Adithya Krishna Murthy
Research Student - PhD
Research Program: International Centre for Neuromorphic Systems
Brain-Inspired Neuromorphic Compute Architectures for Edge Computing
Deep neural networks (DNNs) have become pivotal for solving several cognition and learning problems and have achieved unprecedented accuracy on many modern ML applications. DNNs are widely used in various edge computing platforms to perform direct inference on-chip, reducing latency, conserving bandwidth, and improving privacy. However, the state-of-the-art DNNs developed for an edge computing framework perform around a million operations and require KBs to MBs of memory to store activations and parameters. This becomes a significant bottleneck in deploying DNNs on an edge device constrained by power, area, and memory.
My research aims to address this issue by optimizing the entire software and hardware stack. We employ event-driven algorithms that inherently reduce the number of computations to save power and latency on the software front. On the hardware front, I propose to develop a low power, low area, low latency, and high throughput DNN architecture and implement it on an ASIC/FPGA. The developed architecture will leverage sparsity in parameters and activations and utilize low precision quantization to reduce the cost of memory, bandwidth, latency and overall data movement.
- B.Tech in Electronics and Communication Engineering from PES University, Bangalore, India (2017)
|Phone||+61 493 318 697|
|Location||Western Sydney University Penrith campus (Werrington South)|