6.6. Connectivity, AM26x, MSPM33 and Radar Devices

This page covers the device families that do not have a page of their own: the SimpleLink connectivity MCUs (CC27xx, CC13xx, CC35xx), the AM26x Arm Cortex-R5F family, the MSPM33C family, and the IWRL6432 radar sensor. For the per-device task matrix, SDK example and memory-estimation support see Device Overview. AM13E2 is listed in Device Overview as well; it has a hardware NPU (see NPU Guidelines).

Note

The tinyml-sdk repository ships on-device AI examples for C2000 (c28, c29), AM13 and AM26 only. There are no Tiny ML ModelZoo SDK examples for the CC, MSPM33 or IWRL6432 devices. For those devices ModelMaker produces the compiled model library and golden vectors, and you integrate them into your own SimpleLink / MSPM33 / mmWave project. The hardware numbers for the CC devices in the 1.4.0 guide (flash, RAM, energy per inference, recommended model sizes) could not be verified against 1.5.0 sources and are not repeated here; use the device data sheets.

6.6.1. Connectivity Devices (CC)

No CC device has a hardware NPU (only F28P55, MSPM0G5187, AM13E2 and MSPM33C34 do). They use one of two compilation styles, set in tinyml-modelmaker (timeseries/constants.py).

target_device

Compiler core flags

Compilation style

CC2755, CC2745

Cortex-M33, thumbv8.1-m.main+cdecp0

Default preset uses the Arm Custom Datapath Extension (CDE), ti-npu type=m33cde. A software-only (“soft”) preset is also defined.

CC35X1

Cortex-M33, thumbv8.1-m.main+cdecp0

Same as CC2755: CDE preset by default, soft preset available.

CC1354, CC1314

Cortex-M33, armv8-m.main, FPv5-SP

Software (soft) preset only.

CC1352, CC1312

Cortex-M4F, armv7e-m, FPv4-SP

Software (soft) preset only.

Note

The 1.4.0 guide listed CC2755, CC1352, CC1354 and CC35X1 (as a Cortex-M4 Wi-Fi part). In 1.5.0 the device list also contains CC2745, CC1312 and CC1314, and the compiler flags for CC35X1 are those of a Cortex-M33 with CDE. The CC devices are listed by classification, motor fault and PIR detection models (see Device Overview); regression, forecasting and anomaly-detection models do not list them.

Example configuration:

common:
  task_type: 'generic_timeseries_classification'
  target_device: 'CC2755'

compilation:
  enable: True

Choose a model that lists your device; ModelMaker rejects a model that does not.

6.6.2. AM26x (AM263, AM263P, AM261)

Arm Cortex-R5F devices (VFPv3-D16 floating point), compiled without an NPU. AM263 supports classification, arc fault, motor fault, blower imbalance, regression, forecasting and anomaly detection models; AM263P and AM261 support classification, regression, forecasting and anomaly detection (see Device Overview). The tinyml-sdk folder am26/ai/examples contains the same twelve example projects as the AM13 folder (arc_fault, blower_imbalance, fan_blade_anomalydetection, forecasting_pmsm_rotor_temp, generic_timeseries_*, hvac_indoor_temp_forecast, motor_fault, torque_measurement, washing_machine_load_weighing). Deployment follows Non-NPU Deployment.

6.6.3. MSPM33C (MSPM33C32, MSPM33C34)

Arm Cortex-M33 devices. MSPM33C32 is a CPU-only target (MSPM33 SDK mspm33c321a). MSPM33C34 has an NPU compilation profile and is accepted on the command line, but no 1.5.0 model lists it, so every task is rejected for it. Use MSPM33C32 for MSPM33 projects.

6.6.4. IWRL6432 (Radar)

The IWRL6432 mmWave radar sensor (Arm Cortex-M4F, no NPU) is the target of the radar_classification task and the only device that task lists in 1.5.0.

6.6.5. Where to Go Next