Accelerating DSP Workloads on RISC-V Cores and PPA Analysis

  • Unique Paper ID: 211192
  • Volume: 13
  • Issue: 6
  • PageNo: 3682-3695
  • Abstract:
  • Small RISC-V cores in edge devices spend most of their cycles in a few digital signal processing (DSP) loops. These loops can be accelerated inside the pipeline (in-core), by a bus-attached engine (off-core), or by both. Published designs usually report one bundled extension, so the value of each mechanism stays hidden. This paper measures every mechanism separately and in combination on a single system-on-chip. From one five-stage RV32IM core we derive 17 configurations: two baselines, five in-core variants (packed-SIMD multiply–accumulate, post-increment loads, zero-overhead loops, complex arithmetic), five off-core engines (memory-mapped MAC, DMA dot product, scratchpad GEMV, 16-tap streaming FIR, 64-point FFT) and five hybrids. All 17 run four Q15 kernels, a 45-workload sweep and five multi-stage applications (ECG, speech, image, radar, OFDM). Every run is checked bit-exactly against a golden model and against MATLAB double precision. Power, performance and area (PPA) are measured from gate-level simulation and an open-source RTL-to-GDSII flow on SkyWater 130 nm. All 17 layouts are DRC-clean and LVS-matched and meet hold timing in all nine corners. The best hybrid needs 7.48× fewer cycles and 5.0× less energy than RV32IM at 3.5× the area, and its energy–delay product is 37.5× better. After layout the clock is set mainly by repair buffering on long nets, not by the DSP logic.

Copyright & License

Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

BibTeX

@article{211192,
        author = {Likhith G and Kovuru Sai Karthik and G P Vishwas and G Mallikarjuna Gowda and Dr. Poornima G},
        title = {Accelerating DSP Workloads on RISC-V Cores and PPA Analysis},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {6},
        pages = {3682-3695},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=211192},
        abstract = {Small RISC-V cores in edge devices spend most of their cycles in a few digital signal processing (DSP) loops. These loops can be accelerated inside the pipeline (in-core), by a bus-attached engine (off-core), or by both. Published designs usually report one bundled extension, so the value of each mechanism stays hidden. This paper measures every mechanism separately and in combination on a single system-on-chip. From one five-stage RV32IM core we derive 17 configurations: two baselines, five in-core variants (packed-SIMD multiply–accumulate, post-increment loads, zero-overhead loops, complex arithmetic), five off-core engines (memory-mapped MAC, DMA dot product, scratchpad GEMV, 16-tap streaming FIR, 64-point FFT) and five hybrids. All 17 run four Q15 kernels, a 45-workload sweep and five multi-stage applications (ECG, speech, image, radar, OFDM). Every run is checked bit-exactly against a golden model and against MATLAB double precision. Power, performance and area (PPA) are measured from gate-level simulation and an open-source RTL-to-GDSII flow on SkyWater 130 nm. All 17 layouts are DRC-clean and LVS-matched and meet hold timing in all nine corners. The best hybrid needs 7.48× fewer cycles and 5.0× less energy than RV32IM at 3.5× the area, and its energy–delay product is 37.5× better. After layout the clock is set mainly by repair buffering on long nets, not by the DSP logic.},
        keywords = {Design-space exploration, Digital signal processing, Hardware accelerator, Open-source EDA, PPA analysis, RISC-V},
        month = {November},
        }

Cite This Article

G, L., & Karthik, K. S., & Vishwas, G. P., & Gowda, G. M., & G, D. P. (2026). Accelerating DSP Workloads on RISC-V Cores and PPA Analysis. International Journal of Innovative Research in Technology (IJIRT), 13(6), 3682–3695.

Related Articles