Will Young Americans Want to Work in Semiconductor Manufacturing?
If you want to watch the video, it is below:
The feedback in the comments for this video has been mostly: “Yes, if only those greedy tech giants would pay us enough”
What is “enough”? $100K a year? $120K a year? 200K? What do you think?
Recently, an article about semiconductor degree programs in the US got posted on HackerNews. It quickly received a flood of comments about the career path. I was really struck by just how negative they were.
If you're a fab person, you'll work long hours, be on call often (and you will get woken up often), and eventually will own a tool that you'll be responsible for, even when not on call ... Lots of abusive and pathological behavior, as well.
Here's another one discussing the long term career prospects of a TSMC engineer. Spoiler alert, they say it sucks. A PhD is mandatory. Post-doctorate experience is also mandatory. And you also have to put 10+ years in an R&D sweatshop.
To sum it all up, another comment asserts that semiconductors are a "problematic career" with long hours, subpar labor practices, and low pay.
"People with other skills get out" - it makes it sound like a prison camp here.
Today, American politicians and social elites can't shut up about building fabs for semiconductor independency and finding industry "talents" to staff those new fabs.
But how can America “reclaim” its title as a world leader in semiconductor manufacturing if Americans actively discourage their peers from joining it? In this video, just a few thoughts that I have been thinking on with regards to semiconductor manufacturing careers.
Manufacturing Versus Design
Before we get started, I have to say. Not everyone at TSMC or Samsung works on the manufacturing line - and the workers for these jobs can't smoothly switch between them.
Broadly speaking, the semiconductor industry has manufacturing and design jobs. Design jobs help create the designs that the foundries fab out into real chips. Manufacturing jobs are the guys who might have to put on a clean suit and go into the fabs.
Interestingly enough, TSMC hires a lot of IC design engineers to work with customers on how to best prepare their chips for fabrication. They also hire a lot of software developers too. For instance, they hire AI engineers to develop algorithms that help make their machines perform better.
I would say that these non-manufacturing jobs are a lot like software engineering. You sit in a comfortable air-conditioned office or work from home and earn lots of money. These jobs are fine. Let's talk about the other kind.
On the Line
The goal of semiconductor manufacturing is to deliver extremely complicated products at staggering scale. This requires a lot of operations discipline. Companies like Intel, Samsung and TSMC succeed on the back of military-style execution.
So what are these manufacturing jobs like? I can only describe what I can see here in Taiwan.
On the manufacturing lines, there are two types of workers. First, you have line technicians and operators, led by a foreman. Their responsibilities are to debug manufacturing issues, operate the line, and inspect products.
In Taiwan, these guys work two days on, two days off - 12 hour shifts from 7 AM to 7 PM and vice versa. This is legal in Taiwan as it is also in the United States - though some state laws will require overtime pay.
You only need a high school or vocational school degree for these jobs. However, the internal job progression for this track is somewhat limited. People can ascend to foreman status - but that’s about it. Some TSMC people here spend years as a foreman, living life.
The pay is significantly better than what someone can get working at 7-11. Though I wouldn’t say a 7-11 job here in Taiwan would have anywhere near the anxiety and high pressure of being a foreman of a TSMC manufacturing line. Except maybe at rush hour when everyone needs their coffee.
And then there are the engineers. They co-work with the technicians to guide the manufacturing process flow and fix problems. TSMC employs several major categories of engineers.
Process engineers inspect and analyze the various processes in semiconductor manufacturing. They need to optimize the processes' variables and test for optimizations. When yield rates are low, this guy's life is hard.
Equipment engineers monitor and inspect the equipment and are on the hook when it breaks. They do require a technical degree, but it does not always have to be electrical engineering.
In Taiwan, equipment engineers will go to the temple - Taiwan is full of them - and buy charms in hopes of making the device act more obedient.
R&D engineers work in new product development. They are in a race against time and the competition to create new processes and products.
These jobs face particularly high pressures. They regularly require PhDs and work 10+ hours a day on experiments. However, those who succeed win plaudits and street cred from their peers - it is a genuine way to help make a big impact on the whole company.
Production management engineers work on the production forecasting and logistics side. They make sure that products are being properly dispatched, there are enough resources to keep production smooth, and that the customers are aware of progress.
Because they have to interact with Western customers, their English skills have to be better than that of the average Taiwanese.
And finally there are the smart manufacturing engineers and AI engineers. I mentioned the latter earlier. They use big data and AI techniques to help improve production.
Semiconductor foundries in general have had a hard time filling these positions. But the situation is especially tricky in the United States because of the adjacent riches offered by the software development industry.
Few jobs - engineering, non-engineering or otherwise - can compete with American software development. I would argue that it’s the single best job category in the world.
It is relatively easy to break into. There are a lot of companies in the US needing software developers. The perks and benefits are legendary. The salaries can get obscenely, ridiculously high - especially when you add stock compensation. And so on.
There is a reason why 9 out of 10 Berkeley EECS majors graduating in 2018 ended up going into software engineering.
EECS literally has Electrical Engineering in it, yet for them EE is just an intriguing extra thing they have to do before joining Google or Meta.
Or if they do stay in the semiconductor field, many of them end up going into chip design. Which has a lot in common with software development.
It makes sense. Working on the semiconductor manufacturing line may require them to put on a clean suit and head into the factory. Most very smart, highly educated human beings - social elites - would rather work remote from their bedrooms writing code.
This problem is intercontinental. A few months ago, I read a fascinating article about China's own struggles to hire and keep experienced line technicians for their growing microelectronics industry.
