Andes’ new deep-learning accelerator addresses convolutional neural networks in edge applications. Accompanied by vector CPUs, it forms an AI subsystem that can be scaled up for higher vision- and audio-workload performance. Capable of 8 TOPS peak AI performance, the AndesAire AnDLA unit runs autonomously once set up, allowing control and other CPUs to perform additional tasks in parallel. It’s a simple core, with no bells and whistles, handling only INT8 data—at least in this first edition. Software support is basic, but it provides flows both for SoC developers putting fixed functions onto their chips and for end users with TensorFlow Lite for Microcontrollers as a model interpreter. Pairing with a vector CPU provides a fallback for nonlinear or other functions that AnDLA currently lacks but may support in the future.