AIVA®3D AI ASIC is designed with the heterogeneous architecture, combining 3D Depth engine and AI inferencing engine. It introduces a new-generation neural network inference fusion engine (NPU), the latest 3D visual depth sensing computing engine (VDE), and an accelerator visual tracking engine (VTE), Enabling the ASIC to lead the consumption area in terms of real-time high-performance and power efficiency.
AIVA has a very powerful and flexible 3D vision technology: it can measure depth information in the range of 0.2 meters to 6 meters at a baseline of 50 mm, which improves the depth error to less than 1% at 1 meter, enabling intelligent device the ability to understand and perceive the world in 3D.
It allows improved environmental awareness: exposure, backlighting, dimness, and other complex environments.
NPU (Neural Network Inference Engine)
It supports PyTorch, MXNet, TensorFlow, and other mainstream framework models. The resource reusability of this ASIC architecture enables flexible allocation of hardware computing units to adapt to the computing needs of different scenarios.