Sipeed MaixCAM AI Vision Board (SG2002 RISC-V, 1 TOPS NPU, WiFi 6)

Specifications

• SoC: SOPHGO SG2002 (1GHz RISC-V / ARM A53, 700MHz RISC-V, 8051)

• NPU: 1 TOPS @ INT8 (Supports BF16, YOLO, MobileNet)

• Memory: 256MB DDR3

• Storage: TF (MicroSD) Card Boot / SD NAND

• Camera: Supports up to 5MP, includes 4MP MIPI CSI Camera

• Display: 2.3-inch HD IPS Capacitive Touchscreen (552x368)

• Audio: Built-in analog mic, PA amplifier (up to 1W speaker)

• Networking: WiFi 6 + Bluetooth 5.4

• USB: Type-C USB 2.0 (Host & Device mode)

• Dimensions (Case): 62.3mm x 44.3mm x 12.6mm (34.22mm depth with camera)

• GPIO Logic Level: 3.3V (Strictly NO 5V direct connections)

Downloads:

• Sipeed MaixCAM Development Board NPU - Buying Guide
• Sipeed MaixCAM Development Board NPU - Starter Guide 

Key Applications and Projects

• STEM Education: Using the intuitive MaixPy SDK and graphical MaixVision IDE to teach students how to train and deploy their first image classification models.

• Production Line Quality Inspection: Deploying traditional OpenCV algorithms alongside hardware-accelerated AI to instantly recognize defective parts on a high-speed conveyor belt.

• AI Algorithm Prototyping: Providing AI engineers with a cost-effective platform to quickly deploy and verify their trained AI models on real hardware using the provided SDKs.

Unlock New Capabilities (Project Evolution) The AI-Powered Sorting Robot: Stop relying on unreliable color sensors for sorting objects! By integrating the Sipeed MaixCAM with a standard 3D-printed robotic arm, you can utilize the onboard NPU to run a custom YOLOv8 object detection model. Train the model on the MaixHub online platform to recognize specific tools, fruits, or recycling materials. The MaixCAM will identify the objects via its 4MP camera, display the bounding boxes on its 2.3-inch touchscreen, and use its UART/I2C pins to send exact coordinates to the robotic arm's motor controller for perfectly accurate, automated sorting!

PACKAGE INCLUDED

• 1 x Sipeed MaixCAM Main Board (SG2002)

• 1 x 4MP Camera Module with FPC Cable

• 1 x 2.3-inch IPS Touchscreen (Pre-assembled)

• 1 x Black Protective Case (Pre-assembled)

• 1 x USB Type-C Cable

FREQUENTLY ASKED QUESTIONS (FAQ)

1. What processor does the MaixCAM use? It is powered by the SOPHGO SG2002 SoC, which includes a 1GHz RISC-V C906 big core, a 700MHz RTOS core, and an 8051 low-power core. The big core can alternatively run as a 1GHz ARM A53.

2. How much AI processing power does it have? The integrated Neural Processing Unit (NPU) delivers 1 TOPS (Tera Operations Per Second) at INT8 precision.

3. What kind of AI models can it run? It natively supports a wide variety of models, with rich operator support for architectures like MobileNetv2, YOLOv5, and YOLOv8.

4. How much memory is on the board? The MaixCAM features 256MB of DDR3 RAM.

5. How do I program the MaixCAM? You can program it using Python via the MaixPy SDK, graphically through the MaixVision IDE, or in C/C++ using the MaixCDK.

6. Does it have built-in wireless connectivity? Yes, the board features an onboard WiFi 6 and BLE 5.4 module for high-speed wireless networking.

7. Can I connect 5V sensors directly to the GPIO pins? No, the GPIO pins operate strictly at a 3.3V logic level. Connecting 5V directly will severely damage the CPU.

8. What resolution is the included screen? The rear-mounted 2.3-inch HD IPS capacitive touchscreen has a resolution of 552x368.

9. Can it record audio? Yes, it features a built-in analog silicon microphone for audio input and processing.

10. Can I connect a speaker to it? Yes, it features an onboard PA power amplifier that can directly drive a speaker up to 1W via the pin header.

11. What are the overall dimensions of the case? The main body of the case measures 62.3mm wide and 44.3mm tall.

12. How thick is the device with the camera installed? From the back of the case to the tip of the camera lens, the total depth is 34.22mm.

13. What is the spacing for the mounting holes? The two mounting holes on the bottom edge tabs are spaced exactly 44mm apart (center-to-center).

14. What size screws do the mounting holes take? The holes have a radius of 1.35mm (2.7mm diameter), making them perfect for standard M2.5 mounting screws.

15. How is the device powered? It is powered via the onboard USB Type-C port. Ensure your supply can provide at least 5V/2A to handle power spikes during intensive AI tasks and WiFi usage.

16. Does it support hardware video decoding? Yes, it supports H.264, H.265, and MJPEG hardware decoding (up to 2k@30fps).

17. What interfaces are available on the pin headers? The two 14-pin headers expose UART, I2C, SPI, ADC, PWM, WDT, and general GPIO pins.

18. Where do I store the operating system and AI models? The board boots and runs its Linux environment from a TF (MicroSD) card inserted into the onboard slot.

19. How do I train my own AI models for this? Sipeed offers the MaixHub online platform, which allows you to train custom AI models without needing local expensive hardware or deep AI expertise.

20. What is the purpose of the built-in push buttons? The board features two physical buttons: one is a Reset (RST) button, and the other is a programmable USER (function) button.