AI 模型和平台
NRF71芯片中的北欧半导体 Wi-Fi和边缘AI组合
在 Google 上的首选来源中添加 Unite.AI
北欧半导体宣布 2026年11月3日的nRF71系列，描述了电池操作设备的低功率芯片，它们结合了Wi-Fi，蓝牙LE,Tread，应用程序处理和端侧 AI支持，其备用功率比其目前的nRF70系列Wi-Fi产品低20倍。
应用程序和连接的一个芯片
第一个芯片是nRF7120和nRF7120E. 两者结合了Wi-Fi，一个2.4GHz的多protocol无线电和一个设备上的主机处理器。 北欧早期的NRF70系列产品是用于在其蓝牙LE SoCs上构建的设计中加入低功率Wi-Fi的Wi-Fi伴生IC. 新架构取消了单独的主机微控制器的需求，意在简化设计，同时减少其板块空间和组件需求。
北欧列出了在DTIM 10连接时的平均电流，在2.4GHz的14个微amps和5GHz的8个微amps中，两个SoCs都是相同的。 在nRF Connect SDK中默认启用目标Wake Time和Wi-Fi电源，但带有专用的低功率接收器。 有了这些功能，一个设备就可以留在网络上，利用现有基础设施在传输之间消耗的电流很少。 这就是北欧电池电源无线网络定位背后的操作模式：保持连接而不让备用消耗主导电力预算。
AI 硬件和同步无线电
nRF7120支持双波段Wi-Fi 6在2.4和5GHz. nRF7120E 新增 6GHz波段用于三波段Wi-Fi 6E并集成Axon神经处理单元；基芯片没有专用的AI加速器。 两者都使用256MHz Arm Cortex-M33应用处理器和额外的256MHz RISC-V共处理器，用于实时任务和外围仿真。 他们的4 MB的非挥发性内存和1 MB的用户可访问RAM为无线堆栈，Matter，客户应用程序和固件更新提供了空间。
北欧称的Axon NPU运行的机器学习模型比Cortex-M33单机运行速度快10至17倍，使得关键词的定位，异常检测和图像分类在电池动力设备上实用。 NRF7120E还可以在当地运行传感器聚变和音频模型。 通过在芯片上处理这些工作量，它可以仅仅传递无线网的推论结果，减少电源和带宽要求。
开发者还可以用北欧的Neuton软件从自己的数据中构建超小型模型，仅用于对应用处理器的推论。 该软件路由给基础nRF7120提供了端侧推论选项，即使没有专用加速器。 无 Cloud支持超空模型更新。
独立的Wi-Fi和2.4GHz多protocol无线电具有独立的RF和处理块，可以同时接收。 因此，一个产品可以保持其与云的Wi-Fi连接，同时继续收听蓝牙LE或线程流量。 芯片都支持 Matter 和 NFC，并且使用与北欧的nRF54和nRF70系列相同的nRF Connect SDK. 对于 Matter over Wi-Fi 设备，蓝牙 LE在Wi-Fi 提供网络连接的同时进行调试。
北欧正在针对电池动力智能家庭枢纽，Matter 桥，Thread 边境路由器，智能锁，安全板，可穿戴器和工业传感器。 这些设计可以使用集成的无线电和应用程序处理器，而不是通过多个设备集成这些功能。
模块选项和早期取样
nRF7120 模块包有SOC，有RF匹配和滤波器，一个传输/接收开关和二聚体，一个64MHz晶体，一个屏蔽器和一个10×14毫米包中的PCB天线。 它的针兼容的nRF7120-LAAA和nRF7120E-LAAA的变体允许一个板在双波段Wi-Fi 6和三波段Wi-Fi 6E之间移动，与Axon加速度没有重新设计。 北欧计划FCC、CE和日本对该单元的认证；这些认证尚未完成。
NRF71系列是用选定的客户进行采样，产品页描述的是基于项目的早期采样而不是一般的可用性。 因此，该公告对一芯片结构进行了详细配对，初始可获性有限：早期客户项目将决定集成无线电、应用处理器和可选Axon加速器如何转化为电池动力成品。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
AI Models & Platforms
Nordic Semiconductor Combines Wi-Fi and Edge AI in nRF71 Chips
Add Unite.AI to your preferred sources on Google
Nordic Semiconductor announced the nRF71 Series on November 3, 2026, describing low-power chips for battery-operated devices that combine Wi-Fi, Bluetooth LE, Thread, application processing and on-device AI support, with up to 20 times lower standby power than its current nRF70 Series Wi-Fi products.
