MeMPA: A Memory Mapped M-SIMD Co-Processor to Cope with the Memory Wall Issue
1. IntroductionElectronic systems need more and more computational power to fulfill the increasing complexity of modern applications. Such applications, like Neural Networks, for example, require the execution of a vast number of instructions in parallel, implying stringent constraints on the design of application-specific accelerators or CPU-centric architectures. Nowadays, the von Neumann paradigm is the most diffused computing approach: it consists of a Central Processing Unit (CPU) and memor...
Read more at mdpi.com