Principal Investigator: Assistant Professor Gong Xiao, ECE

Co-Principal Investigator: Professor Aaron Thean, CDE

The data intensive and data centric computing calls for a paradigm shift of computing hardware platform to address tremendous but important challenges facing future computing that cannot be merely answered by commonly pursued approaches such as continuous down-scaling and performance improvement in Si-based transistors, distributed and parallel computing based on von- Neumann architecture, and 2D/2.5 system integration approach. There is an urgent need to search for innovative ways to tackle ‘the memory wall’ where a large amount of data needs to be fetched and stored during the computation which is both time and energy consuming.

Two effective solutions would be (1) co-locating computation blocks and memories for highband width data traffic; and (2) empowering memories with computational capability such as neuromorphic computing where chips are directly inspired by biological neural circuits so that they can process new knowledge, adapt, and learn in real time at unprecedented low power levels. A neuromorphic architecture stores the data (weights) in the non-volatile memory devices (analog synapses) and uses crossbar structures to realize large-scale matrix operations, which is similar to the behaviour of the synapses and neuron in the human brain. Essentials to the neuromorphic architecture include the analog synapse device with multiple states, low latency, high endurance, and high retention, as well as the integration of these high density synapse devices with peripheral circuits in a 3D manner.

In this proposal, we aim to work on amorphous-oxide semiconductor (AOS) thin film transistors (TFTs) and dynamic random access memory (DRAM) having back-end-of-line (BEOL) compatible processes, extremely low power, good scalability, and low cost, to enable efficient near and inmemory computing in a 3D monolithic integration manner that promises to deliver significant improvement in latency, energy, and cost as compared with the state-of-the-art nano-electronic systems, as shown in the figure below. Our proposed work would cut across the disciplines of materials, devices, circuits, and systems and is highly relevant to many major industry players and promises to spearhead the transformation of nano-electronic hardware platform for future computing.