Highly advanced semiconductors are indistinguishable from cakes

Hello, this is Kakao Ventures.

Our investment team always works closely with early-stage startups while keeping an eye on market trends. In this process, we often have questions and engage in many discussions. We believe that the more diverse the thoughts and the deeper the conversations, the better. That's why we would like to share some of our thoughts with you one by one in the future. We hope this will be of some help to entrepreneurs, investors, or anyone interested in the market.

Highly advanced semiconductors are indistinguishable from cakes.

[Deep Tech Stories by Zero] 1. Overview of Semiconductors

Written by Youngmoo Kim

With the launch of ChatGPT, an intense wave of large language models (LLMs) has emerged. Alongside the rise of LLMs, semiconductors have become a hot topic.

As the industry has discovered emergent abilities in LLMs and accepted scaling laws, these models demand immense memory bandwidth. Consequently, existing semiconductors struggle to keep up with the computing power and memory throughput required by the rapidly growing LLM parameters. This has brought previously obscure components, such as OSAT's interposer, into the spotlight.

But wait, you say you don’t understand any of this at all?

You're not alone. In 2020, OpenAI's declaration that "AI models get better the larger they get" spurred rapid advancements in both software and hardware. This swift evolution has introduced numerous challenges, shaking up technology and the market.

The rapid transformation over such a short period has changed the flow of technology and market dynamics, inundating us with unfamiliar concepts. It's no wonder the brief introduction earlier was a bit difficult to understand.

Today, rather than diving into overly complex topics, we want to start by discussing what semiconductors are in general, get familiar with difficult terminology, and then begin a series.

There are countless semiconductor-related terms like MOSFET, 3D stacking, EUV, 2nm, etc., that can get confusing in terms of what exactly they mean when you hear them. This confusion often arises because people are only exposed to surface-level information without understanding the fundamental composition, structure, and manufacturing process of semiconductors. However, if we look at the basics of semiconductors, it's not that difficult—semiconductors are actually quite similar to cakes.

Much like cakes, which are made up of various layers and ingredients, semiconductors are composed of different materials and layers that work together to form a functional device. Understanding these basics can demystify the terminology and make the complex world of semiconductors more accessible.

Semiconductors = Cakes?

When you hear "semiconductor," you might think of silicon wafers often mentioned in the news, chips with multiple pins, or something like the graphics card shown below.

However, the semiconductors that take those common forms are actually chip modules that have been well-packaged for commercial release. Inside those modules are countless electronic components, and each of those components is itself a semiconductor. The term "semiconductor" encompasses a lot of hidden content.

To better understand semiconductors, we need to focus on their most basic unit. The basic unit of a semiconductor is much smaller and has a different shape than what we typically imagine.

Is this the basic unit of a semiconductor you're referring to?

That's right! The electronic device in that photo is called a transistor, and it's the basic unit of a semiconductor.

However, the transistor in the photo is huge—far too big.

It may seem odd to call something the size of a fingernail "huge," but the transistors used in current chips are much, much smaller.

The photo above shows semiconductors from Intel and AMD.

It might be confusing because it's in black and white, but if you recall the fingernail-sized transistor we saw earlier, it had three pins. In the Intel and AMD semiconductors, you can see similar three-pinned structures repeated over and over. Each of those nanometer-scale structures performs the same role as the fingernail-sized transistor.

Can you get a sense of how incredibly small they are?

To provide a bit more context, the disc shown is likely something you've seen in the news before—it's called a silicon wafer, which is used to manufacture semiconductors.

If you look closely at the wafer, you'll see countless small rectangular patterns etched onto it; each of those rectangles is called a "Die."

We cut out an individual die, package it, and that's how we get the modular chip form that typically comes to mind. With today's technology, we can cram anywhere from tens of billions to hundreds of billions of those tiny transistors into a single small die.

The photo above shows the very first transistor created at Bell Labs. This transistor had a bipolar junction transistor (BJT) structure.

Compared to the size of transistors back then, you can get a sense of just how incredibly small transistors have become. If they hadn't shrunk this much, we would have had difficulty even using calculators, let alone laptops and smartphones.

