Forlinx Embedded News

Dedicated to discussing issues related to embedded systems on modules (SoMs) and single board computers(SBC). www.forlinx.net

[forlinx]

RV1126BJ-based FET1126BJ-S SoM delivers 3 TOPS AI, low power, Linux support, and industrial reliability for edge AI vision and embedded appl

Scaling Edge AI shouldn't feel like a compromise. 🚀

In 2026, the shift from "Cloud-heavy" to "Edge-native" is officially here. As Small Language Models (SLMs) and agentic AI become the standard for industrial automation, selecting the right hardware is the difference between a proof-of-concept and a successful mass production rollout.

Enter the FET1126BJ-S SoM by Forlinx Embedded.

Designed around the Rockchip RV1126BJ, this System-on-Module is built to handle 2B-parameter models locally—no cloud dependency, no latency, just pure performance.

Why it’s a game-changer for your next project:

🧠 3 TOPS AI Computing: Optimized for mixed-precision and Transformer models.

👁️ Smart Vision: Built-in AI-ISP supporting 8M@30fps with hardware-level HDR.

❄️ Industrial Resilience: Stable operation from -40°C to +85°C.

⏱️ Faster TTM: Pre-installed Linux 6.1 and comprehensive BSP packages.

Whether you’re building autonomous security systems or predictive maintenance tools for the factory floor, the FET1126BJ-S provides the stability you need to scale.

[forlinx]

Explore Forlinx Embedded's latest SoMs and edge computing gateways at embedded world 2026. Featuring i.MX 95/94, RK3588J, and TI AM62L solut

Forlinx Embedded is proud to participate in embedded world 2026 in Nuremberg, Germany, presenting a wide range of embedded computing platforms and application solutions at Booth 3-561.

At this year’s event, we are showcasing our latest embedded SoMs, development boards, industrial computers, and edge computing gateways built on leading platforms from NXP, Rockchip, Allwinner, and Texas Instruments. These solutions support a variety of applications including edge AI, industrial automation, smart displays, and EV charging.

Live demonstrations at the booth include:

i.MX9596 + Ara240 perception and understanding platform

i.MX94x smart display solution

FCU3501 edge AI gateway (RK3588J-based) supporting 16-channel AI video processing

RV1126B four-camera panoramic stitching system

T536 8-channel CAN communication module

TI AM62L interactive EV charger solution

We are also presenting the FCU2601 embedded control unit, which has obtained IEC 62443-4-1, IEC 62443-4-2, and EN 18031 cybersecurity certifications, supporting compliance with the EU Cyber Resilience Act (CRA) for secure industrial deployments.

During the event, Forlinx is also collaborating closely with ecosystem partners including NXP, Rockchip, and Texas Instruments, with multiple joint demonstrations and product showcases across partner booths.

With over 20 years of experience in embedded technology, Forlinx continues to expand its global presence by delivering reliable hardware platforms and integrated solutions for industrial and edge AI applications.

Visit us at Booth 3-561 to explore our latest innovations and discuss how embedded technologies are shaping the future of intelligent systems.

[forlinx]

Learn to enable Core Dumps on OK3568 Buildroot systems. Fix segmentation faults and stack overflows using GDB for efficient embedded Linux a

Stop Guessing Why Your Embedded Linux Apps Are Crashing! 🛠️

Debugging segmentation faults and stack overflows on platforms like the OK3568 can be a nightmare when serial logs aren't enough. If you’re working with Linux Kernel 5.10 and Buildroot, enabling Core Dumps is a game-changer for your development workflow.

A Core Dump is essentially a "black box" recorder for your software. When a process fails, it saves the memory image, registers, and call stack, allowing you to perform offline analysis using GDB without needing the hardware in front of you.

Key Takeaways from my latest guide:

✅ How to bypass the default "disabled" status in Buildroot.

✅ Setting persistent storage paths (e.g., /userdata/core).

✅ Essential GDB commands like bt (backtrace) to pinpoint the exact line of code that failed.

✅ Best practices for production vs. R&D environments.

