Manufacturing you'd need billions and the support of your government. But a design business? There's quite a lot of those around.
Bare minimum estimates, working backwards:
- $? for marketing and distribution, physical inventory costs
- $100k/y for one field engineer (FAE)
- $250k for first successful manufacturing run
- $250k for first full mask run with bugs
- $100k bringup boards, test equipment, engineering time fixing it
- $25k shuttle run for initial testing. This will find at least one bug.
- $100-200k outsourced layout: this is boring, specialised, and low value-add, so get someone else to do it
- $500k/y misc software and testing staff or consultants
- $100-500k/y each: 3-5 senior design engineers. For best results, these are people you already know and are spinning out of their job at Big Boring Semi Co
Excellent description, I wished I could've pushed your comment higher.
People are always shocked when I tell how tiny is the semi industry, but it really is.
Besides the super-concentration of semi manufacturing which starts to get more coverage, designs needs some exposure too.
When Apple bought PA Semi, it went rather unnoticed, but people didn't know that the amount of logic designers of a such calibre who can design cores like ZEN, or Firestorm is probably less than 100 in the whole of North America.
Sounds very dramatic, but America is less than 100 senior logic designers away from getting out of design business too.
Do you know how it is to work on software in this industry ? Be it for the tools or embedded software programming or in VHDL / verilog.
I'm still in university and I've always liked hardware and low level related stuff but I've heard bad things about software in hardware companies like Qualcomm.
The wage disparity in between semi industry at large, and almost everything else is a matter of legends...
Almost every part of semi industry is very bad on effort to salary ratio, and players like Intel, or Qualcomm are far from the worst, they actually do very well on salary front, and attracting talent. It's Asian companies who score the worst on that.
TSMC process development job is 100:1 lottery win + 20 years of your life for $50k a year salary. Know people with first hand experience of that.
The professional tools are licensed at 100k-200k/year per seat each, there is no way around them if you want to actually tape out the chip. An experienced backend engineer needs 3-4 months to take a design from Verilog to a file that can be send to a Foundry. You also need access to a PDK from a manufacturer like TSMC and probably will have to license some of the digital-analog components (PLL, IO-Pads) from someone. Manufacturing itself is fairly expensive as well, we taped out via europractice ( https://europractice-ic.com/schedules-prices/), you pay about 12k per square millimetre in TSMC 65nm. Even older processes are more affordable, but you have to keep in mind that each of them requires at least some customisation and access to blocks specific to the process (memories for instance).
On the design and verification side the story is slightly better, verilator (https://www.veripool.org/wiki/verilator) for instance is an excellent tool that lets you simulate system-verilog code with high performance. It has seen wide industry adoption in the last years and now a couple of companies like Google, SiFive, etc. are investing in open source tooling for hardware. You can go a long way to building your design without having to pay for expensive licenses. The downside is that most of the interface type components (PCI-Express, DDR4, etc.) are prohibitively complicated to build yourself, so you will need to rely on external IP at some point.
Yes, I can totally believe that, the software itself is actually distributed in a fairly convenient format, given that some of it is only compatible with some obscure RedHat version, so installation amounts to copying files from an iso image and pointing it to a license server. It probably is also easy to patch out the License server code. I would be fairly surprised if those places also had access to current gen PDKs and the required memory compilers. We have to basically order memory blocks by mail and don't get access to the tools to generate them ourselves.
To build chips similar to what TSMC produces (i.e. some of the most cutting-edge fab tech in the world)? Hundreds of billions of dollars. Decades of experience. Growing silicon crystals and turning them into transistors barely bigger than a few dozen atoms is immensely difficult. And then turning that into a process you can repeat with precision and churn out at high yield is another enormous hurdle.
Serious question as someone totally naive on this. A few hundred billion dollars seems like an extremely good deal to lead this area of technology - why don't we see funding for this at a government level? I mean, we now have a 2 trillion dollar infrastructure plan.
