UCLA

Recent trends indicate that we will generate exponentially more data from a variety of devices. Neuromorphic computing (a.k.a. brain-inspired computing), is required to cognitively extract useful information in the Internet of Things (IoT). In order to achieve cognitive computing, ultra-large-scale systems that contain billions of neurons and trillions of synapses as the interconnection between those neurons will be needed. In this thesis, we first discuss system integration technologies and the scaling limitation of the von Neumann (vN) machines. Then, based on the system integration technologies, we propose neuromorphic computing systems with anon-von Neumann (NvN) architecture. The proposed systems are modeled, simulated, evaluated and compared with different system integration technologies. We show that the 3-dimensional-wafer-scale-integration (3D-WSI) technology is a potential candidate to enable neuromorphic computing systems at the scale of the human brain, while keeping the system latency and energy consumption at an acceptable level.