Whether you are doing system integration or application development, mastering the ulimit and core_pattern configurations will save you hours of "blind" debugging.

[forlinx]

Achieve microsecond-level jitter control and high-performance real-time communication with IgH EtherCAT Master and Preempt-RT on the Forlinx

🚀 Unlocking Microsecond-Level Real-Time Control in Industrial Automation 🚀

In the world of industrial automation, achieving microsecond-level jitter control for high-precision applications like robotics and multi-axis motion control has always been a challenge. But with the combination of IgH EtherCAT Master and the Preempt-RT real-time kernel running on the Forlinx Embedded OK3576-C development board, we can now push the limits of real-time communication and control!

🔧 Key Highlights:

⚫ Achieved exceptional real-time performance with microsecond-level jitter reduction.

⚫ Cycle Synchronous Velocity Mode (CSV) reduced jitter from 6.308 μs to 3.579 μs.

⚫ Cycle Synchronous Torque Mode (CST) reduced jitter from 50.047 μs to 2.113 μs under full load stress.

⚫ CPU isolation strategy applied for even better performance.

This groundbreaking solution opens up new possibilities for industries that require precise control and reliability.

👉 Learn more about the real-time performance of this industrial control solution and how it can transform your applications.

[forlinx]

Learn how to train a custom YOLOv5 model, convert .pt to RKNN, and deploy on RK3568. Covers environment setup, LabelImg, ONNX export, and rk

The journey from a local training environment to real-time inference on embedded hardware can be a complex puzzle. I’ve just documented a complete pipeline for training a custom YOLOv5 model and successfully deploying it on the Rockchip RK3568 NPU.

The Workflow Highlights:

🛠️ Environment Setup: Navigating the specific Python 3.8/3.9 requirements for YOLOv5 and LabelImg on Windows 11. 🏷️ Dataset Engineering: Implementing a clean directory structure and YOLO-format annotation for high-quality training. 🔄 Model Transformation: Converting the PyTorch .pt weights into .onnx and finally into .rknn using the RKNN-Toolkit2 on Ubuntu 22.04. 🚀 Hardware Deployment: Compiling the C++ RKNPU2 demo and optimizing the post-processing headers for high-performance edge inference.

The result? A seamless transition from a standard PC environment to efficient, hardware-accelerated detection on an embedded development board.

Check out the full breakdown below! 👇 https://www.forlinx.net/industrial-news/yolov5-training-rk3568-rknn-deployment-guide-775.html

[forlinx]

Learn how to convert YOLOv8 models to RKNN and deploy on OK3576-C with RK3576 NPU for real-time embedded AI object detection.

🚀 Edge AI Made Easy with OK3576-C & YOLOv8!

With the rapid growth of intelligent vision applications, real-time on-device object detection has become a core capability for industrial vision, smart devices, and surveillance.

At Forlinx Embedded, we’ve demonstrated a full workflow from model to edge deployment using the OK3576-C board with RK3576 NPU:

1️⃣ Convert YOLOv8 pre-trained models (.pt) → RKNN format 2️⃣ Deploy the model on the OK3576-C board 3️⃣ Run real-time object detection using a USB camera (~30 FPS)

This guide helps developers quickly move from model training to edge inference with minimal setup, leveraging the RK3576 NPU’s efficiency and power.

[forlinx]

Enable fast device-to-device communication on OK3588/RK3562 using USB NICs. Stable, low-cost, and easy embedded networking solution.

🔌 Efficient and Low-Cost Device-to-Device Communication in Embedded Systems 🚀

In embedded system development, a stable, cost-effective, and easily deployable communication solution is key for rapid testing and integration. On the Rockchip OK3588 platform, we've implemented a USB-based networking solution that enables seamless TCP/IP communication between devices, without the need for additional Ethernet ports or complex switching equipment.

Key Features:

✅ USB NIC Virtualization: Automatically creates a virtual NIC (usb0) in Host/Device mode using the Linux USB Gadget mechanism.

✅ Cross-Platform Support: Facilitates collaboration between different platforms like OK3588 and RK3562.

