AI Inference

The cube:evk is built on the NXP i.MX 8M Plus, a highly capable edge-AI SoC featuring a quad-core Arm Cortex-A53 @ 1.8 GHz and an integrated Neural Processing Unit (NPU) delivering up to 2.3 TOPS. This dedicated NPU enables machine-learning workloads to run directly on the device, dramatically reducing CPU usage while providing real-time inference even under tight latency constraints.

The platform provides support for LiteRT (TensorFlow Lite), enabling efficient on-device inference with low latency and a compact binary footprint. LiteRT is an open-source software library designed for executing machine learning models on mobile and embedded devices.

VX Delegate

LiteRT provides multiple options for executing compute operations across different compute units. One of these is the VX Delegate, which streamlines accelerated inference by offloading supported operators to the on-chip hardware accelerator embedded in the i.MX 8 series.

Operating in direct conjunction with the hardware accelerator driver (OpenVX with extensions), the VX Delegate optimizes the utilization of the accelerator’s capabilities. Available as an external delegate, it is a specialized TensorFlow Lite delegate that is initialized by loading a dynamic library encapsulating the actual delegate implementation.

The corresponding library is located at /usr/lib/libvx_delegate.so.

Inference Example

Developers can start immediately using our public example repository:

👉 https://github.com/cubesys-GmbH/tflite-inference-example

This example demonstrates image detection with bounding boxes, showing how to load TFLite models, run inference on the NPU, and visualize detections.

With this architecture and tooling, the cube:evk can easily execute advanced perception workloads such as object detection, classification, scene segmentation, or sensor fusion — right at the edge, without cloud dependency or external accelerators. This makes the platform an excellent fit for V2X-enhanced applications where reaction time, reliability, determinism, and privacy matter.

Coupled with the cube:evk’s rich connectivity — Ethernet, CAN, GNSS/IMU, and native DSRC / C-V2X link-layer support — the NPU enables a new class of intelligent cooperative agents: vehicles, VRUs, robots, and roadside units capable of understanding their environment locally and sharing that insight via V2X in real time.