Autobot for Effective Design Space Exploration and Agile Generation of RBFNN Hardware Accelerator in Embedded Real-time Computing

This paper presents a method of employing Autobot to replace humans in the task of efficient hardware design for radial basis function neural network (RBFNN) in real-time computing applications. Autobot applies quick iterations using hardware generation and supports various number systems such as floating-point, half-floating point, and mixed-precision and hardware architectures to perform possible design space exploration, enabling an agile analysis for those requests. We have implemented and employed Autobot to successfully test with the applications of RBFNN-based Mackey-Glass chaotic time series prediction, servo motor control, and data classification. Analysis of these results shows that Autobot is able to deliver the hardware accelerator with less execution time than previous works, which also shortens the design time from days to minutes. Therefore, the proposed methodology is a useful alternative for agile real-time hardware development on FPGA.