Pyramid: Machine Learning Framework to Estimate the Optimal Timing - - PowerPoint PPT Presentation

▶
pyramid machine learning framework to estimate the
SMART_READER_LITE
LIVE PREVIEW

Pyramid: Machine Learning Framework to Estimate the Optimal Timing - - PowerPoint PPT Presentation

Oct 21, 2023 333 likes •515 views

Pyramid: Machine Learning Framework to Estimate the Optimal Timing and Resource Usage of a High-Level Synthesis Design Hosein Mohamamdi Makrani, Farnoud Farahmand, Hossein Sayadi, Sara, Bondi, Sai Manoj PD, Houman Homayoun, And Setareh

slide-1
SLIDE 1

Pyramid: Machine Learning Framework to Estimate the Optimal Timing and Resource Usage

  • f a High-Level Synthesis Design

Hosein Mohamamdi Makrani, Farnoud Farahmand, Hossein Sayadi, Sara, Bondi, Sai Manoj PD, Houman Homayoun, And Setareh Rafatirad Presenter: Rashmi Agrawal September, 2019 International Conference on Field-Programmable Logic and Applications (FPL 2019)

1

slide-2
SLIDE 2

HLS-Based Development and Benchmarking Flow

2 Outline: Motivation, Experimental Setup, XPPE, DSE, Conclusion

High-Level Synthesis HDL Code Manual Modifications (pragmas, tweaks) HLS-ready C code Reference Implementation in C

Not accurate!

slide-3
SLIDE 3

Static Timing Analysis using CAD Tools

3

Bitstream

Logic Synthesis, Technology Mapping, and Place&Route

Time consuming! Not Optimal!

slide-4
SLIDE 4

Solution

4

slide-5
SLIDE 5

Goal of this Research

5

  • We propose Pyramid, a framework that uses ML to

accurately estimate the optimal performance and resource utilization of an HLS design

  • For this purpose:
  • we first create a database of C-to- FPGA results from a

diverse set of benchmarks using Minerva to find maximum clock frequency

  • We use the database to train an ensemble machine learning

model to map the HLS-reported features to the results of Minerva

slide-6
SLIDE 6

General ML Models

6

ML models: regression model, artificial neural network (ANN), support vector machine (SVM), and random forest (RF) Tuning technique: Grid Search Average Error: LR=23%, ANN=14%, SVM=19%, and RF=15%

slide-7
SLIDE 7

Ensemble Model

7

Stacking Regression Key Idea: Cross Validation and Batch Training Model Parameters: threshold for the accuracy and the maximum number of iterations

slide-8
SLIDE 8

Evaluation Results

8

slide-9
SLIDE 9

Questions

Thank you!

9

slide-10
SLIDE 10

Supporting Slides

10

slide-11
SLIDE 11

Inaccuracy of HLS Report

11

slide-12
SLIDE 12

Challenge of Hardware Evaluation

12

slide-13
SLIDE 13

Impact of Optimization on the Outcome of HLS: Average of 10 numbers

13

slide-14
SLIDE 14

Using Minerva to Tackle Finding Optimal Results

14

Bitstream

Logic Synthesis, Technology Mapping, and Place&Route Minerva

  • F. Farahmand et al., “Minerva: Automated hardware
  • ptimization tool,”

in ReConFig, 2017.

Extremely time consuming!

slide-15
SLIDE 15

Studied Applications

  • Benchmarks:
  • Development suites:
  • Total number of applications:
  • Applications’ types:

15

Machsuit, S2CBench, CHStone, Rosetta, and xfOpenCV

Vivado and Vivado HLS version 2017.2

90 Machine learning, Image/Video Processing Cryptography, and Mathematical

slide-16
SLIDE 16

Data Collection

16

  • Total features: 72
  • Data set: 60% for training, 20% for Validation, and 20%

for testing (unseen data)