OAK-SoM-Pro¶

left: OAK-SoM-Pro-S3 (RVC3), right: OAK-SoM-Pro (RVC2)¶

Overview¶

The OAK-SoM-Pro is a System on Module (SoM) designed for integration into top-level systems with a need for a low-power, high performance, real-time AI, and depth perception. It is ideal for use in industrial automation, robotics, and security systems. The OAK-SoM-Pro interfaces with the system through two 10-Gbps-rated 100-pin board-to-board mezzanine connector which carries all signal I/O as well as 5v input. The on-board SMPS system regulates the 5V input and provides all necessary digital and analog power. An auxiliary power port is offered to interface without connection to a baseboard.

OAK-SoM-Pro comes in two variants: Robotics Vision Core 2 based and Robotics Vision Core 3 based SoM. Robotics Vision Core 3 provides quadcore ARM CPU, INT8 quantization, and enhanced stereo depth perception compared to the Robotics Vision Core 2.

OAK-SoM-Pro-S3 (Robotics Vision Core 3) was designed to be a drop-in replacement for OAK cameras that use OAK-SoM-Pro (Robotics Vision Core 2) (list below).

Devices that use OAK-SoM-Pro MX

All devices that have SoM on-board are also open-source. You could redesign them, for example to change stereo baseline. If you are interested in integrating DepthAI (via SoM) into your product, see documentation here.

Robotics Vision Core 2 inside:

This OAK device is built on top of the RVC2. Main features:

4 TOPS of processing power (1.4 TOPS for AI - RVC2 NN Performance)
Run any AI model, even custom architectured/built ones - models need to be converted.
Encoding: H.264, H.265, MJPEG - 4K/30FPS, 1080P/60FPS
Computer vision: warp/dewarp, resize, crop via ImageManip node, edge detection, feature tracking. You can also run custom CV functions
Stereo depth perception with filtering, post-processing, RGB-depth alignment, and high configurability
Object tracking: 2D and 3D tracking with ObjectTracker node

OAK-SoM-Pro-S3 has an Robotics Vision Core 3 and quadcore ARM CPU on board. It was designed to have the same dimensions and pinout as the OAK-SoM-Pro, so it can easily replace the OAK-SoM-Pro on OAK cameras that use it (list here).

Note that for OAK POE devices (OAK-D-PoE, OAK-1-PoE) this won’t work, as S1/S2 OAK POE devices use an PCIe to ETH bridge which isn’t (yet) supported by RVC3 (kernel). RVC3 has ETH PHY already on the SoC, so S3 PoE devices won’t use the same bridge.

It was primarily designed for evaluation and backwards compatibility. For new products, we would suggest using OAK-SoM MAX (differences below).

RVC3 inside

This OAK device is built on top of the RVC3. Main features:

3.0 TOPS for AI with INT8 quantization support
Quad-core ARM A53 @ 1.5GHz, running Yocto Linux, acting as a host computer
Imaging: ISP, max 6 cameras, 500 MP/s HDR, 3A
Run any AI model, even custom architectured/built ones - models need to be converted.
Cloud platform - Robothub - connectivity out-of-the-box
On-device SLAM / VIO support
Encoding: H.264, H.265, MJPEG - 4K/75FPS, Decoding: 4K/60FPS
Computer vision: warp/dewarp, resize, crop via ImageManip node, edge detection, feature tracking. You can also run custom CV functions
Stereo depth perception with filtering, post-processing, RGB-depth alignment, and high configurability
Object tracking: 2D and 3D tracking with ObjectTracker node

OAK-SoM-Pro-S3 vs OAK-SoM-Max

These are both our initial SoMs that have on-board Robotics Vision Core 3 (Robotics Vision Core 3). The OAK-SoM-Pro-S3 was designed first for evaluation and backwards compatibility with the OAK-SoM-Pro. The OAK-SoM MAX was designed later for maximum performance and extensibility.

Compatibility with existing models
- Pro-S3 is compatible with boards that have OAK-SoM-Pro integrated (list here). This allowed us for quicker evaluation of the new Robotics Vision Core 3.
- Max is not yet compatible with any boards.
Connectors
- Pro-S3 has 2x 100-pin mezzanine connector, with the exact same pinout as SoM-Pro.
- Max has 3x 100-pin mezzanine connector, which exposes additional MIPI RX/TX lines.
MIPI lines
- Pro-S3 has 2x 4-lane and 2x 2-lane MIPI RX lines.
- Max has 6x 2-lane MIPI RX and 2x 2-lane, 1x 4-lane MIPI TX lines.
RAM
- Pro-S3 has 1x 2GB DDR RAM due to size/shape constraints, even though Robotics Vision Core 3 supports dual channel RAM.
- Max has 2x 2GB DDR RAM on-board, which provides maximum performance, as Robotics Vision Core 3 supports dual channel RAM.
Size
- Pro-S3: 30mm x 45mm
- Max: 40mm x 40mm

OAK-SoM-Pro-S3 on the OAK-FFC-4P baseboard, which has OAK-SoM-Pro integrated by default.¶

Board Layout¶

Dimensions and Weight¶

Width: 30 mm
Height: 45 mm
Length: 4 mm
Weight: 5g

General information¶

Robotics Vision Core 3 unique features: 2 GB RAM (16GBit)
Robotics Vision Core 2 unique features: 512 MB RAM (4GBit)
Interfaces with the system through two 10-Gbps-rated 100-pin connectors (DF40C-100DP-0.4V(51))
2 x 2-lane MIPI channels
2 x 4-lane MIPI channels (for multiple camera inputs)
USB 3.1 Gen 2
Multiple Boot Modes supported: NOR, USB, Ethernet (EEPROM), or eMMC
SD-card supported
Design files produced with Altium Designer 20