Edge AI Expansion Board for Raspberry Pi 5
Request a quote- Ships from Germany in 1–3 business days
- 2-year warranty
- Secure checkout
Description
Build connected edge AI systems on Raspberry Pi 5. Sixfab Edge AI Expansion Board combines DEEPX NPU acceleration, NVMe SSD storage, and LTE/5G connectivity in a single under-board platform designed for connected field deployments.
25 TOPS · INT8 (DX-M1 M.2 NPU) · PCIe Gen 2/3 x1 · triple M.2 (AI + NVMe + LTE) · USB-C PD, 27 W min, 45 W recommended. For Raspberry Pi 5 only.
Performance at a glance
- 25 TOPSINT8 · DEEPX DX-M1 M.2
- Triple M.2AI + NVMe + LTE/5G
- USB-C PD27 W min · 45 W recommended
Highlights
- Swappable M.2 AI moduleDEEPX DX-M1 (25 TOPS at INT8) on a removable M.2 module. Bundled in the “Board + DEEPX DX-M1” variant · not included in “Board only”.
- PCIe Gen 2 / Gen 3 x1Dedicated PCIe lane for NPU ensuring stable high-throughput inference.
- Triple M.2 architectureSeparate lanes for AI (PCIe), NVMe storage, and LTE/5G connectivity running in parallel.
- USB-C PDSingle-cable power input that also back-powers Raspberry Pi 5 via pogo pins.
- DXNN SDK + Model ZooPre-compiled models or full ONNX → DXNN deployment pipeline.
- On-device inference, no cloudFully offline AI processing with no cloud dependency.
Technical specifications
| AI accelerator | DEEPX DX-M1 · 25 TOPS at INT8 · on swappable M.2 module |
|---|---|
| NPU memory | 4 GB LPDDR5 @ 5600 MT/s on DX-M1 module |
| Host interface | PCIe Gen 2/3 ×1 over 40 mm FFC cable |
| M.2 slots | 3× M.2 · AI M-Key · NVMe M-Key · LTE/5G Key-B |
| NVMe storage | 2230/2242/2260/2280 · 380-440 MB/s read · 350-410 MB/s write |
| Cellular | M.2 Key-B · 1× nano SIM · eUICC supported |
| Power input | USB-C PD · 27 W min · 45 W recommended |
| Pi back-power | 2× 5V + 2× GND pogo pins · 3 A per pin |
| Form factor | 88.46 × 89.19 mm · ~36.26 mm Z-stack with Pi 5 |
| Operating temp | 0-70 °C commercial |
| Supported host | Raspberry Pi 5 |
| Host OS | Raspberry Pi OS (Trixie) |
| Runtime | dxrt-runtime · APT install · Python & C++ APIs |
| Model pipeline | ONNX → DXNN via DX-COM compiler |
| Hot-plug | Not supported. Power off Pi 5 before mounting |
| Compliance | |
| Certification | Certification in progress: CE, FCC, UKCA, RoHS, REACH |
Use cases
- Connected smart camerasRun real-time object detection or face recognition on-device. Transmit metadata and alerts over LTE/5G. Raw video stays local.
- Robotics & autonomous dronesEquip mobile platforms with real-time vision AI and always-on cellular uplink. Make on-device decisions at full frame rate.
- Outdoor vision systemsCount and classify vehicles, pedestrians, or wildlife from remote edge units. Aggregate metadata over cellular while keeping raw video fully on-device for privacy.
- Remote monitoring & sensorsMonitor equipment or infrastructure in unattended environments. Run anomaly detection locally with the NPU and transmit only event-based alerts over LTE.
- IoT sensor fusionCombine camera streams, sensor data, and telemetry for multi-modal edge inference. Store processed results locally on NVMe, sync summaries to the cloud on your schedule.
- Live field testing & stagingValidate AI models in real-world environments before ALPON X5 AI deployment. The Expansion Board runs the same DEEPX SDK and models, zero porting effort when you scale up.
What's in the box
- AI Expansion Board×1
- USB 3.0 Bridge×1
- PCIe Cable×1
- M2.5 5mm M-F Plastic Spacer×4
- M2.5 15mm F-F Plastic Spacer×6
- M2.5 Screw×6
- USB Type-C Plastic Cap×1
- M2 Flathead Screw×3
- M2 Module Plastic Spacer×3
Board only – board + mounting kit (9 items above). Board + DEEPX DX-M1 – board + mounting kit + AI module: adds the DEEPX DX-M1 M.2 AI Module (25 TOPS · INT8) ×1.
Not included (sold separately)
| Required | Raspberry Pi 5 (host board) · USB-C PD power supply (27 W min, 45 W recommended) · DEEPX DX-M1 M.2 AI module (for “Board only”; bundled in “Board + DEEPX DX-M1”) |
|---|---|
| Optional | NVMe SSD (M.2 2230, 2242, 2260, or 2280) · LTE/5G M.2 modem module (M.2 Key-B) · nano SIM card (only if using cellular) · cellular antennas (connector type depends on the chosen modem module) |
Compatibility
Supported host
- Raspberry Pi 5
Not supported
- Raspberry Pi 4
- Compute Module 4
- Non-Raspberry Pi SBCs
Frequently asked questions
How is it different from the Sixfab AI HAT+?
The AI HAT+ targets early-stage AI prototyping. The Expansion Board adds dedicated NVMe and LTE/5G M.2 slots alongside DEEPX AI acceleration, with all components running in parallel on the same board through a single USB-C connection.
Does it require cloud connectivity for AI inference?
No. AI inference runs fully on-device using the DEEPX NPU. The cellular connection transmits metadata, alerts, or application-level outputs.
What AI models can I run?
The platform supports models compiled for the DEEPX DX-M1 NPU family. This includes YOLO-family models, pose estimation networks, and CNN-based architectures such as VGG-type models.
Can I run multi-camera vision workloads?
Yes. The Raspberry Pi 5 platform supports multi-camera configurations, enabling multi-angle monitoring, inspection workflows, and spatial vision applications.
How does NVMe storage help in edge AI?
NVMe storage enables high-speed local data logging and buffering for AI pipelines: store datasets locally, buffer inference results, reduce reliance on cloud storage, and improve performance for continuous workloads.
How does the Expansion Board fit into the Sixfab product line?
It is Stage 2 in a unified DEEPX NPU ecosystem: Stage 1: AI HAT+ for rapid prototyping; Stage 2: Edge AI Expansion Board for connected field deployments; Stage 3: ALPON X5 AI for rugged industrial systems.
What is the difference between the Expansion Board and the ALPON X5 AI?
The Edge AI Expansion Board is designed for Raspberry Pi 5-based deployments, making it ideal for development, testing, and connected field systems. ALPON X5 AI is a fully integrated industrial edge AI computer with a rugged enclosure, extended temperature range, hardware reliability features, and fleet management capabilities.
Why choose the Expansion Board over a standalone AI HAT for field deployment?
A standalone AI HAT handles inference. A field deployment also needs local data logging and a connectivity layer that doesn't depend on Wi-Fi. The Expansion Board solves all on the same board: DEEPX NPU for inference, NVMe SSD for high-speed local storage, LTE/5G M.2 slot for cellular uplink.
Cloud-connected vs. Expansion Board approach?
With a cloud-connected setup, raw video or sensor data leaves the device for remote processing - adding latency, bandwidth cost, and a dependency on network availability. The Expansion Board runs DEEPX NPU inference fully on-device. LTE/5G transmits only the output: metadata, alerts, or event triggers.
