The Siliconomy

As the author of this article, I am deeply committed to maintaining the highest standards of integrity and transparency in my work. For that reason, I want to clarify that no part of the below article was generated or written by ChatGPT or any other AI tool.

I believe in the power of human creativity and intellect, and this article is a testament to that belief. Thank you for taking the time to read, and I hope you find the insights provided both informative and thought-provoking.

Back in the Spring of 2023, my buddy Drew McIllwain, EIT and I took a little weekend trip to Nashville, TN as we went to check out the Country Music Hall of Fame. On the way, instead of blasting BigXthaPlug, we talked about what all normal people talk about on a Saturday afternoon: the global semiconductor industry. Honestly, even if we wanted to blast BigXthaPlug, we wouldn’t have been able to. His truck had some of the worst sounding speakers of any vehicle I’d ever ridden in. On the same token, I can’t really talk about vehicle quality as I was personally driving a multi-colored Volkswagen Jetta held up by zip-ties and, at the time, was eligible for a “Classic Vehicle” license plate in about 2 years. Regardless, there we were on a Saturday afternoon discussing semiconductors. On the drive up, Drew did most of the talking, but I contributed, per usual, with my ears and my questions. Our discussion was ignited by an interest in the stock market, which is an admittedly poor reason for a discussion on the semiconductor industry. The beautiful technology behind the scenes and the accomplishments of the leaders within the industry should be enough to prompt a discussion. Regardless, as we rode and as Drew spoke, he discussed the engineering marvels associated with the machines needed for the manufacturing of today’s most advanced semiconductors. Today, the manufacturing of these machines is fueled by one company: ASML (Advanced Semiconductor Materials Lithography). In addition, he talked about the methods the machines use to achieve the manufacturing, explaining that the technology and innovation going into today’s semiconductor manufacturing is likely the greatest engineering feat in the history of mankind. Paired with the engineering feats, he also told stories about the history of today’s most influential semiconductor companies. One of my favorite stories was the one about Morris Chang who, in 1983, was passed up for CEO at Texas Instruments. I felt like a kid again, reminiscing on the nights where my mother would read “The Very Hungry Caterpillar” or “The Giving Tree” to me. In 1987, four years after leaving Texas Instruments, Chang found himself spearheading a new venture. Backed by the full force of the Taiwanese government, Taiwan Semiconductor was established. Today, TSMC is arguably the most important company in the global semiconductor industry as they provide the chip-fabrication needs for hundreds of global tech companies. Marvell, Sony, AMD, Broadcom, Qualcomm, Nvidia, and Apple are just a few. While TSMC is one of today’s unquestioned leaders, there was a time when Morris Chang’s vision for TSMC was severely doubted. Ironically enough, one instance of this doubt came thundering down from Gordon Moore, co-founder of Intel. After Morris Chang shared his vision for TSMC with Moore, Moore responded by saying: “Morris, you’ve been at the top of the industry and have been a part of many great ideas. "This," referring to Chang’s vision for TSMC, “isn’t one of them.” Today, in terms of overall influence, you can’t get any bigger than TSMC. On the same token, Moore’s Law, for those who are familiar with the concept, has been all but officially pronounced dead.

?

