Personal tools

Difference between revisions of "Developing a Transposition Unit to Accelerate ML Workloads (1-3S/B)"

From iis-projects

Jump to: navigation, search
 
(No difference)

Latest revision as of 21:07, 2 August 2022


Overview

Status: Available

Introduction

Transposing matrices is an important operation used in countless applications from scientific computing to machine learning workloads.

At IIS, we are actively developing a DMA engine to accelerate data movement in various of our platforms. We would now create a transposition unit that transposes matrices while they are copied throughout the system. The accelerator should work of full-precision integer and floating point formats for general purpose scientific computing as well as narrow 4bit / 2bit (and even 1bit?) typed for ML interference.

Project

In this project, you develop, implement, and evaluate a flexible transposition unit able to work on various data widths.

Character

  • 40% Design and implementation of the unit
  • 30% Verification
  • 30% Evaluation

Prerequisites

  • Interest in memory systems
  • Experience with digital design in SystemVerilog as taught in VLSI I
  • Preferred: Knowledge of AXI4

References