradex, and Variscite sell general-purpose SoMs for exactly this reason, alongside AMD’s and Nvidia’s AI-focused lines. What’s changed recently is that AI workloads — running a vision or language model locally — have pushed SoM makers to add dedicated AI accelerators to modules that used to be built around ordinary CPUs, which is why the category now overlaps so heavily with edge AI.

In the news

AMD’s push to win more of that robotics business is a live story: see our report on AMD’s new Kria AI robotics platform, positioned as a direct challenge to Nvidia in the space.

Sources: Wikipedia’s System on module entry; AMD Kria K26 SOM product documentation; DigiKey product listing for the Kria KV260 Vision AI Starter Kit (July 2026).