AndesAIRE (Andes AI Runs Everywhere) includes a first generation 8bit integer AI/ML hardware accelerator intellectual property (IP), the AnDLA I350 (Andes Deep Learning Accelerator), shown above, and the neural network software tools and runtimes, the AndesAIRE NN SDK
The core is aimed at low power AI inference at the edge of the network and at sensor end points for the Internet of Things (IoT) with added sensor fusion and analysis capabilities.
The AnDLA I350 supports the popular deep learning frameworks, including TensorFlow Lite, PyTorch, and ONNX, and performs versatile neural network operations such as convolution, fully-connect, element-wise, activation, pooling, channel padding, upsample, concatenation, etc. in the int8 data type.
The accelerator has been designed with an internal Direct Memory Access (DMA) and local memory to address the memory bandwidth issues of AI inference as well as four 64bit AXI bus interfaces.
The key configurable parameters of the core include the number of multiple accumulate units (MACs) which can scale from 32 to 4096, and the SRAM can scale from 16KB to 4MB. Overall, the core can provide 64 GOPS to 8 TOPS (at 1 GHz) for a wide range of applications.
Tools are a vital part of custom edge AI development. The AndesAIRE NN SDK is a comprehensive set of software tools and runtimes for end-to-end development and deployment. It includes a neural network optimization tool suite called NNPilot and an AnDLA-optimized inference framework running on a host based on TensorFlow Lite for Microcontrollers (TFLM) as well as the AnDLA driver and runtime for bare metal implementations.