Edge AI systems are increasingly required to process complex AI workloads locally with low latency and efficient power consumption. Applications such as industrial automation, smart retail, robotics, and intelligent surveillance demand real-time inference without depending entirely on cloud infrastructure.
The ARA-2 (Ara240) Discrete Neural Processing Unit (DNPU) addresses these requirements by delivering high-performance AI acceleration for embedded and edge computing platforms. When integrated with iWave i.MX95 SMARC System on Module, the platform enables scalable edge AI deployments for next-generation intelligent systems
Why ARA-2 for Edge AI?
ARA-2 is designed to accelerate modern neural network inference workloads while maintaining low power consumption and efficient system performance
Key Highlights
- Up to 40 eTOPS AI compute performance
- PCIe Gen4 x4 host interface
- Support for TensorFlow, PyTorch, and ONNX
- Compact M.2 form factor for embedded integration
- Approximate power consumption of 6.5W
- Support for concurrent multi-model execution
The architecture combines dedicated AI compute engines, high-speed connectivity, and secure execution capabilities to support demanding edge AI applications
By processing AI workloads locally, the platform helps reduce latency, improve responsiveness, and minimize dependency on continuous cloud connectivity.
Edge AI Acceleration on i.MX95 SMARC Platform
The ARA-2 (Ara240) DNPU can be integrated with high-performance embedded processor platforms to enable scalable edge AI systems. One such platform is the i.MX95 SMARC System on Module, which provides a powerful computing environment for AI-enabled embedded applications.
The module is based on the i.MX95 processor, which integrates multiple Arm Cortex-A55 cores along with dedicated real-time processing cores. This architecture allows the system to efficiently manage operating system tasks, multimedia processing, and peripheral control while AI inference workloads can be accelerated using the ARA-2 DNPU.
With support for high-speed interfaces, advanced connectivity, and industrial-grade design, the
iG-G61M® SMARC module provides a reliable hardware platform for deploying edge AI solutions in industrial automation, smart retail, and intelligent embedded systems.
The i.MX95 processor provides:
- 6 × Arm Cortex-A55 cores running up to 2.0 GHz
- Dedicated Cortex-M33 and Cortex-M7 real-time cores
- Up to 16 GB LPDDR5 memory support
- PCIe connectivity for AI accelerator integration
- Wi-Fi 6 & Bluetooth 5.3 Connectivity
- Industrial-grade SMARC form factor
This combination enables the embedded processor to manage operating system tasks, connectivity, and peripheral control, while the ARA-2 accelerator handles AI inference workloads efficiently. The modular SMARC architecture also simplifies integration into industrial and embedded edge systems
Edge LLM Inference on Embedded Platforms
ARA-2 (Ara240), when integrated with the i.MX95 SMARC System on Module, enables execution of Large Language Models (LLMs) directly on embedded edge devices. This allows real-time AI processing without dependency on cloud infrastructure.
In this use case, an LLM-based application runs locally on the embedded platform and processes user inputs through a terminal or application interface. The ARA-2 DNPU accelerates inference, enabling efficient response generation while maintaining low power consumption.
This approach supports intelligent edge systems that require low latency, local data processing, and reliable operation in environments where continuous cloud connectivity may not be available.
Example Interface Output:
Applications
Industrial Automation
Supports intelligent robotics, visual inspection systems, predictive maintenance, and automated production monitoring.
Smart Retail
Enables AI-powered inventory tracking, customer behavior analysis, smart checkout systems, and interactive kiosks.
Home and Entertainment
Enhances smart home devices, media processing systems, and AI-enabled entertainment platforms.
Edge AI Systems
Suitable for AI gateways, embedded computing devices, and edge servers requiring high-performance inference acceleration.
Conclusion
The ARA-2 (Ara240) DNPU provides an efficient and scalable solution for enabling edge AI acceleration on embedded platforms. Integrated with the i.MX95 SMARC System on Module, it delivers the compute performance required for modern AI inference while maintaining low power consumption and compact system integration.
With support for edge AI processing, multi-model inference, and embedded LLM deployment, the platform is well suited for next-generation intelligent systems across industrial, retail, and embedded computing applications.
For more information about iWave’s edge AI platforms and embedded solutions, reach out to mktg@iwave-global.com
