Agenda

1 What is AI, ML, DL & DS – Terminology Know How

2 Machine Learning Overview

3 Life Cycle of a Typical ML Model

4 AI Adaptability in Semiconductors

5 Top Trends in Semiconductors

6 Why AI makes sense in Semiconductors

7 Semiconductor Ideal Expectations from AI

8 AI Technology Verticals to meet The Expectations

9 Digital Image Processing – CNN Architecture

What is AI, ML, DL and DS

Artificial Intelligence (AI)

Programs with the ability to learn and reason like humans ~80%

Machine Learning (ML) ~80% of World’s

Algorithms with the ability to learn without being explicitly programmed Data generated in
last 10 Years

~80% of Resources
Deep Learning (DL) improvement in last
Subset of ML in which artificial neural networks adopt and 10 Years
learn from vast amount of data

Data Science (DS)

DS is a field of study which combines Statistics
and Math.
Data Science
Programming skills – Python / R integrates all the
And most importantly Domain expertise to extract above AI, ML & DL to
meaningful insights from data extracts insight from
data (EDA) and make
predictions from large
datasets (Predictive
Domain Programming
Statistics Analysis)
Knowledge Skills
What is AI, ML, DL and DS

Supervised Learning Unsupervised Learning Reinforcement Learning

Classification Clustering Agent

Regression Association Reward

Optimization Problem Math's Methods: Linear Algebra | Calculus

Supervised Learning – Classification

No
Diabetes
Model

ML

Classification

Diabetes

Important Algorithms for Classification

1. Logistic Regression
2. Naïve Bayes’ Classifier
3. K-Nearest Neighbors – KNN
4. Decision Tree
5. Ensemble Methods: Bagging & Boosting
6. Random Forest
7. Support Vector Machine
8. ………
Supervised Learning – Regression

Important Algorithms for Regression

1. Linear Regression
2. Multiple Linear Regression
3. Decision Trees
4. Random Forests
5. ………
Unsupervised Learning – Clustering
Pre Clustering Post Clustering

Clustering Important Algorithms: Applications

• Find elements (rows, • Recommendation engines
• K-Means
tuples) which are similar. • Market segmentation
• Finding “areas” in space • Fuzzy C-means • Social network analysis
where data is concentrated • Expectation Maximization • Search result grouping
• WYSIWYG : What You • Hierarchal Clustering • Image segmentation
Select Is What You Get • …… • Anomaly detection
Unsupervised Learning – Association

What does the value of one feature tell us Association Rules beyond Market Basket
about the value of another feature? Analysis
People who buy diapers are likely to buy baby
• People who visit webpage X are likely visit
powder
webpage Y.
If (people buy diaper), then (they buy baby powder)
• Nodes which run a web server are likely to
run Linux.
Association rules
• People who have age-group [30,40] & income
• Are statements about relations among features
[>$100k] are likely to own home
(attributes) : across elements (tuples)
• Use a transaction-item set data model
Machine Learning Model Flow
Model Tuning Model Tuning

Gathering Data Data Analysis Data Pre-processing Variable Selection

ML Model Building
(Feature Engineering) (Feature Selection)
• Collecting the data is Answer questions such
Our goal here is to make • Select a subset of Select the ML Model
first and an essential as: family for training which
step towards any • what variables are the data ready for building features out of all
ML models! To this end, features which is suited to best of our
data science and available? need:
many things can be done critical to ML model
Machine Learning • how are they related? performance.
but not limited to such as
project. • what is the • Linear Regression
characteristics of • Filling missing values in • Many feature could • Logistic Regression
• It can be from those variables? the data just be noise so • KNN
business units or (numerical or • Dealing with (e.g., removing them is • DT
from public datasets categorical?) removing) outliers important for • RF
• Ensembles
for building ML • missing values? • Transforming categorical overfitting and
values complexity reduction • ….
models. outliers?
etc.
AI Adaptability in Semiconductors | Market Research

Semiconductor Industry Leads in Artificial Intelligence Adaption, Accenture Report Finds

https://newsroom.accenture.com/industries/electronics-high-tech/semiconductor-industry-leads-in-artificial-intelligence-adoption-accenture-report-finds.htm
Market Opinion

“AI will be a major growth driver for the semiconductor industry in light of high
manufacturing costs and the growing complexity of chip development,”

“To capture this opportunity, chipmakers should leverage AI technologies and partnerships
to increase efficiency across their operations.”

---Syed Alam, Managing Director at Accenture, leads its Semiconductor practice globally.
Top Trends in Semiconductor Industry

Top Trends in Semiconductor Industry to watch out for this year

https://www.businesswire.com/news/home/20180524005772/en/Key-Semiconductor-Industry-Trends-Infiniti-Research
Why AI make sense in Semiconductors

As you can see there is one clear conclusion - the design and manufacturing of advanced node (10 nm and
below) are HUGE and increasing exponentially.

https://www.quora.com/How-much-does-it-cost-to-design-and-fabricate-an-integrated-circuit
Semiconductor Ideal Expectations from AI

No repeated iterations

Tools should not return unexpected results

Achieve predictability from the user’s point of view

Use old design data to improve new designs

Focus on reducing design time , design efforts

What is the solution?

