The RAMAN accelerator is designed to leverage data and weight sparsity to deploy deep neural networks at the edge, ensuring low power consumption, minimal storage requirements, and reduced processing latency. 100x more energy-efficient than industry standard GPUs, Mythic's analog processing units (APUs) promise a new era of accelerated computing across the AI hardware stack, at the data center and the edge. Figure 1: Top-level architecture The key features of the RAMAN accelerator are: Sparsity: RAMAN leverages activation and weight sparsity in (a) Reducing latency by. Researchers at the Department of Electronic Systems Engineering, IISc, led by Chetan Singh Thakur, have developed an AI co-processor called RAMAN, or Re-configurable And sparse tinyML Accelerator for infereNce. RAMAN is an indigenous low-power AI co-processor designed for edge computing. Many near-sensor machine learning (ML) approaches have been implemented to introduce accurate and energy efficient template matching operations in resource-constrained edge sensing systems, such as wearables. Sparsity, in both activations and weights inherent to.
[PDF Version]