✅ Fast Debugging & System Validation: Simplifies rapid prototyping, debugging, and remote maintenance.

🔧 Tested Modes:

Master-Slave Mode Switching: Easily toggle between Host and Device modes using simple commands.

Ping Connectivity: Reliable testing with Host and Device communication for inter-device testing.

This solution enables developers to quickly establish a communication network between embedded systems—accelerating development and product integration for real-world applications. Whether for debugging, system integration, or cross-platform collaboration, it's a game-changer for embedded network setups.

[forlinx]

Learn how to release GPIO pins from Linux on i.MX8MP so the Cortex-M7 can use them safely for control and interrupts without A53 conflicts.

On the i.MX8MP platform, the Cortex-M7 (M-Core) and Cortex-A53 (Linux) often compete for GPIO resources. By default, Linux owns the pinmux and GPIO controllers, which can prevent the M-Core from using pins reliably—especially for interrupt inputs.

To give the M-Core full control, the A-Core must first “release” those pins in the device tree: ✅ Remove the pinmux configuration for the target GPIO ✅ Comment out all Linux device nodes that reference it ✅ For interrupt use on GPIO2/3/4 groups, disable the entire GPIO controller so Linux no longer registers it

This clean separation avoids resource conflicts, ensures stable interrupts, and allows the M-Core to operate with real-time determinism.

A small device tree change, but a big step toward robust heterogeneous multi-core design on i.MX8MP.

[forlinx]

Learn how to port and configure Nginx on OKMX6ULx embedded Linux 4.1.15, including reverse proxy and load balancing with Tomcat on ARM platf

🚀 Guide to Porting & Configuring Nginx on OKMX6ULx Embedded Linux

We recently tested Nginx (v1.8) on the OKMX6ULx Embedded Linux platform (Kernel 4.1.15). Nginx is a high-performance web server, reverse proxy, and load balancer, optimized for high concurrency and low memory usage.

Key Highlights:

✅ Yocto Compilation: Successfully compiled Nginx on imx6ull14x14evk, handling source path adjustments.

✅ Deployment: Packaged and transplanted Nginx to the development board seamlessly.

✅ Reverse Proxy Setup: Forwarded requests from Nginx to Tomcat 9, enabling efficient web service management.

✅ Load Balancing: Configured two Tomcat instances on ports 8080 and 8081 with Nginx upstream, ensuring traffic is balanced across services.

This workflow demonstrates how embedded developers can leverage Nginx for robust web services in constrained environments.

💡 Takeaway: Embedded platforms can handle high-performance web services with proper configuration, bridging industrial systems with modern web architecture.

[forlinx]

Step-by-step guide to flash boot.img and rootfs.img via TFTP on Forlinx RK3568, configure U-Boot, and verify eMMC partitions for industrial-

🚀 TFTP Flashing Guide for Forlinx RK3568 Development Board

Efficient system flashing is critical during embedded Linux development and debugging. On the Forlinx RK3568 development board, flashing boot.img and rootfs.img via TFTP + U-Boot provides a fast, stable, and hardware-friendly workflow.

🔧 What this guide covers:

✅ U-Boot network environment configuration (MAC, IP, gateway) ✅ TFTP server setup and connectivity verification ✅ eMMC partition layout on RK3568 ✅ Step-by-step flashing of boot.img and rootfs.img ✅ Practical tips for industrial-grade stability and scalability

💡 Why RK3568 + Forlinx?

🔹 Industrial-grade eMMC & power protection 🔹 Dual Gigabit Ethernet for low-latency flashing 🔹 SATA 3.0 & M.2 expansion without reflashing 🔹 Integrated 1 TOPS NPU for lightweight AI at the edge

This approach significantly shortens iteration cycles and is ideal for industrial control, smart devices, and edge computing projects.

👉 If you’re working with RK3568 or similar Rockchip platforms, this workflow is worth bookmarking.

