Wafer Scale Integration for Massively Parallel Memory-Based Reasoning

Book One

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 10

Scaling Up

In this paper, we describe a design of wafer-scale integration for massively parallel memory-based reasoning (WSI-MBR). WSI-MBR attains about 2 million parallelism on a single 8 inch wafer using the state-of-the-art fabrication technologies. While WSI-MBR is specialized to memory-based reasoning, which is one of the mainstream approachs in massively parallel artificial intelligence research, the level of parallelism attained far surpasses any existing massively parallel hardware. Combination of memory array and analog weight computing circuits enable us to attain super high-density implementation with nanoseconds order inference time. Simulation results indicates that inherent robustness of the memory-based reasoning paradigm overcomes the possible precision degradation and fabrication defects in the wafer-scale integration. Also, the WSI-MBR provides a compact (desk-top size) massively parallel computing environment.