One Chip for Application and Connectivity
The first chips are the nRF7120 and nRF7120E. Both combine Wi-Fi, a 2.4 GHz multiprotocol radio and a host processor on one device. Nordic’s earlier nRF70 Series products were Wi-Fi companion ICs used to add low-power Wi-Fi to designs built on its Bluetooth LE SoCs. The new architecture removes the need for a separate host microcontroller and is intended to simplify designs while reducing their board space and component requirements.
Nordic lists average current while connected at DTIM 10 of 14 microamps at 2.4 GHz and 8 microamps at 5 GHz for both SoCs. Target Wake Time and Wi-Fi power save with a dedicated low-power receiver are enabled by default in nRF Connect SDK. With those features active, a device can remain on the network and consume little current between transmissions using existing infrastructure. That is the operating pattern behind Nordic’s battery-powered Wi-Fi positioning: maintaining a connection without making standby consumption dominate the power budget.
AI Hardware and Simultaneous Radios
The nRF7120 supports dual-band Wi-Fi 6 at 2.4 and 5 GHz. The nRF7120E adds the 6 GHz band for tri-band Wi-Fi 6E and integrates the Axon neural processing unit; the base chip has no dedicated AI accelerator. Both use a 256 MHz Arm Cortex-M33 application processor and an additional 256 MHz RISC-V coprocessor for real-time tasks and peripheral emulation. Their 4 MB of non-volatile memory and 1 MB of user-accessible RAM provide room for the wireless stacks, Matter, the customer application and firmware updates.
Nordic said the Axon NPU runs machine-learning models 10 to 17 times faster than the Cortex-M33 alone, making keyword spotting, anomaly detection and image classification practical on battery-powered devices. The nRF7120E can also run sensor-fusion and audio models locally. By processing those workloads on the chip, it can transmit just the inference results over Wi-Fi, reducing power and bandwidth requirements.
Developers can also build ultra-small models from their own data with Nordic’s Neuton software for inference on the application processor alone. That software route gives the base nRF7120 an on-device inference option even without a dedicated accelerator. nRF Cloud supports over-the-air model updates.
The separate Wi-Fi and 2.4 GHz multiprotocol radios have independent RF and processing blocks and can receive simultaneously. A product can therefore maintain its Wi-Fi connection to the cloud while continuing to listen for Bluetooth LE or Thread traffic. Both chips support Matter and NFC and use the same nRF Connect SDK as Nordic’s nRF54 and nRF70 Series. For Matter over Wi-Fi devices, Bluetooth LE handles commissioning while Wi-Fi provides network connectivity.
Nordic is targeting battery-powered smart-home hubs, Matter bridges, Thread border routers, smart locks, security panels, wearables and industrial sensors. Those designs can use the integrated radios and application processor rather than assembling those functions across multiple devices.
Module Option and Early Sampling
The nRF7120 module packages either SoC with RF matching and filtering, a transmit/receive switch and diplexer, a 64 MHz crystal, shielding and a PCB antenna in a 10 × 14 mm package. Its pin-compatible nRF7120-LAAA and nRF7120E-LAAA variants allow a board to move between dual-band Wi-Fi 6 and tri-band Wi-Fi 6E with Axon acceleration without a redesign. Nordic plans FCC, CE and Japan certifications for the module; those certifications have not yet been completed.
The nRF71 Series is sampling with selected customers, and the product pages describe project-based early sampling rather than general availability. The announcement therefore pairs a detailed one-chip architecture with limited initial availability: early customer projects will determine how the integrated radios, application processor and optional Axon acceleration translate into finished battery-powered products.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.