The introduction of FPGAs in High-Performance Embedded Computing and Artificial Intelligence still faces chal-enges regarding the difficulty of getting started. It requires hardware knowledge, familiarity with multiple tooling, libraries and frameworks and long synthesis times. To encourage the usage of FPGAs, this work proposes an ecosystem that includes a library with a set of pre-built accelerators for common Digital Signal Processing and Artificial Intelligence workloads, an engine for runtime arbitrary-precision quantisation and an agnostic API, allowing the development of FPGA-accelerated user applications while abstracting the details about the FPGA design and implementation. Our approach is based on hardware reuse, introducing software resource management of a series of pre-built IP cores, allowing low-end FPGAs to be used as hardware accelerators and multiple applications to share resources. Our work is better than managed FPGA standalone applications with Vitis HLS-based quantisation, accelerating 1.22 x, thanks to our quantisation engine, which accelerates 5.12 x the quantisation and 13.30 x the de-quantisation, while keeping close the accelerator execution times.
Transformers networks have been a great milestone in the natural language processing field, and have powered technologies like ChatGPT, which are undeniably changing people’s lives. This article discusses the characteristics and computational complexity of Transformers networks, as well as, the potential for improving its performance in low-resource environments through the use of hardware accelerators. This research has the potential to significantly improve the performance of Transformers in edge and low-end devices. In addition, Edge Artificial Intelligence, Hardware Acceleration, and Tiny Machine Learning algorithms are explored. The proposed methodology includes a software and hardware layer, with a Linux-based minimal image built on top of a synthesized RTL. The proposal also includes a library of hardware accelerators that can be customized to select the desired accelerators based on the device’s resources and operations to be accelerated.