Silicon Labs ML SDK Version 2.2.0 - Release Notes (Jan 22, 2026)#

Simplicity SDK Version 2025.12.0

The Silicon Labs ML SDK is provided as an extension to the Simplicity SDK. It enables AI/ML development on Series 2 (EFR and SiWG917) devices using the Tensorflow Lite for Microcontrollers (TFLM) framework.

Click here for earlier releases.

Release Summary#

Key Features | API Changes | Bug Fixes | Chip Enablement

Key Features#

  • New ML Model Profiler helps developers understand the execution performance of their ML models on the target device and within existing applications. It is available in 2 formats:

    • ML Model Profiler Tool: run .tflite model on Series 2 devices to analyze performance before integration.

    • ML Model Profiler Component: integrate profiling directly within an existing application to capture model performance during normal execution.

  • New sample applications are available for Profiling ML Models, Data Capture from IMU, and Device Tampering Detection.

  • Changed sample application names.

  • Demos for more EFR and SiWG917 boards.

API Changes#

None.

Bug Fixes#

  • Blink app behaves differently when running on Series 2, compared to SiWG917.

  • Blink and Voice Control applications now support BRD4338a and similar SiWG917 radio boards.

  • Flatbuffer converter tool documentation updated to reflect latest changes.

  • Misleading error on macOS for applications built using software optimizations for ML models.

  • Removed invalid link to I2S pin configuration for SiWG917 voice-based applications.

Chip Enablement#

  • EFR32MG24B310F1536IM48

  • EFR32MG24B210F1536IM48

  • EFR32MG24B220F1536IM48

  • EFR32MG26B510F3200IM68

  • EFR32MG26B410F3200IM48

  • EFR32MG26B420F3200IM48

  • EFR32MG26B510F3200IL136

  • EFR32MG26B410F3200IM68

  • EFR32MG26B420F3200IM68

  • EFR32MG26B510F3200IM48

  • EFR32ZG28B312F1024IM48

  • EFR32ZG28B312F1024IM68

  • EFR32ZG28B322F1024IM68

  • EFR32ZG28B312F1024IM68

  • SiWG917M111MGTBA

  • SiWG917M141XGTBA

  • SiWG917Y111MGNBA

  • SiWG917Y111MGAB4

  • SiWG917Y121MGNB4

Key Features#

New Features | Enhancements | Removed Features | Deprecated Features

Note: See Feature Matrix for a list of any applicable APIs, examples, software variants, modes, hardware, and host interfaces for each feature.

New Features#

  • New tool, ML Model Profiler, to profile .tflite models. For details, see documentation here. This tool currently only supports EFR devices from the Series 2 family. SiWG917 support is forthcoming.

  • New component, ML Model Profiler, to enable profiling of .tflite models in existing customer applications. Adding the "ML Model Profiler" component to an application or the ml-profiler component to the .slcp file of an application will enable this feature. This is not be confused with the ML Model Profiler Tool. This component will eventually be a part of the tool as well. The purpose of keeping this component separate is to allow flexibility to profile ML models in user's existing applications alongside other stacks.

  • New sample applications:

    • AI/ML - ML Model Profiler Firmware: Profiling ML Models built for Tensorflow Lite for Microcontrollers.

    • AI/ML - SoC Anomaly Detection for EFR32: Device Tampering Detection by using IMU data and time-series analysis of that data.

    • AI/ML - SoC Data Capture for EFR32 Baremetal: Data Capture from IMU into a CSV file. This app requires JLink on the user's workstation.

Enhancements#

  • Changed sample app names from ml_<app_name>_<platform> pattern to aiml_[sub_technology_]<soc|rcp|ncp|host>_<app_name>_<platform>_<os|baremetal>. This was done to allow quicker segregation of apps based on their capabilities.

  • Pre-built demo binaries are now available for more EFR and SiWG917 boards. Newly added demos are:

    • BRD2601B

      • AI/ML - SoC Anomaly Detection EFR32 Baremetal

      • AI/ML - SoC Data Capture EFR32 Baremetal

      • AI/ML - SoC Profiler Firmware EFR32 Baremetal

    • BRD2608A

      • AI/ML - SoC Anomaly Detection EFR32 Baremetal

      • AI/ML - SoC Data Capture EFR32 Baremetal

      • AI/ML - SoC Profiler Firmware EFR32 Baremetal

    • BRD4186C

      • AI/ML - SoC Blink EFR32 Baremetal

      • AI/ML - SoC Model Profiler EFR32 Baremetal

    • BRD4187C

      • AI/ML - SoC Blink EFR32 Baremetal

      • AI/ML - SoC Model Profiler EFR32 Baremetal

    • BRD4338A

      • AI/ML - SoC Audio Classifier SiWG917 Baremetal

      • AI/ML - SoC Blink SiWG917 Baremetal

      • AI/ML - SoC Model Profiler SiWG917 Baremetal

      • AI/ML - SoC Voice Control Light SiWG917 Baremetal

    • BRD4339A

      • AI/ML - SoC Model Profiler SiWG917 Baremetal

    • BRD4340B

      • AI/ML - SoC Model Profiler SiWG917 Baremetal

    • BRD4342A

      • AI/ML - SoC Model Profiler SiWG917 Baremetal

    • BRD4343A

      • AI/ML - SoC Model Profiler SiWG917 Baremetal

    • BRD4343B

      • AI/ML - SoC Model Profiler SiWG917 Baremetal

    • BRD4343Q

      • AI/ML - SoC Model Profiler SiWG917 Baremetal

Removed Features#

None.