Jensen Huang

You might not know who Jensen Huang is, but you’ve probably heard of Nvidia. As a child, Jensen Huang, Founder and CEO of Nvidia, moved from Taiwan to a boarding school in Kentucky, where he was assigned to clean the boy’s bathroom every day after class. Later, he moved to Tacoma, Washington, and then eventually moved to San Jose, California where he found work at a local Denny’s. Instead of a lavish coffeehouse in Silicon Valley, the origination of Nvidia began in that same Denny’s. This is interesting because not only does it reflect the concept that a good idea can originate from anywhere, but it also magnifies the importance of a good breakfast before starting each day. Today, Nvidia is the unquestioned chip design leader within the semiconductor industry. For clarity, fabrication and design are not the same. Companies such as Nvidia design the chips while a limited few such as TSMC, Samsung, and Intel specialize in chip fabrication. This split in the industry occurs because of the unreal amount of capital and resources required to build a “fab” (this is what they are called in the industry). For perspective, your typical “fab” would take about three to four years, over $10 billion and 7,000 construction workers to complete. In addition, a facility to fabricate the most advanced logic chips can cost twice as much as an aircraft carrier but will, based on the pace of the industry, only be cutting edge for a couple of years. For chip-design companies like Nvidia, this undertaking isn’t even considered. This way, Nvidia can spend less time on fabrication and more time on designing their AI-tailored chips. Nvidia’s early GPUs, including the release of their very first GPU “GeForce 256”, have played an increasingly important role in the development of AI. In the beginning, Nvidia’s GPU’s (graphical processing units) were targeted for gaming and general graphical processing tasks such as video streaming. When Nvidia was still a nascent company, Jensen Huang, Nvidia’s CEO, saw that university students in California were using his company’s chips in a very clever way. In fact, the students weren’t using the GPUs for graphical processing at all. Instead, the students were building artificial intelligence models which, at that point in history, required the computing power supplied by Nvidia’s GPUs. Jensen Huang, catching wind of this, quickly turned the direction of Nvidia’s ship. From that point on, the design of Nvidia’s chips were 100% focused on the alignment with the quickly, although still relatively unpopular, field of AI.

?

Silicon Valley’s Trailblazers

Today, we see the monumental headquarters of Apple at Apple Park. We see Bill Gates’ energy-efficient mansion that overlooks Lake Washington. We see the internet blow up over a photo of Jeff Bezos because “he looks jacked.” Today, the tycoons of Silicon Valley have surpassed those of Hollywood. California’s biggest stars aren’t in Hollywood, they’re in Silicon Valley. While a spot atop the highest ranks of Silicon Valley is akin to a top spot in Hollywood, this wasn’t always the case. We recognize the power and influence that comes with a seat at the top of Silicon Valley, but none of that would be possible if it weren’t for the true trailblazers that made Silicon Valley. The people that had no map - the Christopher Columbuses, the Lewis and Clark’s. The men and women that had to create their own map. The ones that saw what nobody else could – who continued pushing the needle amidst uncertainty at its highest level. That’s what a trailblazer does. In the moment, it’s not popular and it’s not glamorous, but that’s the sacrifice a trailblazer makes. They’re not fueled by money or fame but, instead, their journeys are ignited by the greatest fuel there is: purpose. The men and women that pushed the semiconductor industry forward when pushing forward wasn’t popular are the reason for the successes of Bill Gates, Jeff Bezos, and the late Steve Jobs. We, as a society, owe a great deal of appreciation to their contributions.

?

In 1971, Integrated Electronics, better known as Intel, created the world’s first microprocessor (the Intel 4004). In turn, this invention laid the foundation for the rapid development of computers and digital electronics, which wouldn’t have been possible without Gordon Moore, co-founder of Intel and the man who Moore’s Law is named after. In fact, Moore’s Law may have never gained traction if it wasn’t for Carver Mead, a consultant and close friend to Gordon Moore who gave the trend of doubling-computational-power-every-two-years (Moore’s Law) a name. In addition to Gordon Moore, Intel might not be here today if it wasn’t for Robert Noyce, co-founder and long-time CEO of Intel. The collaboration between Gordon Moore and Robert Noyce at Intel would not have been possible if Robert Noyce hadn’t invented the “planar process,” a technique for manufacturing integrated circuits using silicon. With respect to time periods, the effect that the “planar process” had on the world was akin to that of Johannes Gutenberg’s printing press in 1440. Furthermore, Robert Noyce’s invention wouldn’t have been possible without Nobel Prize winner Jack Kilby. One year before Robert Noyce’s creation of the “planar process,” Jack Kilby was credited with the creation of world’s first integrated circuit while at Texas Instruments in 1957. As it goes, Nobel Prize winner Jack Kilby’s integrated circuit would have never come to be if it wasn’t for John Bardeen, Walter Britain, and William Shockley, who are all credited with the invention of the point-contact transistor during their time at Bell Labs. Today, we know Bell Labs as AT&T. ?This, in turn, would not have happened if it wasn’t for the German engineer Karl Ferdinand Braun. In 1874, Braun sparked the semiconductor industry through his observations of conduction and?rectification?in metal sulfides. In English, this was the beginning of the AC-DC convertor. Finally, and while there has been extensive research on Silicon replacements within the semiconductor industry, we must recognize one man in particular: J?ns Jacob Berzelius. In 1824, the Swedish chemist discovered the element Silicon and, in doing so, unknowingly lit the fuse to an industry that wouldn’t take off until at least another century.

