Industrial Edge AI Platform Integrates Multi-Modal Acceleration

[Erkan Teskancan]

## Industrial Edge AI Platform Integrates Multimodal Acceleration

Edge AI platforms capable of processing multimodal data locally to support real-time decision-making in industrial settings are becoming increasingly important. In this context, Kontron has collaborated with SiMa.ai to develop the KBox A-151 EAI, an industrial edge AI computer, for high-performance AI inference.

### Edge AI Architecture for Multimodal Industrial Workloads

The KBox A-151 EAI is an edge computing platform designed for industrial automation, autonomous systems, and AI inference tasks. The system combines 13th Gen Intel® Core™ or Intel Atom® processors with the SiMa.ai MLSoC™ Modalix AI accelerator, delivering over 50 TOPS (trillions of operations per second) of AI processing performance.

The dual-processor architecture separates traditional application processing from AI inference tasks, allowing operational logic and edge AI workloads to run independently. This structure aims to support system stability in industrial environments where deterministic processing behavior is crucial.

Application areas include industrial automation, energy infrastructure, medical technology, and smart transportation systems. Edge processing reduces latency while minimizing reliance on cloud connectivity in distributed digital infrastructure.

### Hardware Design Suitable for Industrial Environments

The system is designed with a fanless, passively cooled enclosure, maintaining thermal performance without requiring active cooling. This approach reduces maintenance needs and prevents performance degradation due to thermal limitations.

The platform supports various AI workloads, including real-time video analytics and generative AI models, large language models (LLMs), large multimodal models (LMMs), and convolutional neural networks (CNNs. It also enables end-to-end machine vision processes and multi-sensor data integration.

With the integration of the Palette™ software development kit and Edgematic™ development environment, multimodal models can be easily deployed using standard machine learning frameworks. This accelerates the model development and deployment process without requiring significant hardware changes.

### Empowering Physical AI in Edge Environments

The platform targets physical AI applications where AI systems need to interpret sensor data in real-time and interact with physical processes. It enables the integration of image, audio, and other sensor data into operational decision workflows.

The KBox A-151 EAI, with its configurable input/output interfaces, scalable memory options, and industrial certification support, is suitable for environments requiring long product lifecycles and predictable system behavior.

By integrating the SiMa.ai MLSoC™ accelerator into an industrial edge platform, the system meets both AI inference performance requirements and the reliability standards sought in industrial computing platforms.