Research Projects
Fault-Tolerant PIM accelerator
A novel fault-tolerant framework called FARe that enables on-device GNN training using ReRAM-based architectures..
Dynamic Resource Management
Machine learning based dynamic resource management policies for mobile, large-manycore systems and PIM-based architectures
Chiplet-based Architecture Design
A novel server scale 2.5-D manycore architecture called SWAP for deep learning (DL) applications.
DNN Training and Graph Learning at the Edge
In-memory acceleration of DNN training workloads.