Executive Summary : | The rapid growth of technology has led to a need for more efficient and optimized computation methods, particularly in biological systems. Cognitive information processing in biological systems can provide a more efficient architecture than traditional Von Neumann architecture. The fourth fundamental element, the memristor, has changed the perspective of neuromorphic computation and opened a new yet efficient approach. However, challenges in designing memristor-based neuromorphic computation can be encountered at both device and circuit levels. Device level randomness causes a lack of uniformity, making the system vulnerable due to nonlinear dynamics of memristor switching. The fundamental challenges include the absence of a GPDK for memristor crossbars, the requirement of complex peripheral systems and interface systems for accessing the memristor crossbar, the unique scheme for reading and writing conductance values, the lack of standard procedure for reconfigurability of memristor crossbars, the limitation of crossbar array size for commercially available memristor crossbars and its operating conditions, and the integration of CMOS circuits with memristor in a single chip. The proposal proposes a methodology to address these challenges by defining a novel reading and writing scheme of memristor, reducing integrated circuits of peripheral and interface systems to support access to the memristor crossbar. The proposed method also includes computation for training and classification of images using single layer perceptron networks (SLP), with in-memory and parallel computation through the memristor crossbar. This generalization of standard procedure for reconfigurability and system-level approach will pave the way for further biology-inspired neuromorphic computation. |