The explosion of edge AI devices—from smart cameras and industrial gateways to autonomous drones and in-vehicle systems—has placed new demands on embedded storage. Unlike cloud-based AI, edge AI processes data locally, requiring a delicate balance between code storage, model storage, and data logging. Choosing the wrong memory can lead to boot failures, corrupted data, high power consumption, or even project failure.
从智能摄像头、工业网关到自动驾驶无人机和车载系统，边缘AI设备的爆发式增长对嵌入式存储提出了全新要求。与云端AI不同，边缘AI在本地处理数据，需要在代码存储、模型存储和数据日志之间实现精妙平衡。选错存储芯片可能导致启动失败、数据损坏、功耗过高，甚至项目遭受不必要损失。
This guide will help you understand the strengths and weaknesses of the four main memory types—NOR Flash, NAND Flash, eMMC, and UFS/SD—and provide a clear framework for selecting the right storage for your next edge AI project.
本文将帮助你理解四种主流存储芯片——NOR Flash、NAND Flash、eMMC 和 UFS/SD——各自的优缺点，并为你的下一个边缘AI项目提供清晰的选型框架。
Each memory type serves a distinct role in edge AI systems. The table below summarizes their key characteristics.
每种存储芯片在边缘AI系统中扮演不同角色。下表总结了它们的关键特性。
| Memory Type | Primary Role | Read Speed | Write Speed | Reliability | Cost (per GB) | Typical Application |
|---|---|---|---|---|---|---|
| NOR Flash | Code storage, fast boot | Fast (random access) | Slow | Very high (no bad block management needed) | High | BIOS, firmware, FPGA config, secure boot |
| NAND Flash | Bulk data storage | Fast (sequential) | Medium | Medium (requires ECC & wear leveling) | Low | Large files, media storage, AI models |
| eMMC | Embedded system drive | Medium | Medium | High (built-in controller) | Medium | OS, application storage |
| UFS / SD Card | Removable / high‑speed storage | Very fast (UFS) | Very fast (UFS) | High (built‑in controller) | Medium | Data logging, external storage |
| 存储类型 | 主要用途 | 读取速度 | 写入速度 | 可靠性 | 单位成本 | 典型应用 |
|---|---|---|---|---|---|---|
| NOR Flash | 代码存储、快速启动 | 快（随机访问） | 慢 | 非常高（无需坏块管理） | 高 | BIOS、固件、FPGA配置、安全启动 |
| NAND Flash | 大容量数据存储 | 快（顺序访问） | 中等 | 中等（需ECC及磨损均衡） | 低 | 大文件、媒体存储、AI模型 |
| eMMC | 嵌入式系统盘 | 中等 | 中等 | 高（内置控制器） | 中等 | 操作系统、应用存储 |
| UFS / SD卡 | 可移动/高速存储 | 非常快（UFS） | 非常快（UFS） | 高（内置控制器） | 中等 | 数据日志、外置存储 |
Different edge AI applications have different priorities. Here’s how to match memory types to typical use cases.
不同的边缘AI应用有不同的侧重点。下面是如何将存储类型与典型用例相匹配。
Requirements | 需求
Fast boot time | 快速启动
High reliability (no file system corruption) | 高可靠性（无文件系统损坏）
Low power consumption | 低功耗
Code executed in place (XIP) | 代码原地执行（XIP）
Recommended | 推荐
NOR Flash (e.g., IS26KL256S) for code and model storage.
使用 NOR Flash（如 IS26KL256S）存储代码和模型。
Why NOR Flash? It allows direct execution of code without copying to RAM, eliminates the need for a complex file system, and provides the highest reliability for critical boot code.
为什么选NOR Flash？ 它允许代码原地执行，无需拷贝到RAM，省去了复杂的文件系统，并为关键启动代码提供了最高的可靠性。
Requirements | 需求
Run an operating system | 运行操作系统
Store large AI models (several hundred MB) | 存储大容量AI模型（几百MB）
High read/write throughput for video streams | 视频流读写吞吐量高
Recommended | 推荐
eMMC for the system drive.
NAND Flash or SD card for data and model storage.
eMMC 作为系统盘。
NAND Flash 或 SD卡 用于数据和模型存储。
Why eMMC? It provides a standardized, reliable embedded storage with built‑in wear leveling and bad block management.