[forlinx]

Explore Forlinx Embedded’s 2025 portfolio: High-performance i.MX95 SoMs, RK3588 AI computers, and RK3506 platforms designed for 10-15 year i

Forlinx Embedded 2025 Recap: Powering the Next Wave of Industrial Edge AI Intelligence

2025 marked a year of deep collaboration, platform innovation, and real-world deployment for Forlinx Embedded.

As an NXP Gold Partner and official i.MX95 ecosystem partner, we continued aligning closely with NXP’s roadmap for secure, safety-ready edge computing—while further strengthening our strategic partnerships with Rockchip and Allwinner to deliver scalable, cost-effective AI edge platforms.

Throughout the year, we introduced a new generation of industrial System on Modules (SoMs), SBCs, and fanless embedded computers, supporting applications from safety-critical automation and smart energy to machine vision, transportation, and IoT gateways.

Across all platforms, three principles remain consistent: 🔹 Industrial-grade reliability over time 🔹 Predictable system behavior for certification and deployment 🔹 A smooth path from evaluation to mass production, with 10–15 years lifecycle support

From i.MX95 high-end edge controllers to RK3588 AI edge computers and ultra-low-power industrial control SoMs, we are bringing AI closer to where data is generated—securely, reliably, and at scale.

Looking ahead, we remain committed to building long-life embedded platforms that reduce integration risk and accelerate industrial AI adoption worldwide.

👉 Explore our 2025 product portfolio and ecosystem highlights.

[forlinx]

FET1126Bx-S is a compact industrial edge AI SoM based on Rockchip RV1126B, featuring a 3 TOPS NPU, Linux 6.1, and rich video interfaces.

🚀 New Product Release | Industrial Edge AI SoM

As AI continues moving from centralized cloud platforms to edge-side intelligence, industrial applications demand real-time AI inference, lower latency, and higher reliability.

Forlinx Embedded is proud to introduce the FET1126Bx-S Industrial System on Module, powered by Rockchip RV1126B/RV1126BJ—a compact, cost-effective edge AI solution designed for real-time video analytics.

🔹 Quad-core Arm Cortex-A53 🔹 3 TOPS NPU for on-device AI inference 🔹 Linux 6.1 with RKNN AI toolchain 🔹 Advanced ISP & multi-camera support 🔹 Industrial-grade design (−40°C to +85°C)

Ideal for smart manufacturing, public safety, AIoT, and edge vision systems, FET1126Bx-S enables developers to deploy AI closer to where data is generated—without cloud dependency.

👉 Free engineering samples are available for qualified projects.

[forlinx]

FCU3501 is an RK3588J-based AI edge computing box enabling real-time perception and autonomous decision-making for intelligent flexible lane

🚦 When traffic demand changes in seconds, lane control should too.

Morning rush hour.

Left-turn lanes backed up for hundreds of meters — while straight lanes sit nearly empty.

Yet traditional flexible lane systems still rely on fixed schedules or cloud-only control.

🔍 The problem?

Long perception-to-decision cycles and network latency.

💡 The solution: Edge intelligence at intersections.

With the FCU3501 AI Edge Computing Box, built on the industrial-grade RK3588J, traffic perception and decision-making move directly to the roadside:

☑ Real-time AI video analysis

☑ Local, autonomous lane control decisions

☑ Reduced cloud dependency and network latency

☑ 24/7 reliable operation in harsh outdoor environments

From timed to real-time,

from centralized to distributed,

from preset to autonomous.

This is how intelligent flexible lanes truly become intelligent.

[forlinx]

Discover how Forlinx embedded products pass rigorous high and low temperature tests to ensure industrial-grade reliability and stability.

❄️🔥 "Ice and Fire" Trials: Ensuring Forlinx Embedded Product Reliability

At Forlinx Embedded, we put our products through rigorous high- and low-temperature tests to simulate extreme environments. From industrial-grade SoMs to commercial devices, these tests ensure adaptability, stability, and reliability in real-world conditions.

By verifying performance across different temperature grades, we safeguard product quality and give users confidence in harsh environments.

[forlinx]

Deploy Piper TTS on the Forlinx OKMX8MP board to enable fast, offline text-to-speech for IoT and edge AI applications. Includes full setup s

🚀 Offline TTS on Edge Devices Just Got Easier!