Deprecated Features#

None.

API Changes#

New APIs | Modified APIs | Removed APIs | Deprecated APIs

New APIs#

None.

Modified APIs#

None.

Removed APIs#

None.

Deprecated APIs#

None.

Bug Fixes#

ID Issue Description GitHub / Salesforce Reference (if any) Affected Software Variants, Hardware, Modes, Host Interfaces
1567613 1463269 Blink app behaves differently when running on Series 2, compared to SiWG917. The LED has a baseline brightness on SiWG917 while blinking but it turns off completely on Series 2. This is due to the different LED driver implementations on the two chip families. None.
  • SiWG917M111MGTBA
  • BRD4338A
  • SoC
1479985 Blink and Voice Control applications now support BRD4338a and similar SiWG917 radio boards. None.
  • SiWG917M111MGTBA
  • BRD4338A
  • SoC
1438989 Flatbuffer converter tool documentation updated to reflect latest changes. None.
  • Documentation
1479970 Misleading error on macOS for applications built using software optimizations for ML models. None.
  • Silicon Labs ML SDK
1497650 Removed invalid link to I2S pin configuration for SiWG917 voice-based applications. None.
  • SiWG917M111MGTBA
  • SoC

Chip Enablement#

Chip Family OPNs / Boards / OPN Combinations Supported Software Variants (if applicable) Supported Modes Supported Host Interfaces
Chip
  • EFR32MG24B310F1536IM48
  • EFR32MG24B210F1536IM48
  • EFR32MG24B220F1536IM48
  • EFR32MG26B510F3200IM68
  • EFR32MG26B410F3200IM48
  • EFR32MG26B420F3200IM48
  • EFR32MG26B510F3200IL136
  • EFR32MG26B410F3200IM68
  • EFR32MG26B420F3200IM68
  • EFR32MG26B510F3200IM48
  • EFR32ZG28B312F1024IM48
  • EFR32ZG28B312F1024IM68
  • EFR32ZG28B322F1024IM68
  • EFR32ZG28B312F1024IM68
  • SiWG917M111MGTBA
  • SiWG917M141XGTBA
  • SiWG917Y111MGNBA
  • SiWG917Y111MGAB4
  • SiWG917Y121MGNB4
 Standard SoC
  • UART
  • SPI
  • I2S
  • I2C

Application Example Changes#

New Examples | Modified Examples | Removed Examples | Deprecated Examples

New Examples#

Example Name Description Supported Software Variants (if applicable) Supported Modes Supported OPNs / Boards / OPN Combinations Supported Host Interfaces

AI/ML - ML Model Profiler Firmware

 Profiling ML Models built for Tensorflow Lite for Microcontrollers. Standard SoC
  • OPN: EFR32xG2x, SiWG917
  • Boards: BRD2601B, BRD2608A, BRD4186C, BRD4187C, BRD4338A, BRD4339A, BRD4340B, BRD4342A, BRD4343A, BRD4343B, BRD4343Q
  • External Hosts: N/A

AI/ML - SoC Anomaly Detection for EFR32

 Device Tampering Detection by using IMU data and time-series analysis of that data. Standard SoC
  • OPN: EFR32xG2x
  • Boards: BRD2601B, BRD2608A
  • External Hosts: N/A

AI/ML - SoC Data Capture for EFR32 Baremetal

 Data Capture from IMU into a CSV file. This app requires JLink on the user's workstation. Standard SoC
  • OPN: EFR32xG2x
  • Boards: BRD2601B, BRD2608A
  • External Hosts: N/A

Modified Examples#

None.

Removed Examples#

None.

Deprecated Examples#

None.

Known Issues and Limitations#

Impact of Release Changes#

Impact Statements | Migration Guide

Impact Statements#

Change Impact Affected Software Variants if applicable Affected Modes Affected OPNs / Boards / OPN Combinations Affected Host Interfaces

Changed sample application names

 Sample application names have been changed from ml_<app_name>_<platform> pattern to aiml_[sub_technology_]<soc|rcp|ncp|host>_<app_name>_<platform>_<os|baremetal>. For example, ml_blink_efr32 is now aiml_soc_blink_efr32_baremetal. You must update any references to the old application names in your build scripts, documentation, or workflows. Standard SoC
  • OPN: EFR32xG2x, EFM32PG26, SiWG917
  • Boards: All supported boards
  • External Hosts: N/A

New ML Model Profiler tool

 A new tool, ML Model Profiler, is available to profile .tflite models on EFR devices. You can now analyze model performance directly on target hardware. See ML Model Profiler documentation for usage instructions. Standard SoC
  • OPN: EFR32xG2x
  • Boards: BRD2601B, BRD2608A, BRD4186C, BRD4187C
  • External Hosts: N/A

