Domain-Specific FPGA-Based Digital Acquisition System for Hardware Accelerators Integration
Cosmin-Andrei Popovici,Andrei Stan,Vasile-Ion Manta
DOI: https://doi.org/10.2478/bipie-2023-0012
2023-06-01
Bulletin of the Polytechnic Institute of Iași. Electrical Engineering, Power Engineering, Electronics Section
Abstract:Abstract Domains like IoT (Internet of Things), ADAS (Advanced Driver Assistance System), Smart Homes and Smart Cities and XaaS (Everything-as-a-Service) put a great pressure in the last fifteen years on computer engineers for designing processors capable of higher precision computations performed in shorter times, interconnected in ultra-high-speed communication networks and consuming less electrical power in order to be supplied by batteries and clean renewable energy solutions. For almost two decades now, classical VLSI laws such as Moore’s and Dennard’s cannot be applied anymore for embedding twice as many transistors in the same silicon area every two years and keeping the energy density constant at the same time. The most feasible solution for mitigating this problem is designing DSAs (Domain-Specific Architectures) like application-specific hardware accelerators. Hardware accelerators replace the necessity of implementing domain specific algorithms programmatically by embedding hardware implementation variants of the same algorithms in modules running in parallel with the software developed for classical CPUs. An electronic field which can benefit from using domain-specific architectures instead of classical general-purpose microcontrollers is Digital Signal Processing (DSP). Designing logical circuits or complex digital systems won’t be possible today without using logic analyzers, protocol decoders, digital oscilloscopes, and digital recorders. These tools are used in the processes of prototyping and verifying digital systems by recording, displaying, decoding, and analyzing internal or external signals. This article proposes an example of a domain-specific architecture, a Digital Acquisition System, named FpgaDaqEth, acquiring data from 32 digital channels at 1 MS/s maximum acquisition rate, deployed on FPGA, operating at 100 MHz, using DDR3 RAM memory and 100 Mbps Ethernet for the communication with hosting PC running a custom GUI (graphical user interface) application. This design offers between 8x and 43.34x better propagation delay (constant 20 ns) for performing and outputting logical operations over multiple inputs (between 2 and 31 signals) than a general-purpose microcontroller which runs at 6x times higher frequency. It is also initiating UDP transmissions containing digital records and their timestamps 19.31x times faster than a Xilinx (now AMD) MicroBlaze processor implementation running at 2x times higher frequency on the same FPGA. The solution can detect changes in parameters of PWM inputs and send them to the hosting PC right after the first cycle of the modified signal ends. The proposed design is deployed on AMD XC7A100T FPGA as a standalone solution and may be integrated as processor extension for RISC-V cores in a future project.