RK3588 AI Module7

RK3588 AI Module7 (AIM7) features a 64-bit octa-core processor, a 6 TOPS NPU, 8 GB of memory, PCIe 3.0 support, and software-defined pin functionality. It is open hardware, and we have published our design documents online, including schematics and interface definitions. Compatible with the NVIDIA Jetson Nano interface standards, its heterogeneous computing architecture is ideal for edge computing, AI inference, VR/AR, smart NVR systems, and general-purpose applications.

AIM-IO is an open-source expansion board designed for the Jetson Nano ecosystem. It pairs with the RK3588 AI Module7, giving you a miniature AI development platform that enables custom carrier board development and accelerates machine learning deployment for prototyping smart devices. AIM-IO empowers developers, educators, and makers by providing open flexibility and supporting rapid iteration. We have published the schematics, PCB design files, and pin-mux tables.

Native AI Support

Developers can deploy AI applications rapidly using the following tools without complex hardware programming:

RKNN-Toolkit2

One-click conversion of TensorFlow/PyTorch models to RKNN format for NPU acceleration
INT8 quantization tools reduce model size to 1/4 with <1% accuracy loss
Real-time performance analysis: layer-wise compute visualization and memory optimization

Performance Benchmarks

Object Detection (YOLOv5s@1080p): 55 FPS
Semantic Segmentation (DeepLabv3+@4K): 25 FPS
Video Analytics (8x 1080p streams processed synchronously): <100 ms latency

Architectural Advantages & Applications

The RK3588 + AIM-IO combo covers high-value domains with hardware flexibility and a robust software ecosystem. Applications might include:

Smart Vision Hub

Industrial inspection: 12x MIPI CSI camera synchronization, defect detection accuracy >99.9%
Smart retail: 8K crowd analytics + dynamic ad placement with RTMP live streaming

Edge Cloud Servers

Micro data centers: 4x RK3588 cluster, <50 W power consumption, replacing traditional x86 solutions
MLOps platform: Supports TensorFlow Serving/Kubeflow edge deployment

Robotics Control

Multi-axis robotic arms: Hard real-time control (<1 μs jitter) + visual servoing
Autonomous driving: Multi-sensor fusion (LiDAR + camera + radar)

Outperforms Jetson Nano

The RK3588 AI Module7 is powered by RK3588, which outperforms the Jetson Nano with:

A faster CPU, delivering four times the performance
Ample LPDDR4X memory, two times greater than Jetson Nano
Advanced codec capabilities, four times more powerful
State-of-the-art high-speed interfaces, fully upgraded for PCIe, HDMI, and DP
Comprehensive support for a variety of mainstream operating systems, including but not limited to Debian, Ubuntu, and Armbian

Unlock NPU Potential

With the AIM-IO Expansion Board, you can quickly implement advanced applications. Below are a few hardware-configuration recommendations:

Lightweight AI Inference: Add 64 GB eMMC
Multi-Modal Edge Computing: Add NVMe SSD expansion
Industrial Deployment: Add AIM-IO for PoE support

Why RK3588 AI Module7?

Traditional edge computing devices often struggle to balance performance and energy efficiency. The RK3588 groundbreaking design changes the game:

1. Heterogeneous computing architecture, a revolution in collaborative computing

Triple acceleration (CPU+GPU+NPU): Parallel processing of 8K video codec and AI inference tasks, reducing latency by 70%.
6 TOPS NPU custom optimization: Tailored operator libraries for models like YOLO/Transformer, boosting inference efficiency by 5x.

2. Full interface compatibility & seamless ecosystem migration

260-Pin gold finger design: Mechanically 100% compatible with Jetson Nano carrier boards, reusing existing hardware investments.
PCIe 3.0 x4 high-speed channel: Supports external AI accelerators (e.g., TensorRT accelerators).

3. Risk Mitigation

Rockchip's production capacity assurance + ArmSoM strategic inventory: Ensures supply stability.
ArmSoM's open-source commitment: Avoids platform lock-in caused by closed ecosystems.

Features & Specifications

RK3588 AI Module 7 (AIM7)

	Details
Processor	Quad-Core ARM Cortex-A76 at 2.4 GHz + Quad-Core ARM Cortex-A55 at 1.8 GHz
GPU	ARM Mali-G610 MP4 (supports 8K HDR display)
Memory/Storage	8 GB LPDDR4x + 64/128GB eMMC 5.1
Video Codec	8K @ 60fps Decode (H.265/VP9/AVS2); 8K @30 FPS Encode (H.265/H.264)
Key Interfaces	12x MIPI CSI-2 (multi-camera sync); PCIe 3.0 x4 (AI accelerator support); HDMI 2.1 + DP 1.4 (dual 8K output)
OS Support	Linux 5.10 (Ubuntu, Debian, Armbian)
Compatibility	260-pin edge connector, compatible with NVIDIA Jetson Nano ecosystem
Dimensions	69.6 x 45 mm (footprint smaller than a credit card)

AIM-IO Carrier Board

	Details
Display Output	Dual 4K @ 60Hz support (DP + HDMI)
High-Speed I/O	M.2 E-key for Wi-Fi and 5G modules; 4x USB 3.0 + Gigabit Ethernet
Camera Support	3x MIPI CSI (up to 8 cameras)
Industrial Design	PoE power support; 40-pin GPIO + UART/SPI/I²C
Rapid Deployment	One-click firmware flashing (USB Type-C)
Size Comparison	Same footprint as Jetson Nano dev kit (100x80 mm), compatible with existing enclosures
Jetson Nano Compatibility	Provides both 5 V and 12 V power inputs for Jetson Nano

