Hardware Kits

This section introduces the four hardware platforms selected for the TinyML curriculum. Each platform represents a different point along the spectrum of embedded computing capabilities, from ultra-low-power microcontrollers to full-featured edge computers. These platforms illustrate distinct engineering trade-offs in power consumption, computational capability, and development complexity.

The selected platforms are widely used in commercial applications, thereby ensuring that the skills developed through these exercises translate directly to embedded systems development.

System Requirements and Prerequisites

Before selecting a hardware platform, ensure your development environment meets the following requirements:

Development Computer Requirements:

  • Operating System: Windows 10/11, macOS 10.15+, or Linux (Ubuntu 18.04+)
  • Memory: 8GB RAM minimum (16GB recommended for Raspberry Pi development)
  • Storage: 10GB free space for development tools and libraries
  • USB Ports: At least one USB 2.0/3.0 port for device connection
  • Internet Connection: Required for software installation and library downloads

Software Prerequisites:

  • Arduino IDE 2.0+ for Arduino-based platforms
  • Python 3.8+ for Raspberry Pi development
  • Git for version control and example code access
  • Text Editor/IDE (VS Code, PyCharm, or similar)

Hardware Accessories:

  • USB-C or Micro-USB cables (data transfer capable, not power-only)
  • SD Card (32GB+ Class 10) for Raspberry Pi
  • Power adapters appropriate for each platform
  • Camera modules (included with most kits or available separately)

Hardware Platform Overview

Our curriculum features four carefully selected platforms that span the full spectrum of embedded computing capabilities. Each platform shown in Table 1 has been chosen to illustrate specific engineering trade-offs and learning objectives.

Table 1: Platform selection strategy table.
Platform Primary Learning Focus Cost Power Profile Best For
XIAOML Kit IoT & Wireless ML $15-50 Low Power Cost-sensitive deployments
Arduino Nicla Ultra-low Power Design $120 Ultra-low Battery-powered devices
Grove Vision AI Hardware Acceleration $30 Medium Industrial applications
Raspberry Pi Full ML Frameworks $60-150 High Advanced edge computing

Platform Comparison

Table 2 provides a comprehensive technical comparison of all four platforms.

Table 2: Platform comparison matrix.
Characteristic XIAOML Kit Raspberry Pi Arduino Nicla Grove Vision AI V2
Cost Range (USD) $15-50 $60-150 $120 $30
Power Consumption Low High Ultra-low Medium
Processing Power Medium Very High Low High (NPU)
Memory Capacity 8MB 1-16GB 2MB 16MB
Primary Use Case IoT networks Edge computing Battery devices Industrial AI
ML Framework TF Lite TensorFlow, PyTorch TensorFlow Lite SenseCraft AI
Development Env. Arduino/ PlatformIO Python/Linux Arduino IDE Visual/Code

Platform Selection Guidelines

Selecting the appropriate platform depends on specific learning objectives and project requirements. Table 3 provides a systematic mapping to guide these decisions.

Table 3: Platform capabilities matrix.
Learning Objective/Application XIAOML Kit Ras Pi Arduino Nicla Grove Vision AI V2
Embedded Systems Basics βœ“ Limited βœ“ βœ“
Wireless Connectivity βœ“ βœ“ βœ“
Ultra-Low Power Design βœ“
Full ML Frameworks βœ“
Hardware Acceleration βœ“
Real-time Vision Limited βœ“ βœ“ βœ“
Edge-Cloud Integration βœ“ βœ“ βœ“
Production Deployment βœ“ βœ“ βœ“

Hardware Platform Specifications

This section provides detailed technical specifications for each platform, including processor architecture, memory hierarchy, sensor capabilities, and development toolchain requirements.

XIAOML Kit (Seeed Studio)

Best For: IoT & Wireless ML

The XIAOML Kit excels at wireless connectivity and cost-sensitive deployments. It’s perfect for learning IoT sensor networks, remote monitoring systems, and wireless ML inference where you need reliable connectivity in a compact, affordable package.

The XIAO ESP32S3 represents the category of ultra-compact, wireless-enabled microcontrollers optimized for IoT applications. The name β€œXIAO” (小) translates to β€œtiny” in Chinese, reflecting the board’s 21Γ—17.5mm form factor.