Unlike those early, bulky transistors, today's transistors look like this diagram when viewed up close: Deviating from the BJT structure, this form is called a MOSFET, with three pins coming from the source, gate, and drain. So, fundamentally, transistors have this same three-pinned structure.\

Unlike BJTs, MOSFETs are designed to be printed directly onto silicon wafers, enabling incredible miniaturization. This is because, instead of having to solder each component individually, you can simply load wafers into dedicated manufacturing equipment and have the printing done automatically. It's like how we went from pottery makers individually hand-shaping each piece on a wheel to being able to mold and mass-produce ceramics using forms. Using forms allowed for more precise, mass manufacturing of semiconductors.

The old soldering method also meant that if even one component was poorly soldered or failed, it could bring down the whole system—a problem that became more serious as systems grew more complex (the tyranny of numbers). But by implementing entire systems on MOSFETs printed on wafers, we eliminated that risk, much like how using molds prevents human error from ruining ceramic pieces.

By utilizing MOSFETs and wafers, we finally ushered in the era of integrated circuits (ICs) and integrated semiconductors. This rapid advancement in semiconductor performance also gave rise to Moore's Law.

So, you've been following along up to this point. But how does this relate to cakes, you ask?

The image above is a side view cross-section of a completed semiconductor. At the bottom, you can see the repeated MOSFET structures we looked at earlier, with many layers stacked above.

The reason these layers build up is directly related to the 8-step manufacturing process for making semiconductors.

In simple terms, semiconductors are essentially 3D printed onto silicon wafers. A layer is created on the wafer, then circuits are etched onto that layer with lasers. Another layer is then deposited over those etched circuits, unnecessary portions are removed, and the process repeats by stacking yet another layer.

Thinking about this layer-by-layer stacking process, it's quite similar to how cakes are made.

You lay down a base, add fruit or snacks, spread frosting, then lay another base on top and repeat.

Explaining it this way, you can easily visualize how semiconductors are structured. Are you starting to feel a bit more familiar with what semiconductors are?

But the analogy doesn't end there.

When you think about it, this method of stacking layers is similar to how a cake is made. You lay down a layer of cake, add fruit or cookies, spread whipped cream, place another layer of cake on top, and repeat the process.

We've looked at semiconductors at the tiniest component level.

But just like with cakes, is simply acquiring ingredients like cake bases and frosting and stacking them up enough to sell cakes?

Of course not.

If we're a bakery making and selling cakes, we wouldn't just make round strawberry cakes. Especially if we have a multi-trillion won industrial oven for baking cakes. Different customers would want different types of cakes - some might want wedding cakes, others basic birthday cakes, and others Christmas cakes. So the bakery would have to design completely different cakes to meet those varied demands.

For the same reason, semiconductor companies also design different chips tailored to what their customers want.

However, designing semiconductors and manufacturing them require different capabilities, so multiple companies often collaborate on production. It's like Rebellions designing an NPU for AI, which is then manufactured by Samsung Electronics.

In fact, the semiconductor industry forms a far more complex and massive market ecosystem compared to cakes.

Companies without their own fabs like Rebellions are called fabless, while companies that accept orders for manufacturing are called foundries. There are also IDM (Integrated Device Manufacturer) models like Samsung that do everything from design to sales and distribution.

And there are even more diverse business models in the semiconductor industry.

Some companies act as design houses that help design semiconductors and prepare for outsourced manufacturing. Others provide EDA (Electronic Design Automation) software for simulating whether designed semiconductors will function properly.

We've even invested in a Korean EDA startup called Baum Design Systems before, which you can read more about in the article below.

?? Baum Showcases Innovative EDA Solutions to Shift-Left and Accelerate Sustainable Silicon

We're also tackling technical challenges around how to build more power-efficient semiconductors, stack them higher, design more optimal chip architectures, and create smarter packaging solutions.

One example is the case of SK Hynix and Apple.

Previously, SK Hynix mass-produced standardized memory for general sale. But by shifting to a foundry model taking custom orders, they were able to design more tailored semiconductors for clients like Apple's new R1 chip in the Vision Pro headset.

Up to this point, we've taken a broad look at what semiconductors fundamentally are, how these tiny components are manufactured, and made sense of various vaguely used terminology by comparing them to cakes. I hope this has helped you understand these concepts more easily.

In the upcoming series, we will utilize the basic terms mentioned today to delve into more in-depth topics. Starting from semiconductors, we plan to cover diverse fields such as AI, Aerospace, Cybersecurity, Quantum Computing, and more.

The next topic I would like to share is <The Future of Semiconductors? - New Devices and New Materials>.

Thank you.