Industry Insights

Embedded Storage Selection: eMMC vs. UFS vs. SPI NOR Flash by Application

2026-03-02embedded storage / eMMC / UFS

Storage device selection directly shapes an embedded product's performance, cost, and reliability. eMMC, UFS, and SPI NOR Flash — the three mainstream embedded storage options — differ in bandwidth, capacity, boot speed, and cost. Starting from typical scenarios in industrial control, IoT terminals, and edge AI, this article walks through the selection logic and design considerations for embedded storage.

Three Embedded Storage Options at a Glance

SPI NOR Flash: small capacity (1MB-256MB), simple interface (SPI/QSPI), XIP (execute-in-place) support, and fast boot. Used mainly for system boot code, FPGA configuration files, and code storage in embedded systems.

eMMC: mid-range capacity (4GB-256GB), a standardized interface (eMMC 5.1, 400MB/s in HS400 mode), and an integrated flash controller with FTL. Used mainly for operating system and application storage in embedded Linux/Android systems.

UFS: large capacity (32GB-1TB), high bandwidth (2.1GB/s dual-lane on UFS 3.1, 4.2GB/s on UFS 4.0), full duplex, and multi-queue command processing. Used mainly in high-performance embedded systems, edge AI devices, and automotive compute platforms.

Industrial Control

Industrial equipment (PLCs, HMIs, industrial gateways) typically runs a slim Linux or RTOS, with moderate capacity needs (8-64GB) but very high demands on reliability and wide-temperature support. Industrial-grade eMMC (rated -40°C to +85°C) is the recommended choice, operated in pSLC mode to improve endurance. Where fast boot matters, a dual-storage architecture — SPI NOR Flash for the boot code plus eMMC for the operating system — works well.

IoT Terminals

IoT devices span a wide spectrum, from simple sensor nodes to complex smart gateways. Low-end IoT devices (sensors, actuators) can usually meet their storage needs with SPI NOR Flash alone; mid- and high-end IoT devices (smart cameras, voice assistants) need eMMC to store AI models and multimedia data.

Edge AI

Edge AI devices such as inference boxes and smart vision terminals need to store fairly large AI model files (typically hundreds of MB to several GB) along with real-time inference data. UFS's high bandwidth and low latency significantly shorten model load times and improve inference efficiency. For scenarios with frequent model updates, UFS's strong write performance is another key advantage.

A Selection Decision Matrix

Selection should weigh the following factors together: performance requirements (bandwidth, IOPS, latency), capacity needs, temperature range, endurance requirements, BOM cost budget, and supply chain stability. JH Semiconductor supplies the full range of eMMC, UFS, and SPI NOR Flash products, and offers selection consulting tailored to specific applications.

Keywords
embedded storageeMMCUFSSPI NOR FlashIoTedge AIindustrial control