Is there an impediment I'm not aware of? Or maybe it's not as good of a deal as I think?
> why don't we see funding for this at a government level
Because America fell asleep at the wheel while Taiwan did not. TSMC and other Taiwanese chip manufacturers benefited from a government that saw the need to be a part of high value manufacturing. Meanwhile America of the 90s and 2000s assumed that it would be the most powerful country in perpetuity and so it wouldn't matter where something was being produced as long as they could pay for it.
The events of the 2010s has shown the flaws of this thinking. It is now possible for American firms to be cut off completely from semiconductor manufacturing, similar to how Huawei was cut off. There is now support from the American government to restore semiconductor manufacturing. They are footing a part of the bill for TSMC's new plant. I'm sure Intel's lobbyists are skilled enough to get part of their new plants paid for as well. China is doing the same - the state has deployed all it's resources behind SMIC to ensure that what happened to Huawei never happens again.
It's more complicated than this though. You don't just need semiconductor fabs, you also need rare earth metals, almost all of which is mined and refined in China. America is attempting to reshore this too (https://www.economist.com/finance-and-economics/2021/03/31/g...)
> need rare earth metals, almost all of which is mined and refined in China
This is usually presented as an ace up the sleeve that China has secured through shrewd strategic thinking. In reality, rare earth metals are neither particularly rare, nor are they expensive, nor are they in high demand.
Rare earth metals are more common than silver or mercury, somewhat rarer than cobalt. The ones with excellent magnetic quantities go for ~$50 per kilo (allegedly only 200g is required per electric car), the ones used for catalysts and alloy making go for as low as $2 per kilo, cheaper than copper.
They're not nearly as much of a constraint as claimed.
This is my understanding as well; one of the reasons why China has such a high percentage of that market isn't because no one else can, but rather that they don't want to. The environmental damage is often horrendous [1][2] so everyone that can outsources it. I believe there's also been a lot of work on reducing or replacing relience on those materials in quite a few areas, as I believe a parent comment mentioned.
Intel's new fab is in the US (where I reside), I hope they get subsidized as I'm worried about our manufacturing capability across the board. It's affecting what I'm looking to purchase as well, having a few issues with my Ryzen 5900X system and instability matters far more than performance so I'm thinking hard to an Intel alternative. "Slow" but stable, and investing in the US? Sign me up.
Modern semiconductors were invented in America, literally why we call it Silicon Valley. We used to have the best fabs in the world but 80's and 90's corporate American business culture decided it was better to shut down all that manufacturing and move it overseas for more short-term profit. They're now coming to the hard realization that maybe that was a massive mistake.
The story is not anywhere near that straightforward for semiconduction fabrication. A huge portion of that fabrication, for example, just moved out of california to other places in the US. Also there's been a general consolodation of fabrication plants over time as more and more advanced nodes put the price of a plant up by orders of magnitudes. There just isn't enough money for the same number of fabs at the bleeding edge. Also, intel are still US based manufacturing and they have only quite recently been toppled from the throne of most advanced process. They have simply lost to the advancement of overseas tech companies (generally it seems because they made a bad bet on their strategy for further shrinks to their process), nothing to do with US companies outsourcing overseas.
> There just isn't enough money for the same number of fabs at the bleeding edge.
There definitely is but it will come at the cost of share buy backs and keeping share price up which affects compensation of management. Failure to keep the share price up by investing in manufacturing in US will also invite predatory share holders to topple the management. The cult of "free" markets driving wealth growth at the top of the society seems to have trumped strategic common sense.
20 years ago I had the line, American businessman and political leaders are stupid. Because they think they own those factories in China and the far east.
The 70s started the trend of financializing (is that the word?) everything, that meant that at every step of every productive process someone packaged it into a tradable "paper", which got packaged with other paper from other productive processes and then re-sold and repackaged again and again to the market as proxies of the value from those productive processes.