为什么选eMMC？ 它提供标准化、可靠的嵌入式存储，内置磨损均衡和坏块管理。
Requirements | 需求
Extended temperature range (-40°C to +85°C or higher) | 宽温工作范围
Long data retention (>10 years) | 长数据保持时间
High endurance (frequent power cycles) | 高耐久性（频繁上下电）
Recommended | 推荐
Automotive‑grade NOR Flash for secure boot and critical code.
Industrial‑grade NAND or eMMC for data storage.
车规级NOR Flash 用于安全启动和关键代码。
工业级NAND 或 eMMC 用于数据存储。
Key considerations: Check the datasheet for operating temperature, data retention, and program/erase cycles.
关键考虑因素：查阅数据手册中的工作温度、数据保持时间和擦写次数。
Use this checklist to narrow down your options:
使用此清单缩小选择范围：
Capacity | 容量
How large are your OS, AI model, and data logs?
操作系统、AI模型和数据日志各有多大？
Interface | 接口
Does your processor support SPI, Quad SPI, Parallel, or MMC?
处理器支持SPI、Quad SPI、Parallel还是MMC？
Speed | 速度
What is your required boot time? Model load time?
启动时间要求？模型加载时间要求？
Power | 功耗
Is the device battery‑powered or mains‑powered?
设备是电池供电还是外部供电？
Reliability | 可靠性
Do you need ECC, wear leveling, or industrial temperature range?
是否需要ECC、磨损均衡或工业级温度范围？
Cost & Availability | 成本与供应
What is your budget? Does the supplier have stock for long‑term production?
预算多少？供应商是否有长期生产所需的现货？
Scenario: You are designing a vibration sensor for industrial machinery that runs a TinyML model to detect anomalies. The MCU is an STM32H7. The firmware occupies 512KB, and the model requires an additional 256KB. The sensor must boot in under 200ms and operate in an ambient temperature of -25°C to +85°C.
场景：你正在设计一款工业设备振动传感器，运行TinyML模型检测异常。主控采用STM32H7。固件占512KB，模型需要额外256KB。传感器必须在200ms内启动，并在-25°C至+85°C的环境下工作。
Analysis | 分析
Total storage needed: ~768KB.
XIP capability is critical for fast boot.
Industrial temperature range is mandatory.
所需总存储：约768KB。
原地执行能力对快速启动至关重要。
必须支持工业级温度范围。
Recommended | 推荐
A 256Mb (32MB) Parallel NOR Flash with industrial temperature support, such as the IS26KL256S-DABLI00.
256Mb（32MB）并行NOR Flash，支持工业级温度，例如 IS26KL256S-DABLI00。
This device offers:
XIP (Execute in Place) – no RAM copying, instant boot.
Industrial temperature grade – reliable operation in harsh environments.
Parallel interface – fast random access for code execution.
Sufficient capacity – easily stores both firmware and the AI model.
该器件提供：
XIP（原地执行） – 无需拷贝到RAM，启动瞬间完成。
工业级温度范围 – 严苛环境下可靠运行。
并行接口 – 快速随机访问，适合代码执行。
足够容量 – 轻松容纳固件和AI模型。
Selecting the right memory for your edge AI device is not a one‑size‑fits‑all decision. By understanding the strengths of NOR Flash, NAND, eMMC, and UFS, and by aligning them with your application’s requirements for speed, reliability, power, and cost, you can build a robust foundation for your product.
为边缘AI设备选择合适的存储芯片并非一刀切的决策。通过理解NOR Flash、NAND、eMMC和UFS各自的优势，并根据应用对速度、可靠性、功耗和成本的要求进行匹配，你可以为产品奠定坚实的基础。
Whether you are prototyping with an MCU or scaling to industrial‑grade production, the right memory partner can make all the difference.
无论你是在用MCU进行原型开发，还是转向工业级量产，合适的存储合作伙伴都能带来天壤之别。
Our engineering team specializes in industrial and automotive‑grade memory solutions. We offer:
Free technical consultation – we’ll help you match the right memory to your processor and application.
Free samples – test our recommended parts in your own design.
Stock availability – we maintain deep inventory for long‑term production.
我们的工程团队专注于工业和车规级存储解决方案。我们提供：
免费技术咨询 – 帮助你为处理器和应用匹配合适的存储。
免费样品 – 在你自己的设计中测试我们推荐的产品。
现货供应 – 我们为长期生产保持充足库存。
Contact us today for a free consultation or to request samples.
立即联系我们获取免费咨询或申请样品。