You are on page 1of 20

Design & implementation of deep learning

neural networks based on convolutional &


Laplacian filter with image controlling on
DLAU controllers

B.Deepa (17Q61D5708)
M.TECH (VLSI )
Contents
ABSTRACT

INTRODUCTION

LITERATURE REVIEW

EXISTING SYSTEM

PROPOSED SYSTEM

RESULTS

SYNTHESIS REPORTS

COMPARISON TABLE

IMPROVEMENTS OBSERVED

CONCLUSION
ABSTRACT
 In this paper we design DLAU, which is a scalable
accelerator architecture for large-scale deep learning networks for
CNN structure using DLAU controller.
The DLAU accelerator employs three pipelined processing
units to improve the throughput and utilizes techniques to explore
locality for deep learning applications.
INTRODUCTION
 In the past few years, machine learning has become
pervasive in various research fields, applications and
achieved satisfactory products.
 The emergence of deep learning speeded up the
development of machine learning and artificial intelligence.
 Consequently, deep learning has become a research hot
spot in research organizations.
 In general, deep learning uses a multilayer neural
network model to extract high-level features which are a
combination of low-level abstractions to find the
distributed data features, in order to solve complex
problems in machine learning.
EXCITSING SYSTEM
TMMU , PSAU & AFAU
TMMU is in charge of multiplication and
accumulation operations.
PSAU is responsible for the accumulation operation.
PSAU architecture, which accumulates the part sum
produced by TMMU.
 If the part sum is the final result, PSAU will write the
value to output buffer and send results to AFAU in a
pipeline manner.
Conc………………
Finally, AFAU implements the activation function
using piecewise linear interpolation .
This method has been widely applied to implement
activation functions with negligible accuracy loss. In
activation function implementation of sigmoid
function is done.
PROPOSED SYSTEM
Description
In order to explore the locality of the deep learning
application, we employ tile techniques to partition the
large scale input data.
The DLAU architecture can be configured to operate
different sizes of tile data to leverage the trade-offs
between speedup and hardware costs.
Consequently the FPGA based accelerator is more
scalable to accommodate different machine learning
applications.
DLAU Controller
Description
The DLAU accelerator is composed of three fully
pipelined processing units, including TMMU, PSAU,
and AFAU.
Different network topologies such as CNN, DNN, or
even emerging neural networks can be composed from
these basic modules.
Consequently the scalability of FPGA based
accelerator is higher than ASIC based accelerator.
CNN (Convolution with laplacian filter stage 1)
if(enable) begin output1 <= 0;
output2 <= 0; output3 <= 0;
output4 <= 0; output5 <= 0;
done <= 1'b0; end
else begin
output1 <= {1'b1, ~(input2)} + 5'b00001;
output2 <= {1'b1, ~(input4)} + 5'b00001;
output3 <= {2'b00, input5} << 2;
output4 <= {1'b1, ~(input6)} + 5'b00001;
output5 <= {1'b1, ~(input8)} + 5'b00001;
done <= 1'b1;
FSM diagram for AFAU
RESULTS
SYNTHESIS REPORT
Area Utilization Power Utilization
COMPARISON TABLE
IMPROVEMENTS OBSERVED
CONCLUSION
In current design scenario we have seen Author () have proposed
DLAU with CNN architecture with 2x2 and 4x4 as separate
models
Our design criteria improvises a novel concept of CNN structure
for both 2x2 and 4x4 for each internal modules utilized in design.
This proposed structure utilized only 2 L-Filters, 3 adders and 1
max-pooling circuits where as for exitisng scenario we need to
provide at least 2 filters and 8 adders with one max-pooling
modules.
We propose the design of deep learning accelerator unit using
FSM which can reduce power consumption and speed up device.
THANK YOU

You might also like