XIAO ESP32S3 development board

XIAO ESP32S3 development board

Processor Architecture: ESP32-S3 dual-core Xtensa LX7 running at 240MHz

Memory Hierarchy: 8MB PSRAM and 8MB Flash storage

Connectivity: WiFi 802.11 b/g/n and Bluetooth 5.0

Integrated Sensors: OV2640 camera sensor, digital microphone, 6-axis inertial measurement unit

Power Characteristics: 3.3V operation with multiple low-power modes

Development Environment: Arduino IDE and PlatformIO support with extensive library ecosystem. Supports C/C++ programming with Arduino-style abstractions and direct ESP-IDF for advanced users.

Application Focus: IoT sensor networks, remote monitoring systems, wireless ML inference, cost-sensitive deployments

Arduino Nicla Vision

Best For: Ultra-Low Power Design

The Arduino Nicla Vision is optimized for battery-powered devices and always-on sensing applications. It’s ideal for learning ultra-low power design, image classification systems, and object detection applications where battery life is measured in months, not hours.

The Nicla Vision exemplifies professional-grade embedded vision systems built around the STM32H7 microcontroller. This platform demonstrates how specialized hardware design enables sophisticated ML inference within severe resource constraints.

Arduino Nicla Vision with camera module

Arduino Nicla Vision with camera module

Processor Architecture: STM32H747 dual-core ARM Cortex-M7/M4 running at 480MHz

Memory Hierarchy: 2MB integrated RAM and 16MB Flash storage

Integrated Sensors: GC2145 camera sensor, MP34DT05 digital microphone, 6-axis IMU

Power Characteristics: 3.3V operation optimized for battery-powered deployment

Development Environment: Arduino IDE and OpenMV IDE support with specialized computer vision libraries. MicroPython support for rapid prototyping alongside C/C++ for production deployments.

Application Focus: Battery-powered devices, image classification systems, object detection applications, always-on sensing

Grove Vision AI V2

Best For: Hardware Acceleration

The Grove Vision AI V2 features dedicated neural processing hardware for orders-of-magnitude performance improvements. It’s perfect for learning industrial inspection systems, real-time video analytics, and advanced object detection where you need NPU-accelerated inference capabilities.

The Grove Vision AI V2 incorporates dedicated neural processing hardware (NPU) to demonstrate hardware-accelerated ML inference. This platform illustrates how specialized AI processors achieve orders-of-magnitude performance improvements over software-only implementations.

Grove Vision AI V2 with NPU

Grove Vision AI V2 with NPU

Processor Architecture: ARM Cortex-M55 with integrated Ethos-U55 NPU

Memory Hierarchy: 16MB external memory for model and data storage

Neural Processing Unit: Dedicated hardware accelerator for ML inference

Camera Interface: Standard CSI connector supporting various camera modules

Audio Input: Onboard digital microphone

Development Environment: SenseCraft AI visual programming platform for no-code development, with Arduino IDE support for custom applications. Supports both graphical programming and traditional C/C++ development workflows.

Application Focus: Industrial inspection systems, real-time video analytics, advanced object detection, NPU-accelerated inference

Raspberry Pi (Models 4/5 and Zero 2W)

Best For: Full ML Frameworks

The Raspberry Pi bridges embedded systems and traditional computing, providing a complete Linux environment for advanced ML applications. It’s ideal for learning edge AI gateways, advanced computer vision systems, language model deployment, and multi-modal AI applications where you need full computing capabilities.

The Raspberry Pi family bridges embedded systems and traditional computing, providing a full Linux environment while maintaining educational accessibility. This platform demonstrates how increased computational resources enable sophisticated ML applications.

Raspberry Pi 5 and Pi Zero 2W comparison

Raspberry Pi 5 and Pi Zero 2W comparison

Processor Architecture: ARM Cortex-A76 (Pi 5) or Cortex-A53 (Zero 2W)

Memory Hierarchy: 1-16GB DDR4 RAM depending on model

Storage: MicroSD card primary storage with USB 3.0 expansion

Connectivity: Gigabit Ethernet, WiFi, Bluetooth, multiple USB ports

Camera Interface: Dedicated CSI connector plus USB camera support

Operating System: Debian-based Raspberry Pi OS (full Linux distribution)

Development Environment: Full Linux development environment with native Python, C/C++, and JavaScript support. Package managers (apt, pip) provide access to extensive ML libraries including TensorFlow, PyTorch, and OpenCV.

Application Focus: Edge AI gateways, advanced computer vision systems, language model deployment, multi-modal AI applications

Getting Started

To get started with the hardware kits used in this course, you can purchase them directly from the following official sources:

Check each site for educational discounts, bundles, and regional availability. Most kits are available as starter packages that include the board and basic accessories.

Back to top