Your idea is: Use machine learning to detect specific flaw in chips that have been identified as defective
Things you did well:
I think you have a great manageable scope. There is a very specific problem you are trying to solve and function that you are trying to take over.
You did an awesome job lining up the reader to understand the technology
The three improvement explanations in your second section are great
Things I would change:
I like your section titles but I think you could improve them by making it clear they are a sequence. Here is a (not very creative) example of what I mean:
-- The Old Way: Human Classifiers
-- The New Way: Convolutional Neural Networks
-- The Result: A Smarter Factory
Your first section is great! Very clearly introduces the problem. However, you don't actually say what's wrong with human classifiers (slow, prone to error, expensive). Make sure you add this!
In the middle section, although it is clear by context that you intend to replace the human classifiers with CNNs, you don't actually say it directly. I would recommend editing the first sentence of the second section to read "...the time is right to replace human defect classifiers with modern computerized technology such as..."
Make sure you go and read the assignment page that Brooks had for Big Idea #2. It's basically the same as #1 but he also wanted to see a precise research question and a description of the data and methods uses to make the product. You are missing both of these! Here's an example:
-- Question: What impact could CNN have on chip defect classification?
-- Data & Methods: use the same images the technicians currently use but apply them to a convolutional neural network. I don't really know how these work but I assume that you would probably train it on the same catalog of reference images or something similar.
I think a bit of detail on the implementation scope would be an interesting addition. Is going to be directly implemented in the production line to sort defects out from non-defective products? Or is this used to help determine which defects can be fixed vs which ones need to be scrapped? Or is this simply going to be a research tool where defective chips are taken in to help improve design?
Your last section could use some beefing up! This section is meant to explain how the solution helps and how it will be used in addition to some drawbacks. Maybe this is a better place for the three improvements you outlines (speed, accuracy, classification).
Overall comment:
This is a very cool idea and you did a good job of explaining everything clearly. I think you just left the reader to "read between the lines" on a lot of the major parts of this assignment so make sure you go back and fill it all in/be sure no question is left unanswered.
Your idea is: Use machine learning to detect specific flaw in chips that have been identified as defective
Things you did well:
Things I would change:
Overall comment: