Energy Efficient Open-Source RISC-V Processor for Automotive Workloads

Sammanfattning

Modern vehicles integrate a growing number of electronic control units to performkey tasks, such as driver assistance, smart braking, and steer by wire. This presents achallenge for the transition towards battery electric vehicles as any power consumedby embedded systems cannot be used for driving, directly reducing vehicle range.As a result, energy efficiency is of central concern as automotive software growsmore complex. Another factor is ease of development, with closed systems requiringexpensive and specialized software for development, making them expensive anddifficult to maintain. Moreover, processors traditionally used in the automotivespace are proprietary which enables vendor lock in. Exploring alternative processorarchitectures has great potential for meeting the evolving requirements. The openand flexible nature of RISC-V makes it particularly well-suited to address theperformance, safety, and energy-efficiency challenges of modern automotive systems.The instruction set being open allows anyone to design a compatible processor whichcan enable competition and result in cheaper and more efficient designs. Additionally,the design can be tailored to the specific performance and power requirements of theapplication. Finally, software development can leverage open-source tools, reducingthe need for expensive licenses. This thesis investigates the suitability of the NOEL-V,an open-source RISC-V processor, for use in automotive systems. It is comparedto a commercial automotive ECU based on Infineon TriCore TC399XP in terms ofperformance and power consumption by porting a climate system to RISC-V andevaluating it on an FPGA implementation of NOEL-V. The performance is assessedthrough cycle count and schedulability of real-time tasks, while power consumptionon a 45 nm process is estimated using EDA tools. For the ECU, the performancemetrics of the TriCore are collected on the real hardware through Lauterbach debugtools, and power consumption is estimated indirectly via the system-level powerincrease during workload execution. The results show that although the NOEL-Vplatform does not achieve the same raw performance as the TC399XP, it is sufficientfor the climate application and requires fewer clock cycles per instruction. Theproject also identifies major power hotspots in the RISC-V processor under thetarget workload, and proposes directions on optimizing power dissipation for futureRISC-V-based automotive processors.