Three Dimensional Integration Technology Applied to Neuromorphic Hardware Implementation

Neuromorphic computation is recognized as a break-through technology which is very energy efficient and has a huge potential for significant contribution in hardware architecture using integrated circuit technology. Due to the huge parallelism between the neuron layers, hardware implementation of neuromorphic system using three dimensional (3D) integration could provide a sustainable and promising alternative to the existing conventional two dimensional (2D) planar integrated circuit. In this study, the neural architecture, their implementation in 3D model, associated limitation and design challenges are discussed. Through Silicon Via (TSV) is used in 3D chip which is a key technology that vertically stacks multiple neuron dies to make the signal path shorter and more concise. In TSV based 3D neuromorphic IC, crosstalk is an important issue that appears due to electromagnetic interference between the neighboring TSV. We analyzed the crosstalk and coupling between the proposed TSV arrangement of 3D neural structure by means of EM field simulation.