Skip to content

Profiling

Use the Profiling feature in CodeFusion Studio to either capture profiler traces from your Zephyr project or prepare CTF traces for visualization. The profiler also supports inference-level profiling for TFLM models, allowing detailed analysis of AI workloads.

Beta feature

The Profiling tool is currently in beta. Some options or trace formats may change in future releases.

About Zephelin

Zephelin is a middleware layer developed by Antmicro for Analog Devices. It extends Zephyr RTOS with advanced tracing, profiling, and runtime metric capture. Zephelin records system-level events such as task switches, interrupts, and timing information, and can also capture AI operator-level performance data when using TFLM.

Before you start

Before you begin, ensure you have created a Zephyr workspace using the Workspace Creation Wizard. For example, create a MAX32690 workspace using the Zephyr Single Core TensorFlow AI Model template.

If you create a workspace from a non-AI template (for example, Zephyr Single Core Blinky), you can still capture basic system traces. However, for more detailed profiling — including AI inference timing — add a model configuration as described in Embedded AI Tools.

Configure and capture a profiling trace

All build, flash, and profiling operations in CodeFusion Studio are performed using CFS tasks. You can run these tasks from the Actions view, Command Palette, or other locations described in Build, clean, flash, and debug tasks.

Step 1. Configure profiling options

Profiling tab in System Planner Profiling tab in System Planner

  1. Open System Planner in your Zephyr workspace.
  2. Click the Profiling tab.
  3. Configure the following options:
    • Enable Zephelin Profiler – Records Zephyr runtime activity such as task scheduling and interrupts.
    • Trace Interface – Select the UART interface used to stream trace data (options depend on your SoC).
    • Enable AI Model Profiling – Enables TensorFlow Lite Micro inference tracing for supported processors. For details, see Supported processors and model formats.

Step 2. Generate, build, and flash

  1. In System Planner, click Generate Code.
  2. Enable Generate AI Models if your project includes an AI model.
  3. Run the following CFS tasks:
    • CFS: build – Compiles your project (for example, creates m4/build/zephyr.elf).
    • CFS: flash or CFS: flash and run – Programs your board using JTAG or OpenOCD.

Step 3. Capture the profiler trace

  1. Run the Capture Profiler Trace task.

    Zephelin Capture Profiler Trace taskZephelin Capture Profiler Trace task

    Note

    Close any serial monitor or terminal before running Capture profiler trace — the task needs exclusive access to the UART port.
    If another process is using the same device, the capture will fail with the error:
    serial.serialutil.SerialException: device reports readiness to read but returned no data

    1. When prompted:

    2. Enter your serial port (for example, /dev/ttyUSB0 or COM3).

      Note

      Ensure the serial port name is entered exactly as it appears in your system, without any trailing spaces or extra text. For example: /dev/tty.usbserial-0001

    3. Select a baud rate (115200).

    4. Specify an output file name, for example zephelin_trace.
    5. When asked Send enable command to device before collecting data?, select No (recommended).
      • Choose No if your application enables tracing automatically at startup (default).
      • Choose Yes only if your project supports host-controlled tracing (CONFIG_TRACING_HANDLE_HOST_CMD=y).

  2. Monitor the Terminal output — live trace data will appear.

  3. If your application finishes execution quickly (for example, the MAX32690 Single Core TensorFlow AI Model example), press Reset on your board.
  4. Press Ctrl + C to stop capture.
  5. The generated trace file appears in your project folder in the Explorer view.

Sample Zephelin AI profiling output

***Booting Zephyr OS build 7ec15cb5b25a***
sys_trace_k_thread_info: 0x20002428
zpl_inference_enter:
zpl_tflm_enter_event: subgraph_idx=0 op_idx=0 tag=FULLY_CONNECTED arena_used_bytes=840 arena_tail_usage=88
zpl_tflm_exit_event: subgraph_idx=0 op_idx=0 tag=FULLY_CONNECTED arena_used_bytes=840 arena_tail_usage=88
zpl_inference_exit:
...

Note

The output format may vary depending on the selected trace mode and platform configuration.

Switch to CTF mode

By default, the profiler uses Text output. To capture traces in CTF for later visualization:

  1. Add the following line to your prj.conf or boards/<board>.conf file:

    CONFIG_ZPL_TRACE_FORMAT_CTF=y

  2. Re-run the CFS: build and CFS: flash tasks to to rebuild and re-program your project.

  3. Follow the steps in Step 3. Capture the profiler trace to run the Capture Profiler Trace task again and record a new trace in CTF mode.

  4. When a trace is captured in CTF format, Zephelin automatically creates two output files in your project’s directory:

    • <output_name>_0 – the binary trace file containing the captured CTF data. This file is used later for TEF conversion.
    • <output_name> – a readable text log showing application output, such as test results or boot messages.

    For example, if you entered zephelin_trace as your output file name, Zephelin will create the following files:

    • zephelin_trace_0 – binary trace data (for TEF conversion)
    • zephelin_trace – readable application log

    This behavior is implemented by Zephelin, including the automatic _0 suffix for the binary trace file. Always select the file with _0 appended when converting to TEF format.

Sample log output

Test 0 passed: pred=0.026405 calc=0.000000
Test 1 passed: pred=0.863044 calc=0.841471
Done – Success
***Booting Zephyr OS build 7ec15cb5b25a***

Prepare CTF trace for visualization

  1. Run the Prepare CTF Trace for Visualization task.
  2. Choose the generated CTF file (for example, zephelin_trace_0).
  3. Specify an output file path for the .tef file.
  4. Open the Zephelin Trace Viewer.
  5. Upload your .tef file to visualize the captured trace data.

Zephelin Trace Viewer showing layer-level profiling results for a TensorFlow Lite Micro model

Zephelin documentation and version information

You can access the official Zephelin documentation at antmicro.github.io/zephelin. The site reflects the latest Zephelin release, which may differ slightly from the version integrated in CodeFusion Studio. CodeFusion Studio 2.0.0 uses the version from September 25, 2025, and the corresponding documentation can be found in zephelin/docs. The documentation describes advanced configuration options—such as memory profiling and tracing scopes—that may also be enabled in CodeFusion Studio through project configuration.