New ML Model Profiler component (part of ML Model Profiler Tool as well)

 A new component, ML Model Profiler, enables profiling of .tflite models in existing applications. You can add the "ML Model Profiler" component to an application or add the ml-profiler component to the .slcp file to enable this feature. This is not be confused with the ML Model Profiler Tool, this component eventually is part of the tool as well. The purpose of keeping this component separate is to allow flexibility to profile ML models in user's existing applications alongside other stacks. Standard SoC
  • OPN: EFR32xG2x
  • Boards: All supported boards
  • External Hosts: N/A

New sample applications

 Three new sample applications are available: AI/ML - ML Model Profiler Firmware (for profiling ML models), AI/ML - SoC Anomaly Detection for EFR32 (device tampering detection using IMU data), and AI/ML - SoC Data Capture for EFR32 Baremetal (data capture from IMU to CSV). You can use these as reference implementations. Standard SoC
  • OPN: EFR32xG2x
  • Boards: BRD2601B, BRD2608A
  • External Hosts: N/A

Expanded demo support

 Pre-built demo binaries are now available for more EFR and SiWG917 boards. You can run demos on additional hardware without building from source. Standard SoC
  • OPN: EFR32xG2x, SiWG917
  • Boards: BRD2601B, BRD2608A, BRD4186C, BRD4187C, BRD4338A, BRD4339A, BRD4340B, BRD4342A, BRD4343A, BRD4343B, BRD4343Q
  • External Hosts: N/A

Using This Release#

What's in the Release? | Compatible Software | Installation and Use | Help and Feedback

What's in the Release?#

This release introduces the ML Model Profiler tool and ML Model Profiler component for profiling .tflite models on EFR devices. Three new sample applications are included: ML Model Profiler Firmware for profiling ML models, SoC Anomaly Detection for device tampering detection using IMU data, and SoC Data Capture for capturing IMU data to CSV files. Sample application names have been updated to a new naming convention for better organization. Pre-built demo binaries are now available for additional EFR and SiWG917 boards. This release also includes bug fixes for the Blink app behavior, SiWG917 board support, macOS error messages, and documentation updates.

Compatible Software#

Software Compatible Version or Variant
Software Development Kit (SDK)

Installation and Use#

To upgrade your existing software with this release, update Simplicity Studio to the latest, Simplicity SDK to v2025.12.0, WiSeConnect SDK to v4.0.0, and ML SDK to v2.2.0 from Studio installation manager, or download the SDKs from the respective links listed in Compatible Software section above. To update the ML SDK, please refer to AI/ML SDK Setup guide.

To run your first demo, see our Getting Started

To kick start your development, see our Developer's Guide

For information about Secure Vault Integration, see Secure Vault.

To review Security and Software Advisory notifications and manage your notification preferences:

  1. Go to https://community.silabs.com/.

  2. Log in with your account credentials.

  3. Click your profile icon in the upper-right corner of the page.

  4. Select Notifications from the dropdown menu.

  5. In the Notifications section, go to the My Product Notifications tab to review historical Security and Software Advisory notifications

  6. To manage your preferences, use the Manage Notifications tab to customize which product updates and advisories you receive.

To learn more about the software in this release, dive into our online documentation

Help and Feedback#

Feature Matrix#

Supported Features | Unsupported Features

Supported Features#

Feature Name Description Quality Related API Names Supported Software Variants, Hardware, Modes, Host Interfaces Related Example Names
ML Model Profiler Tool Tool to profile .tflite models on EFR devices. Analyze model performance directly on target hardware. See documentation: ML Model Profiler documentation. Beta
  • Standard
  • EFR32xG2x
  • SoC
ML Model Profiler Component (part of ML Profiler Tool as well) Component to enable profiling of .tflite models in existing applications. Add the "ML Model Profiler" component or the ml-profiler component to the .slcp file. This is not be confused with the ML Model Profiler Tool. This component will eventually be a part of the tool. The purpose of keeping this component separate is to allow flexibility to profile ML models in user's existing applications alongside other stacks. Production
  • Standard
  • EFR32xG2x, SiWG917
  • SoC
 AI/ML - ML Model Profiler Firmware
Anomaly Detection Device tampering detection using IMU data and time-series analysis. Production
  • Standard
  • EFR32xG2x
  • SoC
 AI/ML - SoC Anomaly Detection for EFR32
Data Capture Capture IMU data into a CSV file for training and analysis. Requires JLink on the user's workstation. Production
  • Standard
  • EFR32xG2x
  • SoC
 AI/ML - SoC Data Capture for EFR32 Baremetal

Unsupported Features#

  • SiWG917 does not use the built-in TensorFlow component directly; instead, support is provided through automatic code generation handled by advanced configurators.

  • Series 2 devices do not support new software optimizations. However, due to their architecture, they are still at least as efficient as SiWG917.

SDK Release and Maintenance Policy#

See our SDK Release and Maintenance Policy.