?

The World’s First Chipmaker

The first semiconductor company, Fairchild Semiconductor, was founded in 1957 in Santa Clara, California, which was only 13 miles from the “heart” of Silicon Valley as we know it today. Fairchild was founded by eight men, two of which included Gordon Moore and Robert Noyce, the eventual co-founders of Intel. Always looking for an edge, Fairchild Semiconductor was also the first company to offshore semiconductor production to Asia. In 1963, they paid South Korean workers 10 cents an hour to manufacture their chips. Shortly after, others began jumping onto the semiconductor scene, both in and outside of the United States. In the early days, semiconductor companies received much of their revenue from the United States government. This took place through military contracts and, while a contract with the U.S. government is hardly ever a bad thing, industry pioneers such as Intel’s Robert Noyce realized that the consumer market held the real potential for earnings. With this in mind, the industry turned their attention to consumer markets. Years later, as foreign semiconductor companies began producing higher quality chips at a lower price, the companies that once turned away from the U.S. government came crawling back, seeking shelter from the tumultuous storm that was beginning to form within the global semiconductor industry.

?

Carver Mead’s Bold Prediction

As the global industry became increasingly competitive, the need to pack more and more transistors into a single chip increased. By squeezing more transistors into a smaller semiconductor, more work can be done in less time. More transistors equal more power and efficiency. As the number of transistors increases, the chip can perform more calculations and store more data. During the earliest hours of the chip industry, the first generation of semiconductors were hand-drawn, even at soon-to-be behemoths such as Intel. This was back when semiconductors had thousands of transistors. The Intel 4004 microprocessor, for example, had 2,300 transistors, which was world-class at the time. In the 1970s, the transistor patterns on a semiconductor were sketched by hand. This was done on a reddish material called Rubylith. Carver Mead, Intel’s 5th employee, predicted that one day semiconductors would have millions of transistors. Well versed in the area, Mead knew that the early method of hand-sketching semiconductors would quickly become impossible, so he set out to find a better way. In his search, he was introduced to a lady named Lynn Conway. Shortly after meeting, the two collaborated on a textbook called Introduction to VLSI Systems. The book, and the principles within, set the chip industry on fire. Computer science and electrical engineering professors, Carver Mead included, began teaching VLSI (very-large-scale integration) system design. Upon the textbook’s publishing in 1978, the maximum number of transistors that had been placed on a semi-conductor was 20,000. With the help of Lynn Conway, Carver Mead, and the new software-based approach used to construct semiconductors, this number was set to double every two years for the next 40 years. Today, Apple, who began designing their own chips in 2010, is packing roughly 114 billion transistors into their M1 Ultra microprocessor. While Mead may have misjudged the potential for semiconductor growth, he was right about one thing. In 1970, back when lava lamps were high tech, Mead stated that “Influence in this new world would accrue to people who could produce computing power and manipulate it with software.”

?

Chip Industry Today

As of April 11, 2024,?Nvidia?was the largest semiconductor company in United States in terms of market capitalization, amounting to 2.27 trillion U.S. dollars. Broadcom ranked second with a market cap of 640.6 billion U.S. dollars, followed by the likes of AMD, Intel, and Qualcomm . The semiconductor industry (at least by the stock market’s standards) has had a meteoric rise this year. If you follow the stock market, you’ll know that companies such as Nvidia, TSMC, and Broadcom have had unprecedented years. As stated above, Nvidia is the most valuable semiconductor company in the world, but this value correlates directly with their market capitalization which, in turn, correlates directly to their stock price. While AI is undoubtedly the future and Nvidia’s chips will undoubtedly play a crucial role, the company’s stock price has raised some eyebrows. For this reason, judging a semiconductor company purely from their market capitalization may not be the best reflection of their dominance within the entire globalized industry. While true, a company’s market capitalization is probably the most efficient way to determine the future role that a company could eventually play. As of April 7, 2024, here are the top 10 global chip companies by market cap:

1.??? Nvidia (United States)

2.??? TSMC (Taiwan)

3.??? Broadcom (United States)

4.??? Samsung (South Korea)

5.??? ASML (Netherlands)

6.??? AMD (United States)

7.??? Qualcomm (United States)

8.??? Applied Materials (United States)

9.??? Intel (United States)

10. Texas Instruments (United States)

?

