Deliverable 5.1 Accelerators models and components: iteration 1

This report presents the progress and the development of several AI (Artificial Intelligence) accelerators, which are built and are evaluated on the robust and efficient system architecture modeling and simulation infrastructure of WP5. We present the comprehensive evaluation of complete computing systems integrating AI and accelerators of the NEUROPULS Horizon Europe project (Grant Agreement n° 101070238) along with RISC-V CPUs. The rise of artificial intelligence (AI) and neuromorphic computing has necessitated the development of specialized hardware accelerators capable of meeting the unique computational demands of these domains. In this context, this deliverable presents a comprehensive overview of our efforts to design, model, and simulate AI accelerators using the gem5 simulation infrastructure. The next iteration of this deliverable (i.e., the deliverable D5.2) will augment this deliverable more AI accelerators modeling and design space exploration, along with neuromorphic accelerators and the NEUROPULS neuromorphic accelerator modeling. This document details our methodologies and experimental results, focusing on performance, area, and reliability assessments of a famous AI accelerator (LeNet5) integrated with RISC-V CPUs. We highlight the implementation and simulation of prominent neural network models, specifically MobileNetV2 and LeNet-5, within our infrastructure. This report details the diverse objectives, including the execution of accelerators in a full system setup, the exploration of the design space of AI accelerators and the assessment of their critical design parameters such as performance, area, and resilience. This deliverable outlines the methodology, tools, and considerations involved in the establishment of the system architecture modeling and simulation of AI accelerators, highlighting its significance in advancing the understanding and optimization of computing systems