Their youngest and brightest workers in Shanghai tried out the industry, and decided that they rather go be software developers for some of China’s massive tech giants. It beats having to wear a clean suit every time.
It is a problem even in Taiwan, where people are obsessed with TSMC. What track and field is to Jamaica is what electrical engineering is to Taiwan.
The entire force of Taiwanese society pushes their young people into it. Every Taiwanese parent thinks of TSMC as an American parent might think of Google or Apple.
And yet despite this, the Taiwanese company has to deal with substantial turnover in its technician and engineer ranks. It is a very demanding work culture.
Americans have been experiencing this as TSMC expands into Arizona. A few months ago, a Glassdoor review left by an American training at TSMC went viral. I read it, and it feels to be a pretty fair and even-handed review of the culture clashes. Even here, people brought up in this culture leave all the time.
The growing presence of various chip design companies in Taiwan are making it easier for those talents to jump ship. MediaTek and power electronics-maker Delta Electronics offer salaries almost as high as TSMC.
Not to mention non-Taiwanese companies like those in Mainland China and the US. Google's main hardware R&D hub outside of the US is in Taiwan, and their presence in Banqiao has been growing quite rapidly. I heard Google has basically poached half of MediaTek's people.
The importance of the issue has been reflected in TSMC's unusually high 20% employee wage raise announced in 2020, taking effect in 2022. The company has also been working towards making it easier for employees to purchase TSMC stock. More pay doesn't solve the work environment issue, but it certainly helps.
Back in the United States, I think Intel has found success in building their own pipelines of semiconductor specialist technicians.
For instance, Intel has set up a long standing partnership with various Arizona community colleges to enroll prospective technicians in 2-year technical education degree programs.
In Malaysia, Intel offered research grants for students to learn semiconductor engineering - particularly in packaging, where the company has a facility. The company also contributed the time of their own senior engineers to help train these people in industry practices.
The benefit of starting on ground zero is a trained, affordable workforce that meets the company's specific needs. But I also see the flip side of this notion that these HackerNews commenters so artfully point out:
Which is that these workers end up having such specialized skills that they cannot leave the field even if they tried. And as a result, they can get underpaid for their specialty and even sometimes poorly treated.
It is a reflection of the fact that there are really only a few companies who do advanced semiconductor manufacturing. Software engineers and even IC chip designers on the other hand have a far greater employer pool competing for their efforts.
Now, there are a lot of ultra-impressive skills that the market similarly undervalues. Like the trombone. But what can we do about the strategically important but economically undervalued situation of semiconductor skills? Individual income credits for hardware technicians? Extra taxes for software engineers? What do you think?
Now that leads into another point: Immigration. Let’s get those foreign semiconductor talents to come here to the United States, right?
For that one, I am not so sure if it is politically feasible. The big question for me is whether or not the problem is perceived to be serious enough to elicit such a radical move in policy-making.
A policy brief from Georgetown says that the now-passed CHIPS Act will create about eight new fabs translating to about 27,000 new jobs.
Of those, the report mentions that some of these jobs will need to be filled by foreigners:
> The industry will likely need at least 3,500 highly skilled foreign workers with prior expertise. Ideally, these workers would have experience working at firms like TSMC or Samsung, bringing tacit engineering knowledge to ... American workers.
What kind of jobs are these going to be, I cannot really identify. Are these TSMC technicians, engineers, R&D people, or high-level executives?
The high-level executives and R&D people can already come to the United States if they want to. Many were educated in the US and many hold dual citizenship. I know one of either TSMC’s chairman or CEO has American citizenship - as does Morris Chang.
Okay, then how about the semiconductor engineers and technicians? For that I am not sure if 3,500 jobs is a serious enough shortage to make a special program just for them.
I look back at the H-1A Visa created in the 1980s to address the nursing shortage back then.
For the six years it was in effect, they issued nearly 40,000 visas - mostly to Filipinos. The program was then sunset, totally ending by the 1990s.
And in this rather divided political environment, I doubt it can happen again. 90% of TSMC's hires are Taiwanese. People are going to point to that. And immigration doesn't solve the fundamental issues of Americans choosing to work for software over hardware.
To conclude, I want to go back to something about young Americans and semiconductors. In his newsletter, SemiAnalysis hit upon something that has been echoed by a lot of other people that I have talked about the issue. Young Americans don't have easy access to hardware anymore.
Young Americans have easy access to software. It is super easy to go to Github and download some open source software like Rails to set up a website. And with that, young people can spark an interest that will eventually lead them to their future career.
That's not so easy with today's hardware. SemiAnalysis points at right-to-repair laws for the reason why this is the case, and it certainly doesn't help. But there should also be something said about complexity.
In the 1960s, your average person can break open a transistor radio and understand it enough to fix it with their own hands. Personal computers were more complex, but most people can go inside and interact directly with the hardware to make it do their bidding.
But your average cellphone is today just too complex - built with stacks of technologies that a single person can spend a whole lifetime studying. The most a lot of people can ever do is change the battery.
So today, users don't interact with the actual hardware everyday. They interact with the GUI software and slick UX. And as a result, that's what they eventually get interested in and go do as a career.
It also means that the average person on the street sees their iPhone or Samsung or Huawei more as a mystical remote talisman than a tool with levers and pulleys that you can manipulate to make do things.
The integrated chip inside the phone is a beautiful thing. But young folks never get to see it. Never get to tinker with it. And as a result, they grow up never thinking about it - leaving them to do something else.