In addition to the above list, there are many other important companies in many different countries that play a crucial role. The absence of even the smallest companies within the ecosystem could put the global industry in jeopardy. Today, most of the companies leading the charge are based out of Japan, South Korea, Taiwan, Netherlands, and the United States, with some other prominent players scattered throughout Europe. As stated, there are companies that fabricate, or manufacture chips, there are those that design chips, there are those that create software for the machines that design the chips, and even companies that create the world’s purest diamonds which eventually make their way into ASML’s machines.

?

To make the world’s most advanced chips, you need the world’s most advanced machines. Today, ASML is the only company in the entire world that is capable of making these machines, which comprise of around 457,329 parts and a price tag of $380 million. This makes ASML’s most advanced machines the most expensive mass-produced machine in the history of mankind. As the semiconductor industry has developed, there has been a constant need for machines that have the capacity to etch (draw) more and more transistors on a semiconductor. The larger the “node” used to cut into the silicon, the less space to put transistors. The smaller the “node,” the more transistors. In 1997, most of the cutting-edge semiconductors were made with a 250 nm (nanometer) node. Today, the standard is 3nm. When speaking of “the greatest engineering feat in the history of mankind,” the ability for an ASML machine to produce chips with a 3nm node is what we’re referring to. The precision and stability needed to achieve 3nm is mind-boggling. Interestingly, this precision and stability provided by ASML’s machines require the world’s purest diamonds. Speaking broadly, these ultra-pure diamonds are used to reflect the 3nm nodes at certain angles where they will eventually make their way onto the Silicon. Using ASML’s 3nm machines, the letter “X” drawn on the silicon would not be “drawn” in the way we would typically think of an “X” to be drawn. On the Silicon, we would see the “X.” Behind the scenes, the 3nm node, based on a good dose of algorithms, is reflecting off different parts of the diamonds which produce the desired result of an “X.” The reason for this, among other things, is to ensure precision and stability within the system. From a mathematical perspective, the precision happening within the machine would be the equivalent of a laser coming from the moon and hitting a golf ball on Earth. This gives us an idea of the engineering challenges involved. If the laser on the moon were to move a micro degree (one-millionth of a degree), the target golf ball would be missed by 10 meters. Not only do ASML's machines hit this golf ball, but they do it repeatedly, 24/7, millions of times a day. All of this is happening while there are people walking around, trucks driving by, and internal temperatures changing. At that scale and level of accuracy, everything influences the machine.

?

While ASML builds the machines, companies such as TSMC, Samsung, and recently Intel, use the machines to fabricate semiconductors. In the fabrication space, TSMC is the world leader, producing 90% of the world’s most advanced chips and 50% of all chips combined.?The former chair and CEO of the Taiwan?Semiconductor, Morris Chang, estimates that it would cost up to 50% more to make leading-edge chips in the U.S. than in Taiwan due to labor costs and different worker norms. Until recently, only Samsung and TSMC could produce the world most advanced chips. Now, Intel is revamping its foundry business. While Intel just reported an operating loss of over $7 billion on their new foundry business (Yes, $7 billion), their prospects look solid. In addition, a foundry that can produce the world’s most advanced chips within the United States is a huge win for the national security of the United States.

?

