Learnings from Semiconductor Manufacturing Workshop by SemiX

Key Learnings from Semiconductor Manufacturing Workshop by SemiX, IIT Bombay

We had a workshop on Semiconductor Manufacturing conducted by SemiX, IIT Bombay, where I learned so many new things. Here I present to you some of those learnings. There were many technical learnings of what is happening in the research areas, technical stuff going on, and many non-technical learnings. So I will put them into two parts. First discuss about non-technical learnings, that is applicable to all of our lives and then move on to technical stuff in this blog.

Non-Technical Stuff

India's Need for Semiconductor Manufacturing

India needs semiconductor chip manufacturing, because if we are denied them, then it is not good for us, as chips are in nearly all of the electronic devices we use — from phones, to laptops, to cars, to aeroplanes, to rockets. Everywhere they are used and required.

Start Small, Scale Smart

Instead of going large initially and trying to do risk mitigation on questions being asked to give you money, go small, make it work and then expand — it would be better.

The Uniqueness of Human Thinking

What makes us humans unique? There are multiple things but one very important thing is our capability to think. This capability to think cannot be beaten by AI or anything. We have built complex things by thinking. And our thinking capability degrades when we don't give it exercise. It also needs exercise to maintain it. You can't outsource thinking.

AI can beat us if we don't give exercise to our brains. Use AI — but you are doing the main thing.

What stops us from thinking is the freely available resources. We are just consuming and consuming information, and not developing our thinking muscles.

Information is ok; you need to know stuff. But if you can think, you can come up with ideas and ways which you can't get just by having information.

You need to become a person who can't be beaten by AI.

A Big Problem to Solve

Long lasting, completely autonomous (energy as well as w.r.t. human interference) sustainable monitoring system, without creating any toxicity in the environment.

Why? These can help us to make decisions driven by data.

Communication Tips

When giving a speech, to make it impactful, give referencing.

Essential Capabilities

• Learnability — ability to learn
• Adaptability — to new situations

What Separates Good Engineers

1. Ability to debug effectively

• Through logical linking
• Designing experimentation
• Doing the modelling

2. Articulation

• Able to say and write what you mean precisely
• Clearly communicate

Design Philosophy

Do design thinking, with beauty in mind.

Debugging Strategy

When you have a fault in your product and want to find and resolve it, think this way:

Try to think of the logic that you need to implement to have that fault as if you were adding it as a feature — then you come across that flawed logic.

Logical Estimation

Develop the ability to make a logical guess/estimation. That would be helpful.

Technical Stuff

Tesla's Success Story

Tesla is popular because on a single charge it can go around 400 km — and this ability is there because of power management, made possible by power semiconductors.

Neuromorphic Computing — Prof. Udayan Ganguly

The Energy Problem with AI

AI guzzles power — to train and use an AI model requires a lot of energy. Projects are coming up to develop nuclear power plants to power these AI machines, but these would dissipate too much heat and are not sustainable.

Natural vs Artificial Intelligence Comparison

Taking example of Chess Game by Watson:

• Watson → a supercomputer → consuming 10 MW of power
• Human → consuming around 10 W of power

This shows the energy efficiency of our brain. If we can compute as efficiently as our brains, we can be far more energy efficient. Neuromorphic Computing aims to do so.

Spiking Neural Networks (SNN)

We are not talking about ANN (Artificial Neural Networks) here but SNN (Spiking Neural Networks). In our brain, the token of information is a spike — an alternative way of representing information. Anything humans can perceive can be represented with these spikes.

Information is encoded like a tabla or drum performance — each click is the same, but the difference lies in their timing.

Circuit Analysis Exercises

Exercise 1: Identifying Circuit Elements

Answer: Resistor

Exercise 2: Circuit Function

Input and output comparison shows that low-frequency sine waves are gone and high-frequency saw waves remain — so it's a high-pass filter.

Research Paper Analysis

Leaky Integrate and Fire Neuron

Credit: M. A. Khanday et al.

The input current increases, and after a threshold is hit, the neurons fire — hence the paper title: Leaky Integrate and Fire Neuron by Charge-Discharge Dynamics in Floating-Body MOSFET.

Reference: https://www.nature.com/articles/s41598-017-07418-y

Visual Analogy

Video shown: YouTube Short
In the video, water collects (leaky integration) and, after a threshold, the bamboo tilts (fire), then resets — a nice analogy.

Speech Classification Implementation

Paper: IEEE Document
Related article: IEEE Spectrum Article

They implemented SNN on a chip for speech classification achieving an energy per spike of 8.2 fJ — extremely efficient.

Memory in Speech Recognition

To recognize words like "bat", previous letter information must persist. Liquid state machine analogy — a drop creates waves that persist briefly, like short-term memory. Similarly, SNNs retain information through recurrent connections.

Energy Efficiency in AI Computing

Traditional AI computing consumes much energy because data must be stored, chunked, and transferred repeatedly — heating wires and wasting energy.

Biomedical Applications

Another research showed an electronic backpack attached to cockroaches to control movement — a step toward prosthetic control, enabling people to regain bodily control.