6.3. C2000 Family

The C2000 family of digital signal processors (DSPs) is TI’s flagship platform for real-time industrial control applications.

6.3.1. Overview

C2000 devices feature:

High-performance DSP cores optimized for control algorithms
Integrated analog peripherals (ADCs, comparators)
PWM modules for motor control
Rich communication interfaces

These capabilities make C2000 ideal for predictive maintenance and real-time fault detection where ML inference runs alongside control loops.

6.3.2. Supported Devices

NPU-Enabled

Device	Core	Frequency	Features
F28P55	C28x + NPU	150 MHz	TINPU neural accelerator, ideal for complex models

C28x Core Devices

Device	Core	Frequency	Features
F28P65	C28x	150 MHz	High-performance, no NPU
F2837	Dual C28x	200 MHz	Dual-core, legacy support
F28004	C28x	100 MHz	Mid-range, motor control
F28003	C28x	100 MHz	Cost-optimized
F280013	C28x	100 MHz	Entry-level
F280015	C28x	120 MHz	Enhanced entry-level

C29x Core Devices (64-bit)

Device	Core	Frequency	Features
F29H85	C29x	300 MHz	64-bit, high performance
F29P58	C29x	300 MHz	64-bit, balanced
F29P32	C29x	300 MHz	64-bit, cost-optimized

6.3.3. F28P55 (Recommended)

The F28P55 is the recommended device for Tiny ML applications due to its integrated TINPU (Neural Processing Unit).

Key Specifications:

150 MHz C28x DSP core
TINPU neural accelerator
512 KB Flash
128 KB RAM
16-bit and 12-bit ADCs
Type-4 CLB (Configurable Logic Block)

NPU Capabilities:

8-bit and 4-bit quantized inference
Hardware-accelerated convolutions
Up to 25x faster than CPU inference
Support for models up to ~60k parameters

Configuration:

common:
  target_device: 'F28P55'

training:
  model_name: 'CLS_4k_NPU'  # Use NPU-compatible models

compilation:
  preset_name: 'compress_npu_layer_data'

6.3.4. F28P65

The F28P65 offers similar performance to F28P55 but without the NPU.

When to Use:

Existing F28P65 designs
Models that don’t fit NPU constraints
Projects not requiring fastest inference

Configuration:

common:
  target_device: 'F28P65'

training:
  model_name: 'CLS_4k'  # Standard models (no _NPU suffix)

6.3.5. F2837

Legacy dual-core device for existing systems.

Features:

Dual C28x cores at 200 MHz
Large memory (1 MB Flash)
CLA (Control Law Accelerator)

Configuration:

common:
  target_device: 'F2837'

6.3.6. C29x Family (F29H85, F29P58, F29P32)

The newest C2000 generation with 64-bit C29x cores.

Features:

64-bit native architecture
300 MHz operation
Enhanced floating-point performance
Backward compatibility with C28x code

Configuration:

common:
  target_device: 'F29H85'  # or F29P58, F29P32

6.3.7. Typical Applications

C2000 devices excel in these ML applications:

Motor Fault Detection

Bearing fault classification
Winding fault detection
Vibration anomaly detection

common:
  task_type: 'generic_timeseries_classification'
  target_device: 'F28P55'

dataset:
  dataset_name: 'motor_fault_classification_dsk'

Arc Fault Detection

DC arc fault in solar/battery systems
AC arc fault in building wiring

common:
  task_type: 'generic_timeseries_classification'
  target_device: 'F28P55'

dataset:
  dataset_name: 'dc_arc_fault_example_dsk'

Power Quality Monitoring

Harmonic analysis
Transient detection
Grid fault classification

6.3.8. Development Tools

Code Composer Studio (CCS)

TI’s official IDE for C2000 development:

Download from: https://www.ti.com/tool/CCSTUDIO
Version 12.x or later recommended
Install C2000 device support

C2000WARE SDK

Software development kit with:

Device drivers
Example projects
Documentation

controlSUITE

Legacy SDK (use C2000WARE for new projects)

6.3.9. Memory Considerations

C2000 devices have different memory architectures:

Device	Flash	RAM	Max Model Size
F28P55	512 KB	128 KB	~60k params (NPU)
F28P65	512 KB	128 KB	~20k params (CPU)
F2837	1 MB	200 KB	~30k params (CPU)

Choose model sizes that fit within device constraints:

# F28P55 with NPU - can use larger models
training:
  model_name: 'CLS_13k_NPU'

# F28004 - use smaller models
training:
  model_name: 'CLS_1k'

6.3.10. Next Steps

Review NPU Guidelines for F28P55 model constraints
See CCS Integration Guide for device programming
Try Arc Fault Detection example on C2000