The NPU by OPENEDGES, known as ENLIGHT, is a cutting-edge neural processing unit tailored for deep learning applications requiring high computational efficiency. It introduces an innovative approach by utilizing mixed-precision computation (4/8-bit), heavily optimizing processing power and reducing DRAM traffic through advanced scheduling and layer partitioning techniques.
This NPU offers superior energy efficiency and compute density, making it significantly more effective than competing alternatives. It is highly customizable, accommodating varying core sizes to meet specific market demands, ensuring a broad application reach from AI and ML tasks to edge computing requirements.
ENLIGHT enhances performance with its DNN-optimized vector engine and advanced algorithm supports, including convolution and non-linear activation functions. Its toolkit supports popular formats like ONNX and TFLite, simplifying integration and accelerating the development process for complex neural network models in high-performance environments.