In many IoT and edge-computing scenarios, devices need to speak — even without an internet connection. That’s why Piper, a fast, lightweight, open-source neural TTS engine, is becoming a preferred choice for embedded systems.

We recently walked through how to deploy Piper on the Forlinx OKMX8MP development board, enabling high-quality, low-latency text-to-speech directly on the device.

🔧 What the guide includes:

• Full deployment steps for Piper on OKMX8MP-C

• How to install voice models and run local TTS

• Commands and sample scripts

• Troubleshooting common issues (missing models, library errors, encoding problems)

• Tips for switching voices and testing via ONNX models

💡 Why this matters:

Local TTS is essential for industrial automation, edge AI applications, robotics, smart terminals, and environments with limited connectivity. Piper delivers impressive performance while keeping everything fully offline.

If you’re working with i.MX8M Plus, edge AI, or embedded Linux, this tutorial will help you get started quickly and reliably.

🌐 Want to join our embedded tech community?

We’ve created a Facebook technical group where engineers share insights, projects, and solutions. Join us and grow together!

[forlinx]

Enhance smart factories with FCU3501 AI edge computing box for real-time visual inspection, equipment monitoring, and reliable industrial AI

Building the Core of Visual Analysis in Smart Factories with FCU3501

As manufacturing moves from experience-based decision making to data-driven operations, visual inspection, safety monitoring, and equipment status visibility are becoming mission-critical. Traditional PCs and cloud-only solutions often fail to meet the needs of real-time, high-reliability industrial environments.

The Forlinx Embedded FCU3501 AI Edge Computing Box is engineered specifically for these challenges.

⚙️ Industrial-grade performance

✅ Based on the Rockchip RK3588J processor ✅ Built-in 6 TOPS NPU, expandable to 32 TOPS via M.2 accelerator ✅ Capable of processing 16 channels of 1080P video streams simultaneously ✅ Wide-temperature operation (-40°C to +85°C), fanless, CE/FCC certified

📌 Multiple application values

⚫ Real-time visual quality inspection ⚫ Equipment status monitoring + alarm output ⚫ Personnel safety behavior detection ⚫ Local HDMI display + wireless debugging

☁️ Edge-Cloud collaborative architecture Only structured results and alerts are uploaded to the cloud—reducing bandwidth, improving real-time response, and enabling scalable digital transformation.

🔧 Future-proof design Start with 6 TOPS and scale to 32 TOPS as models grow or new cameras are added.

The FCU3501 is becoming a core engine for AI-powered smart factories—boosting safety, improving quality, and accelerating the path toward intelligent manufacturing.

[forlinx]

Rockchip RV1126B delivers 3TOPS AI performance, advanced AI-ISP imaging, low-power AOV 3.0, and strong security, making it ideal for edge AI

Rockchip RV1126B is redefining the performance standard for edge AI vision.

As AIoT applications continue to grow—across smart security, industrial inspection, robotics, and intelligent vehicles—the demand for on-device AI processing and high-quality imaging has never been higher.

The newly released Rockchip RV1126B delivers a comprehensive upgrade across architecture, performance, and power efficiency, including:

🔹 3 TOPS NPU with mixed-precision support (INT4/8/16, FP16)

🔹 Dedicated AI-ISP for day/night adaptation and ultra-low-light imaging

🔹 AOV 3.0 for 24/7 audio event detection with standby power as low as 1mW

🔹 8MP 45fps video encoding, 6-DoF digital anti-shake, and multi-camera stitching

🔹 Enhanced security with SM2/SM3/SM4, TrustZone, and keyladder

With the ability to run lightweight LLMs and multimodal models (<2B parameters), the RV1126B opens new possibilities for smarter, more responsive edge-AI devices.

From intelligent cameras to industrial visual inspection and next-gen robots, RV1126B is setting a new benchmark for performance, clarity, and reliability.

Forlinx Embedded will soon launch new hardware solutions based on RV1126B—stay tuned!