- 1. Buy Kits
- 2. Create User Account
- 3. 开发环境
- 4. 探索演示内容
What is AI/ML at Tiny Edge?
In the IoT industry, "edge" refers to devices that perform computation locally instead of relying on cloud computing. The latest development, Tiny Edge, brings computation closer to where data is generated, such as sensor nodes. This shift moves from a centralized, cloud-based solution to a distributed network of edge nodes that collect, process, and infer data locally. By 2027, over 3 billion devices are expected to be sold with TinyML, a subset of AI focused on deploying machine learning models on Tiny Edge devices. This growth is driven by societal trends like the need for speed, privacy and connectivity. Additionally, the transition from wired to wireless technology is further accelerating the adoption of Tiny Edge devices.
Applications of Machine Learning using Silicon Labs’ SoCs
Silicon Labs' Wireless SoCs support a range of ML applications, such as sensor signal processing for predictive and preventative maintenance, bio-signal analysis for healthcare, and cold chain monitoring. They also enable audio pattern matching for security applications, voice commands for smart device control, and low-resolution vision for tasks like people counting and presence detection. The SoCs offer various RAM sizes to accommodate different application requirements. Machine learning models are applied to data from sensors such as microphones, cameras, and those measuring time-series data like acceleration and temperature. These models include audio pattern matching, wake word/command word detection, fingerprint reading, always-on vision, and image/object classification and detection. The detected events can then be further processed according to the requirements.
AI/ML Journey with Silicon Labs
Silicon Labs can accelerate the development of AI/ML devices, starting by outlining each step in the process and helping you along each stage of your project. 我们将简化您的开发之旅，帮助您更快速、更高效地将设备推向市场。
We have outlined below three key stages of the AI/ML Developer Journey, along with what is required to successfully complete each stage.
开始
Build Your Own Solution
Pre-Built Solution
1. 购买套件：硬件和示例
Silicon Labs provides offers several development and explorer kits, from ultra-low-cost, small form factor to compact, feature-packed platforms designed for robust networks. We have several exciting demos, including wake-word detection, Pacman, and gesture control. These feature-rich kits support multiple protocols and come in different memory configurations with a wide variety of sensors and peripherals for quick debugging and rapid prototyping. Based on the demos you are interested in, please select the kit that best fits your needs below. The demos are hardware agnostic.
 |
 |
 |
 |
|
| 套件 | EFR32xG24 开发套件 | EFR32xG28 Explorer 套件 | EFR32xG26 +10 dBm Dev Kit |
SiWx917 Wi-Fi 6 & Bluetooth LE Dev Kit |
| OPN | (xG24-DK2601B) | (xG28-EK2705A) | (xG26-DK2608A) | (SiWx917-DK2605A) |
| 支持的协议 | Bluetooth, Matter, Proprietary, Thread, Zigbee | Bluetooth, Sidewalk, Wi-SUN, Z-Wave | Bluetooth, Matter, Proprietary, Thread, Zigbee | Bluetooth, Wi-Fi |
| 描述 | EFR32xG24 开发套件是一款紧凑、功能丰富的开发平台。它能够快速开发无线物联网产品，并完成原型设计。 | The EFR32xG28 Explorer Kit is small form factor development and evaluation platform based on the EFR32xG28 SoC focused on rapid prototyping and concept creation of IoT applications for Sub-GHz and Bluetooth LE. | EFR32xG26-DK2608A 开发套件是一款紧凑、功能丰富的开发平台。它能够快速开发无线物联网产品，并完成原型设计。 | SiWx917 Wi-Fi 6 和蓝牙 LE 5.4 开发套件是一个紧凑但功能丰富的开发平台，用于快速测试、开发和原型设计无线物联网应用。 |
| 价格 | $79 USD | $34 USD | $89 USD | $40 USD *ML enablement in alpha, contact sales |
| 闪存/RAM | 1536 kB / 256 kB | 512 kB / 32 kB | 3.2 MB / 512 kB | 8 MB Flash / 8 MB external PSRAM |
| MVP | ✔ | ✔ | ✔ | ✔ |
| 传感器 | Inertial Sensor, Stereo Microphones, Pressure Sensor, Ambient Light Sensor | 温度传感器 | Inertial Sensor, Stereo Microphones, Pressure Sensor, Ambient Light Sensor | Temperature Sensor, Humidity Sensor, Inertial Sensor, Digital Microphone, Ambient Light Sensor |
2. 创建用户帐户
在等待开发套件期间，我们建议设置您的用户帐户。
Silicon Labs 帐户：
Silicon Labs 帐户：此帐户将为您提供访问我们的开发人员社区、入门指南、私有 GitHub 存储库和 Simplicity Studio 开发环境的权限。您可在此处创建帐户或验证对帐户的访问权限。
3. 设置开发环境
虽然我们知道您在选择开发环境时有很多选项，但我们相信 Simplicity Studio 才是适用于开发蓝牙设备的正确之选。原因如下：
- Simplicity Studio 包含编程器和调试器功能，因此您不必担心手动设置。
- 认识您购买的电路板，并确定您可以使用的示例应用。
需要帮助设置环境吗？我们的入门指南将让您立即启动和运行。
下载 Simplicity Studio v5 的完整在线安装程序版本：
系统要求
| Windows | Windows 10（64 位） Windows 11 |
| MacOS | 10.14 Mojave 10.15 Catalina* 11.x Big Sur* 12.x Monterey* *如果尝试使用 Keil 8051 或 IAR 工具链，请点击此处 |
| Linux | Ubuntu 20.24 LTS |
| CPU | 1 GHz 或更高 |
| 内存 | 1 GB RAM（8 GB 推荐用于无线协议开发） |
| 磁盘空间 | 最低 FFD 安装需要 600 MB 磁盘空间 支持无线动态协议时需要 7 GB |
4. 探索演示内容
这里列出了一些其他思路，只需对下方建议的参考示例应用进行修改，通过极少量的编码即可轻松将这些思路转变为现实产品。这些用例并非作为即用型演示内容而提供，而是为进一步的评估营造了适当环境。
Voice Control Light
Detects spoken keywords ""on" and "off" to turn on and off LED on board.
