Firefly Launches RK3576-Powered Mini AI SBCs

Firefly CAM-3576Q38 Mini AI SBC

For AIoT, edge AI, smart vision, industrial, and automotive applications, Firefly Technology has released the CAM-3576 series of mini (38 × 38 mm) SBCs based on the Rockchip RK3576 processor with a 6 TOPS NPU. The CAM-3576Q38 (commercial), CAM-3576JQ38 (industrial), and CAM-3576MQ38 (automotive) modules are available in three distinct configurations and are intended for use with dash cams, smart cameras, intelligent security systems, and private on-device AI model deployment. It incorporates a 6 TOPS computing NPU and has the Rockchip octa-core 64-bit AIoT processor RK3576, which enables the private deployment of popular large AI models and different deep learning frameworks. Additionally, it offers 4K@120fps high-definition, high-frame-rate display capabilities, 4K@60fps encoding, and 4K@120fps decoding.

Its 145G FLOPS Mali-G52 MC3 GPU allows for effective heterogeneous computing, satisfying the needs of graphics-intensive applications. Up to 16GB of LPDDR5 RAM, up to 256GB of eMMC flash, and a microSD card for expansion are all supported by the CAM-3576 series. The boards also have Fast Ethernet, Wi-Fi 6, USB 2.0, USB-C (device), RS-485, UART, I2C, ADC, GPIOs, audio input/output, RTC support, and a MIPI CSI input for up to 16MP camera sensors with HDR support.

Previously, Firefly introduced the ROC-RK3506J-CC industrial SBC and the ROC-RK3588S-PC credit-card-sized SBC. The CAM-3576Q38 is its first product based on the Rockchip RK3576 processor.

ICORE-3576Q38 SOM and CAM-3576Q38 Mini AI SBC

Firefly CAM-3576Q38 specifications:

• SoM: iCore-3576Q38
  • SoC: Rockchip RK3576 (Q38 – Commercial) / RK3576J (JQ38 – Industrial) / RK3576M (MQ38 – Automotive)
    • CPU: Octa-core 64-bit processor with 4× Cortex-A72 @ up to 2.2 GHz and 4× Cortex-A53 @ up to 2.0 GHz (1.6 GHz for industrial/automotive)
    • GPU: Arm Mali-G52 MC3 with OpenGL ES 1.1/2.0/3.2, OpenCL 2.0, Vulkan 1.1, 2D acceleration
    • NPU: 6 TOPS INT8 AI accelerator with INT4/INT8/INT16/BF16/TF32 support
    • Video Decoder: 8K@30 fps / 4K@120 fps (VP9, AVS2, AV1), 4K@60 fps (H.264/AVC)
    • Video Encoder: H.264, H.265, MJPEG up to 4K@60 fps
    • System Memory: 4, 8, or 16 GB LPDDR4 / LPDDR4x / LPDDR5
    • Storage: eMMC 16 GB to 256 GB, optional UFS 2.0
• Removable Storage: MicroSD card slot
• Host Interface: 2× 100-pin high-density board-to-board connectors
• Display Support: HDMI / eDP / DP / MIPI DSI (via carrier board)
• Camera Interface: 40-pin 0.5 mm MIPI-CSI DPHY FPC (1×4-lane or 2×2-lane)
• Audio:
  • Audio Input: 1× MIC (2-pin, 1.25 mm)
  • Audio Output: 1× Speaker (4-pin, 1.25 mm, 2×10 W / 6 Ω or 8 Ω)
  • Audio Interface: 4× SAI (4T/4R), I2S / TDM / PCM, up to 192 kHz
• Networking:
  • Ethernet: 10/100 Mbps via 8-pin connector with onboard S18610G PHY
  • Wireless: Dual-band Wi-Fi 6 (2.4 GHz / 5 GHz, 802.11ax) with antenna connector
• USB:
  • USB Type-C (device mode)
  • 1× USB 2.0 via 24-pin 0.5 mm FPC
• Serial: 1× RS-485 (8-pin 1.25 mm), 1× UART (expansion header)
• Expansion Header: 24-pin FPC with USB 2.0, ADC, I2C, UART, 10× GPIO, 5 V output
• Buttons: Reset, MaskROM
• RTC: Battery holder (2-pin 1.25 mm)
• Debug: 3-pin debug header, 2-pin recovery header
• Power Input: 12 V DC via 8-pin 1.25 mm connector
• Dimensions: 38.09 × 38.11 × 11.54 mm
• Weight: ~21 g
• Operating Humidity: 10–90% RH (non-condensing)
• Operating Temperature (Commercial): −20 °C to 60 °C
• Operating Temperature (Industrial/Automotive): −40 °C to 85 °C

CAM-3576Q38 38mm Mini SBC Interfaces

The CAM-3576Q38 is compatible with standard Linux and Buildroot for embedded product development, as well as RTLinux for real-time networking, hardware resource isolation, Flexbus, and DSMC. Additionally, it supports common AI frameworks like TensorFlow, TensorFlow Lite, PyTorch, ONNX, Caffe, and Darknet with support for custom operators, as well as Docker container management and the import and export of RKNN models. Additionally, it facilitates the deployment of real-time AI workloads like YOLO for on-device inference and Transformer-based large models. It was found on the Wiki that Firefly’s Linux software stack, complete with SDKs, BSPs, and documentation, supports the SBC. Firefly’s resource downloads page has more details, including source code, tools, and reference materials. 

iCore-3576Q38 Industrial SOM

AliExpress offers the CAM-3576Q38 RK3576 tiny AI SBCs in commercial, industrial, and automotive versions. The 4GB/64GB commercial model costs $182.36, the industrial model costs $193.45, the automotive model costs $232.26, and the 8GB/64GB industrial version costs $304.36. Taxes are added at checkout, but bulk orders (10+ units) receive an additional 2% discount. The Firefly store also sells the ICORE-3576Q38 38mm System On Module for $139 and the commercial version of the “mini single board computer” (4GB/64GB) for $159. Users can find additional information on the product page. 

Images used courtesy of firefly