Morphable RISC-V CPU Predictive Tree-Based Perceptron Pipeline

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Author: Chun, Dawit
Format: Recurso digital
Published: Zenodo 2025
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866901613843578880
author Chun, Dawit
author_facet Chun, Dawit
contents <p><strong>Abstract</strong></p> <p>I present a morphable RISC-V processor capable of dynamically switching between 3-stage (P3), 5-stage (P5), and 7-stage (P7) pipeline configurations to optimize performance across diverse workloads. Shallow pipelines minimize branch misprediction penalties but limit clock frequency, while deep pipelines enable higher frequencies at the cost of increased branch overhead. The design uses bypass multiplexers to reconfigure pipeline depth at runtime, guided by a hybrid AI predictor combining decision trees with perceptron networks. The predictor monitors workload characteristics—branch frequency, CPI overhead, and stall rates—to automatically select the optimal configuration, achieving 90+% prediction accuracy across ten benchmarks. Implemented on a Lattice iCE40 FPGA, the system demonstrates clock frequencies of 4 MHz (P3), 8 MHz (P5), and 12 MHz (P7), with measured performance improvements on both branch-heavy and compute-intensive workloads.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18006580
institution Zenodo
language
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle Morphable RISC-V CPU Predictive Tree-Based Perceptron Pipeline
Chun, Dawit
<p><strong>Abstract</strong></p> <p>I present a morphable RISC-V processor capable of dynamically switching between 3-stage (P3), 5-stage (P5), and 7-stage (P7) pipeline configurations to optimize performance across diverse workloads. Shallow pipelines minimize branch misprediction penalties but limit clock frequency, while deep pipelines enable higher frequencies at the cost of increased branch overhead. The design uses bypass multiplexers to reconfigure pipeline depth at runtime, guided by a hybrid AI predictor combining decision trees with perceptron networks. The predictor monitors workload characteristics—branch frequency, CPI overhead, and stall rates—to automatically select the optimal configuration, achieving 90+% prediction accuracy across ten benchmarks. Implemented on a Lattice iCE40 FPGA, the system demonstrates clock frequencies of 4 MHz (P3), 8 MHz (P5), and 12 MHz (P7), with measured performance improvements on both branch-heavy and compute-intensive workloads.</p>
title Morphable RISC-V CPU Predictive Tree-Based Perceptron Pipeline
url https://doi.org/10.5281/zenodo.18006580