An Automated Flow for Configuration and Generation of CNN based AI accelerators for HW Emulation & FPGA Prototyping

Machine learning (ML) algorithms have proven to be a concrete component in various fields that aim to be fully automated. Therefore, many researchers have shed the light on the modifications of ML algorithms to be fully automated for more complicated tasks. However, the acceleration of such algorithms is extremely hard due to the high computations and memory required. This paper implements automated flow using Perl scripts and generated LeNet-5 (A Convolutional Neural Network Model). Our target is high throughput, configurable and scalable RTL design that is generated by Perl scripts. Our flow is designing and verifying using Veloce emulator.