Machine Learning will be a key piece of this . . .

AI Technology Verticals used in Semiconductors

Natural Language Processing Computer Vision Machine Learning

Text Processing Technique Computer Vision or CV is a

ML has wide range of
to process raw and digital Image Processing
application which fall in
unstructured text to Technique where we apply
predictive modelling as well
generate information. different Neural Network
as time series forecasting.
Architecture to process image
Ex.
Ex. and classify them.
1. Anomalies Detection
1. Q & A Bot Ex.
2. Predicting upcoming
2. Sentiment Analysis 1. Wafer Defect
market trends
3. Feedback Rating Classification
3. Stock price forecasting
System 2. Congestion Prediction
Etc...
4. Document Retrieval 3. Image to Image
Types: Classification &
Generation
Regression
Important Libraries: NLTK, Etc…
spaCy. Neural Net: CNN
Use-case 1 (Congestion Prediction)

Eliminate human and fixed model subjective bias with statistical important features

Only know routing results after running routing Able to predict routing behavior and improve congestion
with new recipes to achieve 40Mhz

http://www.ispd.cc/slides/2018/s4_1.pdf
Use-case 2 (Wafer Defect Classification)

Test Data

Image Trained
. Augmentation Model
.
.
. CNN
DEFECT CLASS
PREDICTION

Train Data

Defect Images
(15)
Introduction to Image Classification

Input Deep Neural Output

Image Network Models Class

Automobile
Manual
Classification

Airplane
Distance
Metrics

Convolution Bird
al Neural
Network

Ship
How Images are Represented

28 x 28 image
28 columns

28 rows

Pixel Value (brightness) range : 0 - 255

Major Challenges in Image Processing

➢ Size of the input data.

➢ Suppose an image is of the size 68 X 68 X 3. The input feature dimension then becomes
12,288.

➢ This will be even bigger if we have larger images (say, of size 720 X 720 X 3).

➢ Now, if we pass such a big input to a neural network, the number of parameters will
swell up to a HUGE number (depending on the number of hidden layers and hidden
units).

➢ This will result in more computational and memory requirements – not something to
be handled easily.

➢ Convolutional Neural Network solves all these challenges in a step by step Approach.
Architecture of Convolutional Neural Network – CNN
1. Convolution Layer - Edge Detection

https://www.analyticsvidhya.com/blog/2018/12/guide -convolutional-neural-network-cnn/
1. Convolution Layer - Edge Detection

1 1 1 0 0
0 1 1 1 0
0 0 1 1 1 1 0 1
0 0 1 1 0 0 1 0
0 1 1 0 0 1 0 1

Image Filter
1. Convolution Layer - Edge Detection

………

Image Convolved Features

What Weight Matrix or Filter does?

➢ The weight matrix behaves like a filter in an image extracting particular information from
the original image matrix.

➢ One weight combination might be extracting edges, while another one might a
particular color, while another one might just blur the unwanted noise.

➢ When we have multiple convolutional layers, the initial layer extract more generic
features, while as the network gets deeper, the features extracted by the weight
matrices are more and more complex and more suited to the problem at hand.
Importance of Padding in Convolution Layer

➢ Every time we apply a convolutional operation, the size of the image shrinks

➢ Pixels present in the corner of the image are used only a few number of times during
convolution as compared to the central pixels.

➢ Hence, we do not focus too much on the corners since that can lead to information loss

➢ To overcome these issues, we can pad the image with an additional border, i.e., we add
one pixel all around the edges.

➢ This is where padding comes to the fore

Strided Convolutions

➢ Stride is nothing but the steps that filter takes while convoluting over the image matrix.

➢ For example if stride is 2 then filter will jump two steps while convoluting.

➢ Stride helps to reduce the size of the image, a particularly useful feature.
2. Pooling Layers

➢ Pooling is another method to reduce the size of the image and hence speed up the
computation.

➢ Consider a 4 X 4 matrix as shown below:

➢ Applying max pooling on this matrix will result in a 2 X 2 output:

➢ For every consecutive 2 X 2 block, we take the max number.

➢ Here, we have applied a filter of size 2 and a stride of 2.
➢ Another method could be the average pooling instead of Max pooling.
➢ These all are the hyper parameters which we set to fine tune the model.
Formula to calculate the feature size

Input: n X n X nc
Filter: f X f X nc
Padding: p
Stride: s
Output: [(n+2p-f)/s+1] X [(n+2p-f)/s+1] X nc’

Here, nc is the number of channels in the input and filter, while nc’ is the number of
filters.
3. The Output layer
Future of Semiconductors with AI

https://venturebeat.com/2020/04/23/google-claims-its-ai-can-design-computer-
chips-in-under-6-hours/
Contact Details

Website: https://ashutoshtripathi.com Website: https://enetwork.ai

LinkedIn: https://www.linkedin.com/in/ashutoshtripathi1/ Instagram: https://www.instagram.com/ashutosh_ai/