Suggested Kit: EFR32xG24 Dev Kit
Get up and running quickly with
pre-built application in 10 minutes.
Learn to create the ML application from
trained model in 30 minutes.
其他演示内容
Because starting application development from scratch is difficult, our Simplicity SDK comes with a number of built-in demos and examples covering the most frequent use cases.
Pac-Man
Play the popular Pac-Man game using keywords said out aloud – Go, Left, Right, Up, Down, Stop. The application uses keyword detection. Board can be controlled using Simplicity Studio. The demo is also available as part of Simplicity Studio.
推荐的套件：
Audio Classifier
This application uses TensorFlow Lite for Microcontrollers to classify audio data recorded on the microphone in a Micrium OS kernel task. The classification is used to control a LED on the board. The demo is also available as part of Simplicity Studio.
推荐的套件：
Magic Wand
This application demonstrates a model trained to recognize various hand gestures with an accelerometer. The detected gestures are printed to the serial port. The demo is also available as part of Simplicity Studio.
推荐的套件：
Blink
This application demonstrates a model trained to replicate a sine function. The model is continuously fed with values ranging from 0 to 2pi, and the output of the model is used to control the intensity of an LED. The demo is also available as part of Simplicity Studio.
推荐的套件：
1. Build Model
Already have your .tflite file ready to go? Skip to the next step: “Test and Validate” .
Train your model and prepare it for conversion into a deployable format.
If you are familiar with ML development follow these steps –
Customized Code
Begin by designing and training your AI/ML model. This involves gathering and preprocessing data, selecting appropriate model, and setting up training parameters.
To help you build your model from scratch, we provide a Python package with command-line utilities and scripts to assist you with building your own model.
Refer to the TensorFlow documentation for support building on the Machine Learning model. Refer to LiteRT documentation for support on converting the model to .tflite
If you are new to ML development follow these steps –
Low Code
We've partnered with top AI platforms to help you design and build models with minimal coding. These platforms provide user-friendly GUI and automated workflows to simplify the process.
If you are looking for a pre-built Machine Learning solution - jump to the last tab "Pre-Built Solution"
2. Test and Validate
Evaluate your model's performance against the embedded target, validate the model to ensure it meets required performance metrics.
Optional Tool: MLTK Model Profiler
The MLTK model profiler provides information about how efficiently a model may run on an embedded target. The model profiler allows for executing a .tflite model file in a simulator or on a physical embedded target.
Note: This tool is optional and not officially supported by Silicon Labs, yet.
3. Deploy Model
Integrate and deploy your validated model onto the embedded device.
- Add AI/ML SDK Extension
- Configure TensorFlow Micro Component in Studio: set up the component to select the correct kernel for your embedded device
- Include and Run the Model: copy your .tflite model into your application into the config folder in your Simplicity Project
- Implement Post-Processing: add any necessary post-processing steps to handle model’s output and integrate it with your application’s logic
Turn Key Solutions
Pre-built, ready-to-deploy AI/ML solutions on Silicon Labs SoCs that simplify the development process and accelerate time-to-market.
设计合作伙伴
Silicon Labs has pre-screened and certified the following third-party AI/ML design service companies
to help you design and develop your customized AI/ML solution.
开始
1. Buy Kits
2. Create User Account
3. 开发环境
4. 探索演示内容
Build Your Own Solution
1. Build Model
2. Test and Validate
Deploy Model
Pre-Built Solution
合作伙伴
