Developing intelligent edge AI solutions requires more than just precision AI models and high-performance hardware. The focus is on agility across the entire AI lifecycle: How rapidly can models be developed and deployed, and how adaptable are they to hardware and software changes?
Modern edge solutions must be able to handle complex tasks such as image and speech recognition or anomaly detection in real time, all while operating under strict resource and energy constraints. This demands a platform architecture that can be seamlessly integrated across diverse applications, from industry to smart retail.
Development platform with a vision
Edge applications place varying requirements on processing power, interfaces, graphics and AI capabilities, depending on their specific use cases. MediaTek’s Genio platform tackles these challenges with a holistic approach that combines SoCs (Systems-on-a-Chip) specifically designed for edge AI applications with NeuroPilot SDK (software development kit). NeuroPilot SDK supports various interfaces, including those for Nvidia TAO AI models. This provides users with consistent SDK and board support across the entire product range.
Table 1 shows the currently available derivatives that are particularly suitable for demanding edge AI applications. They range from energy-efficient entry-level variants to high-performance variants designed for vision, HMI (Human-Machine Interface) and generative AI applications.
These platforms not only provide a comprehensive range of services but also set the standard for functionality and future readiness:
- Arm Cortex-A5/A7 multicore architectures for efficient processing,
- 10 years of product support for investment security,
- 6 nm manufacturing technology,
- dedicated APU (Accelerated Processing Unit) for real-time AI,
- security functions at SoC level
- and a uniform SDK for rapid development.
Accelerating AI model time-to-market
Fully equipped kits featuring touch display, camera, WLAN and GPIO are available for quick evaluation on all platform variants. They enable direct testing with video, HMI or AI applications and offer numerous interfaces (including MIPI-DSI, MIPI-CSI, LVDS, HDMI, GbE, PCIe, USB and CAN-FD).
Additionally, SBC solutions come in various form factors like SMARC, OSM and other SOM formats, allowing rapid integration into your own systems. The partner solutions cover various sizes, temperature ranges and operating systems (Yocto Linux, Android, Ubuntu). They enable the direct implementation of edge AI projects, making them particularly interesting for prototypes and small-to-medium production runs.
MediaTek’s Genio platform simplifies the development of edge AI applications thanks to a comprehensive technology stack:
- Nvidia TAO integration: Access to over 100 pre-trained models for image and speech recognition. These models can be adapted and deployed immediately for a variety of Vision AI applications, with no additional training required.
MediaTek’s NeuroPilot SDK is a powerful AI framework that enables efficient execution of AI models on MediaTek NPUs.
It offers:
- Seamless model conversion - for converting ONNX models to TFLite for optimized inference.
- Hardware acceleration - as used by MediaTek’s MDLA (Deep Learning Accelerator) for powerful AI processing
- Broad compatibility - for supporting multiple AI frameworks, including ONNX Runtime and PyTorch.
- Cross-platform SDK speeds up development and maintenance across the entire Genio family. Once developed, solutions can be flexibly transferred to other products, saving both time and money.
This combination greatly simplifies and accelerates the development, porting and scaling of AI models.
From home training to production line
MediaTek’s AI-enabled edge platform is already being used in smart retail, industry, health care and logistics. In any case, customers can reuse trained AI models, rapidly deploy them on OEM boards and maintain product lines over time – achieving minimal development deviation and maximum performance per watt. In addition, Rutronik provides access to a wide partner network in the field of embedded AI, including Advantech, Adlink, Seco and Via. They deliver powerful hardware solutions in multiple form factors, ready for direct use with the platform.
Two real-world use cases illustrate how edge AI projects can be scaled, from simple consumer products to high-end industrial systems:
- The interactive boxing workout FightCamp is based on the compact Genio 350 SoC and uses integrated motion sensors and a camera to capture punching techniques and posture locally, analyze them in real time and deliver instant feedback to the user without a cloud connection. Low power consumption, a compact design and sufficient AI performance (0.35 TOPS) make for a portable solution that delivers responsive user feedback. Thanks to NeuroPilot SDK, standard models from TensorFlow Lite, for example, can be run directly on the device.
- The Genio 720 is used in modern production lines with up to 9 TOPS of AI processing power and supports multiple high-resolution video streams, including up to eight virtual Full HD channels at 30 fps. The platform analyzes quality characteristics like surface defects, labeling or dimensional accuracy directly at the network edge. With HDMI In/Out, dual MIPI-CSI, triple display support and a robust software base, it integrates into existing industrial IT systems.
The fully equipped Genio 720 development kit (Figure 1) includes a touch display, camera interface, CAN-FD, M.2 slots and GPIO headers, making it the perfect foundation for AI-driven quality assurance in industrial applications.
Ready for the future of AI
The MediaTek Genio platform is well equipped for upcoming AI requirements at the edge. Thanks to the uniform NeuroPilot SDK, which integrates Nvidia TAO, new AI models can be deployed flexibly, efficiently and across platforms. Its modular design, support for various operating systems and long-term product availability secure investments while enabling rapid adaptation to future developments. The platform thus provides a future-proof foundation for innovative edge AI applications.