RAMAN: A Re-configurable and Sparse tinyML Accelerator for Inference on Edge