Accelerating Edge AI Inference with Vitis AI NPU on iWave’s Versal AI Edge Boards

Edge Artificial Intelligence (AI) is rapidly transforming the way embedded systems process and respond to real-time data. To meet this growing demand for intelligence at the edge, iWave Global successfully integrated Vitis AI NPU (Neural Processing Unit) on the Versal™ AI Edge VE2302 System on Module (SoM) and its evaluation platform.

This integration demonstrates iWave’s commitment to empowering developers with high-performance, power-efficient, and ultra-low latency AI inference solutions. Built on the iW-RainboW-G57D SoM, the platform accelerates complex AI workloads such as real-time object detection delivering edge intelligence closer to where data is generated.

Vitis AI: Bringing FPGA Acceleration to AI Workloads

Vitis AI is AMD’s unified AI inference software stack designed for FPGA-based hardware platforms, including the Versal Adaptive SoCs. It enables seamless migration of AI models trained for GPUs to FPGA architectures without major rework.

Supporting popular deep learning frameworks such as TensorFlow, PyTorch, ONNX, and Caffe, Vitis AI allows developers to deploy trained neural networks efficiently on hardware optimized for parallelism and low power.

The result is an AI acceleration pipeline that merges FPGA flexibility with the performance of specialized neural accelerators—ideal for latency-critical edge applications.

System Overview and Demo Setup

The live demonstration of the Vitis AI NPU running on iWave’s VE2302 Versal AI Edge Evaluation Kit showcases real-time object detection capabilities.

Demo Components:

• iWave VE2302 Versal AI Edge Development Kit
• HDMI Display
• USB Webcam for live video input
• 12V/5A Power Supply and debug cable

Dataflow Process for AI Inference

The AI inference workflow on iW-RainboW-G57D, leverages the VART X API modules and structured for real-time processing. A script is created to run the NPU application on the SoM. This script captures video using the USB camera and converts it to NV12 format. The converted video is then processed by NPU and detected objects are highlighted on the HDMI display.

Dataflow process is as follows:

Live Demo in Action

Experience real-time object detection on VE2302 Versal AI Edge SoM! Watch the demo as iWave experts showcase AI inference using NPU IP for precise and efficient object detection.

Why Versal AI Edge for Edge AI?

The Versal AI Edge family from AMD combines programmable logic, AI Engines, and a heterogeneous processing system in a single chip. This architecture enables sensor fusion, vision analytics, and AI inference on one platform while maintaining deterministic real-time control.

Key highlights include:

• AI Engines & DSP Engines for vision, radar, and LiDAR workloads
• Native MIPI support for up to 8MP resolution
• Single and half-precision floating-point support for diverse AI and signal processing tasks

The combination of AI compute with programmable logic provides a scalable foundation for a wide range of edge AI use cases—from robotics to industrial automation.

Features of iWave’s Versal AI Edge System on Module

• Compatible with VE2302 / VE2202 / VE2102 / VE2002 devices
• Dual-core Arm Cortex-A72 and Cortex-R5F processors
• Up to 328K logic cells and 150K LUTs
• 8 GTYP transceivers at 32 Gbps
• Up to 8GB LPDDR4 RAM and 128GB eMMC storage
• Dual 240-pin high-speed connectors for expansion
• Connectivity: PCIe Gen4, Ethernet, USB 3.0

Measuring compact yet robust, the SoM supports 40G Ethernet, MIPI camera interfaces, and 122 configurable I/O, ensuring seamless integration into edge AI systems that demand high-speed data movement and real-time inference.

Real-World Applications

The Vitis AI NPU on iWave’s Versal AI Edge SoM unlocks new opportunities across a broad spectrum of industries:

• Smart Surveillance: Real-time object and facial recognition for intelligent monitoring
• Automotive & ADAS: High-speed detection for traffic signs and pedestrians
• Industrial Automation: On-device analytics for defect detection and predictive maintenance
• Healthcare: AI-driven diagnostic imaging and patient monitoring
• Smart Retail: Automated checkout and customer analytics
• Smart Cities: Adaptive traffic management using live video analytics

The solution merges high-throughput AI processing with power efficiency delivering precise, real-time performance in environments where milliseconds matter.

Empowering Developers with Edge AI Tools

iWave provides a full suite of software tools, libraries, and board support packages (BSPs) to simplify the AI development cycle on Versal platforms. With support for Vitis AI, OpenCV, and Linux-based development, engineers can deploy, test, and optimize AI workloads faster.

Backed by comprehensive documentation, long-term availability (10+ years), and ODM design services, iWave ensures that customers can scale from prototype to production with confidence.

iWave Global is a trusted engineering solutions provider specializing in FPGA-based System on Modules (SoMs) and ODM design services for industrial, automotive, medical, and defense markets. Leveraging decades of embedded expertise, iWave enables innovation at the edge through reliable, scalable, and high-performance hardware platforms.

To explore how the Versal AI Edge SoM can power your next AI innovation, visit www.iwave-global.com or contact mktg@iwave-global.com