In addition to the big-name-players such as Nvidia, AMD, Qualcomm, and Broadcom, more companies have started designing their own chips. These include Apple, Google, Meta, Tesla, Amazon, and several others. Apple, for example, has been making chips for their iPhones since 2010 when their first in-house chip was put inside the iPhone 4. In addition, cloud companies such as Microsoft (Azure), Google (Google Cloud Platform), and Amazon (AWS), have all began designing in-house chips, primarily for their data centers. Throughout the 2010s, almost all data centers were contractually bind to Intel’s x86 chip, which is a big reason Intel stayed afloat. Google’s first in-house chip Tensor, or TPU (Tensor Processing Unit), first debuted in 2021 within the Google Pixel. Since then, Tensor G2 and G3 have succeeded the original Tensor chip. This chip, which is also designed to have great machine learning and AI capabilities, is perfect for their data centers. For those with a machine learning background, you may be familiar with Python and it’s TensorFlow machine learning library, which interacts directly with Google’s Tensor chips. In addition, Tesla also develops their own chips in-house. This reason, in addition to their phenomenal ERP (Enterprise Resource Planning) system, which is discussed in my last article, is a big reason that they were able to weather the impact that COVID-19 had on the automobile industry. Tesla, born in Silicon Valley, has always written their own code. They never outsource. During the pandemic, Tesla rewrote software so they could replace chips in short supply with chips not in short supply, which is incredible.

?

Chip Industry Future

In 1965, Gordon Moore discovered a trend that would come to be known as “Moore’s Law.” Computing power was doubling every two years, meaning that the number of transistors on integrated circuits were also doubling every two years. Unfortunately, transistors can only get so small, which is why many say that the permanent laws of physics will eventually get in the way of Moore’s Law. For example, TSMC uses ASML’s machine to produce 3nm nodes, which is barely wider than a 2.5nm strand of human DNA. There will always be room to squeeze more transistors on a single chip (in 2021, IBM announced the successful creation of 2-nanometer chips), but this progression has become prohibitively expensive and slow. ?

?

In her article “The Death of Moore’s Law,” Audrey Woods talked about her discussion with MIT Professor Charles Leiserson, who claims that Moore’s law has been over since 2016. He goes onto say that, without the advancement of chips, the only way for more computing power is through larger and larger data centers, which could be devastating for climate. However, there's a silver lining. Moving forward and assuming Moore’s Law doesn’t auto-resuscitate, computational power will need to be paired up with efficient software engineering practices. As software engineers, we’ve never really had to worry about how “efficient” our code is. In the past, Moore’s Law took care of our efficiency problems as we never had to concern ourselves with code efficiency. In the past, software engineers could brute force performance increases because they knew computing power was bound to double every two years. Now, we must turn toward efficient software practices and “work” for our gains.

?

Alan Kay, one of the world’s greatest computer scientists, said “people who are really serious about software should make their own hardware.” Today, the slowing of Moore’s Law has cemented this idea. While chips will always continue advancing, the software and algorithms which interact with the chips of the future is where the real impact will be made. As software engineers, it’s important to have a good understanding of how the software we write operates under the hood. As to Alan Kay’s quote, until an engineer has developed and understands the interworking of the CPUs and GPUs, the lowest levels of software shouldn't be treated as anything other than magic. Without at least some understanding of the underlying hardware, a software engineer can't really understand their software. Most popular high-level languages such as Python and C# don’t require a knowledge of the inter-workings of a computer's hardware. While true, that knowledge will become increasingly beneficial as we look for further increases in computational power.

?

As for the future of fabrication within the semi-conductor industry, Samsung and TSMC still hold the reigns. Intel however, is coming, and they’re backed by the full force of the United States government. Seven years ago, Intel cut a deal with ASML to acquire their very first next generation EUV machine, which was expected to be ready by 2025. In January 2024, Intel received this machine. Ultimately, Intel CEO Patrick Gelsinger foresees Intel’s foundry business as a direct competitor of Samsung and TSMC, which ultimately brings advanced chip manufacturing back to the United States. While Intel did recently report a staggering operating loss of $7 billion on their foundry business, they have contracts and funds coming in from big-time players including Microsoft, Arm, and the U.S. government. From a recent Bloomberg report, the U.S. government is set to invest $3.5 billion in Intel, which will further facilitates the enhancement of the chip-making process within the United States. So, In addition to improving the efficiency of my code, I’ll probably be buying some Intel stock along the way, as well.

?

I hope you enjoyed the read and were able to get something from it. As always, thanks for reading, and good luck.