Comparisons

	8 GB ArmSoM RK3588 AI Module7 (Rockchip)	NVIDIA Jetson Nano	8 GB Turing RK1
USB Ports	1x USB 3.0, 3x USB 2.0	1x USB 3.0, 3x USB 2.0	2x USB 3.0, 2x USB 2.0
Video Encoding	H264, 8k @ 30 FPS VP9/H265, 8k @ 60 FPS AV1/AVS2, 4k @ 60 FPS	1x 4K @ 30 (HEVC), 2x 1080p @ 60 FPS (HEVC), 4x 1080p @ 30 FPS (HEVC)	H264, 8K @ 30 FPS VP9/H265, 8k @ 60 FPS AV1/AVS2, 4k @ 60 FPS
Video Decoding	H265, H264, VP9, AV1, and AVS2 up to 8K @ 60 FPS	1x 4K @ 60 (HEVC), 2x 4k @ 30 FPS (HEVC), 4x 1080p @ 60 FPS (HEVC), 8x 1080p @ 30 (HEVC)	H265, H264, VP9, AV1, and AVS2 up to 8K @ 60 FPS
CSI Interfaces	2× MIPI-CSI RX 4× lanes; 1× MIPI-CSI DPHY RX 4× lanes	12 channels (3x4 or 4x2) MIPI CSI-2 D-PHY 1.1 (18 Gbps)	2× MIPI-CSI RX 4× lanes; 1× MIPI-CSI DPHY RX 4× lanes
Display output port	1x HDMI 2.1; 1x eDP 1.4; 1× MIPI-DSI DPHY 2× lanes; 1× DP 1.4	1x HDMI 2.0; 1x eDP 1.4; 1x x2 DSI (1.5 Gbps/lane)	1x HDMI 2.1; 1× MIPI-DSI DPHY 2× lanes; 1× DP 1.4
PCIe	1x 1/2/4 lane PCIe 3.0 & 1x 1 lane PCIe 2.0	1x 1/2/4 lane PCIe 2.0	PCIe Gen3
GPU	ARM Mali-G610 MP4	128-core Maxwell GPU	ARM Mali-G610 MP4
CPU	Quad-core ARM Cortex-A76 + Quad-core ARM Cortex-A55	Quad-core ARM Cortex-A57 processor	Quad-core ARM Cortex-A76 + Quad-core ARM Cortex-A55
NPU	6 TOPs（INT8）	None	6 TOPs（INT8)
Memory	8 GB LPDDR4x, 2112 MHz	4 GB LPDDR4, 1600 MHz	8 GB LPDDR4x, 2112 MHz
Storage	microSD card, 32 GB eMMC 5.1 flash storage	microSD card, 16 GB eMMC 5.1 flash storage	SD 3.0, 32 GB eMMC 5.1 flash storage
I/O	3x UART, 2x SPIs, 2x I²S, 4x I²Cs, multiple GPIOs	3x UARTs, 2x SPIs, 2x I²S, 4x I²Cs, multiple GPIOs	2x I²S, multiple GPIOs
OS Support	Debian, Android	NVIDIA JetPack software suite	Ubuntu

Performance

Geekbench source data: 1, 2,3, 4

Support & Documentation

ArmSoM supports various open-source operating systems, including:

Debian Bullseye 11: Stable Linux environment for development Download Debian image
Ubuntu 24.10: Popular Linux distribution with extensive package support Download Ubuntu image

Armbian operating systems and additional distributions are continuously being developed and will be available soon.

We warmly invite you to explore our comprehensive documentation. It contains in - depth technical guides. These guides cover every aspect, starting from the basics of getting started and advancing all the way to complex implementations. They ensure that you have all the necessary resources to develop innovative solutions.

Full schematics & hardware design files for RK3588 AI Module7 (AIM7) will be released before your products ship via Mouser!This information will be made publicly available, helping you gain a better understanding and enabling more efficient work.

Should any inquiries arise, please don’t hesitate to contact us via the designated "Ask a question" link below to submit your inquiries. Or, voice your concerns on our Discord platform. There, you can communicate with our team and other users, and get quick help and useful advice.

Manufacturing Plan

Prototype Design (COMPLETED): The functional prototype has been rigorously validated by 15 beta testers,confirming operational reliability and user acceptance.
Small-Batch Trial Production (COMPLETED): A pilot run of 100 units was executed to verify manufacturing scalability, quality control protocols, and supply chain readiness.
Mass Production Timeline: Full-scale production will commence immediately following the crowdfunding campaign to fulfill backer rewards, pre-orders, and inventory buildup for retail distribution.

Fulfillment & Logistics

Packaging & Quality Assurance: Completed units will undergo final inspection and secure packaging to meet global shipping standards.
Partnership with Mouser Electronics: Crowd Supply’s certified fulfillment partner, Mouser Electronics, will manage global distribution, including customs clearance, last-mile delivery, and real-time tracking.
Transparency & Support: Backers can review detailed logistics policies (e.g., shipping timelines, regional tariffs) in Crowd Supply’s Ordering, Paying, and Shipping Guide.

Risks & Challenges

Documentation: Some Chinese materials may require translation. We will progressively improve documentation during the campaign to ensure clarity for backers.
CUDA Compatibility: AIM7 cannot replace Jetson Nano for CUDA workflows, though Rockchip’s improved NPU libraries offer compelling alternatives.
Armbian Support: Armbian is not yet supported, but we intend to adapt it before we ship.
AIM7 Compatibility with the Jetson Nano Carrier Board: While RK3588 AI Module7 is compatible with the Jetson Nano carrier board, it must be powered at 5 V only. Using 12 V power might damage the CPU. (Jetson Nano can be powered at either voltage using our AIM-IO Carrier Board.)
Turing Pi 2 Compatibility: Verification is currently pending due to unavailable schematics.
Tariff Volatility: Rapidly changing policies may impact future pricing and supply-chain stability.