Bio-inspired Hardware Architectures for Memory, Image Processing, and Control Applications

CMOS scaling has been consistently providing increased density for the modern VLSI chips as predicted by the Moore's law, which dictates that the number of transistors in a semiconductor chip doubles approximately every two years. CMOS technology has facilitated Van Neumann architecture based computation paradigms to flourish and dominate the digital world for decades; however, as the transistor scaling is reaching its physical limits, and with the emergence of new technologies that provide interesting physical properties, alternative computation paradigms might need to be adopted in wide range of applications. The nervous systems of living organisms perform many complex tasks with much more energy and computational efficiency than the current VLSI chips. For example, human brain can perform 1017 FLOPS while dissipating around merely 15W. Therefore, bio-inspired neuromorphic computation paradigms have been attracting significant attention. Especially, mimicking neuron and synapse functionalities with CMOS circuitry has been the goal of many researchers in order to investigate if the computational efficiency of the biological systems can be attained on semiconductor chips.

In this talk, the circuit and system level applications of CMOS and variable resistance devices will be presented with bio-inspired computation paradigms as the main focus. The first application that will be presented is a non-volatile multi-level crossbar memory, capable of storing two or more bits per cell, facilitating the design of ultra-dense data storage. The second application is a programmable resistive grid acting as an artificial retina to realize various computer vision tasks such as edge and line detection. The final application that will be presented is a reinforcement learning inspired controller hardware capable of solving the optimal control problem for general non-linear systems.