Holding paper assets turned out to be much more lucrative than the actual production for most companies, not to mention far far easier than competing in the market with products and innovation, so naturally it slowly began to chip away at the quality of the production in favor of holding paper that represented production. Outsourcing became possible and they noticed it didn't matter to the ones buying the paper assets who actually did the work, so it skyrocketed.
Financialization required an immeasurable amount of debt, they began extending credit to anyone so they could buy said paper, and then turned around and made the debt itself financial paper.
That got us the stock market booms of the 80s, the 90s, the dot com, the financial crisis, and of course their busts.
Now we are at the end, and the signal? companies are adding bitcoin to their balance sheets. The cycle of finacialization is complete, paper is no longer required and you don't even have to outsource your production, not that you have it anyways, as all you need to post profits and pretend you are a good CEO is buy crypto and hodl.
Production for a 'modern' business is rapidly becoming a thing of the past. Governments can try to buy their way into any market but it is useless; most businesses have no real capacity to compete delivering cutting edge products with actual real "bare-metal" profitable businesses because they can only make profits "on paper".
(sorry if finacialization is not how it's written)
Most of these are providing goods and services that 100s millions of people are getting value from, with the possible exception of Facebook :p
As an aside I must admit I had never heard of Delta Electronics before at all; nearly 50 years old and 5th most valuable company.
Out of the top 10 you do get some pure play financial companies like Berkshire, Morgan Chase, Visa (maybe).
Cutting edge semi conductor manufacture is arguably some of the most advanced technology and manufacturing we do at scale. It's just plain hard hard. The equipment, the expertise, the lead time to manufacture, it's all big scale problems.
Relatively speaking getting manufacturing of old node sizes would like not be remotely as hard, 32nm maybe even 16nm.
Those companies, maybe with the exception of Facebook and Microsoft, are literally only possible because of paper money. Take out the paper they hold from their balance sheets and their size goes down in half. Take away their ability to own their markets by issuing debt bonds, which would mean competing with the revenue generated by their own products, and you kill them.
Apple became the biggest company in the world not only because they create products everyone wants, but equally because they are masters at managing money and debt. That is what gives them the edge in production as they barely produce anything. They are probably the best at managing financial assets by far.
And Tesla, well Tesla is very bad at building cars, they are bankrupt without paper profits. They shouldn’t be on that list.
Semiconductors is not harder than any other cutting edge industry that has come before. Taiwan is not special, they are 20 million people. A good chunk of them trained and educated in the US.
Also note that as far as hardware production is concerned, Tesla is the only company on that list that actually manufacturers anything at scale in the US anymore.
> Apple became the biggest company in the world not only because they create products everyone wants, but equally because they are masters at managing money and debt.
What, by not having any? The correct amount of cash for a company to have is $0 because it's not a productive asset, as you say, but Apple's strategy ended up with them having hundreds of billions of it overseas. That is literally the opposite of MBA philosophy, not an even better version of it.
> That is literally the opposite of MBA philosophy, not an even better version of it.
Who said anything about MBAs? The Apple way is clearly the better way to do it at the moment, and has been for the last decade. The MBAs at Intel are not even in the same league.
And Apple owns factories in the US too. They own them for other reasons, not because they "need" them to make products or are profitable on their own.
> Now we are at the end, and the signal? companies are adding bitcoin to their balance sheets. The cycle of finacialization is complete, paper is no longer required and you don't even have to outsource your production, not that you have it anyways, as all you need to post profits and pretend you are a good CEO is buy crypto and hodl.
This literally isn't possible because of how GAAP accounting rules work for bitcoin. You can't count the value going up if you hold, but you have to mark it down if it goes down. It's only bad for you.
The US dominant ideology is private enterprise, not government enterprise. This is why the US does not even own its high-tech weapon manufacturers, preferring to have Boeing, Raytheon and Lockheed-Martin at one remove. Of course, they wouldn't exist without government money, but they're private companies with private shareholders.
So when you're talking a few hundred billion dollars for a US-first technological capability, the important question is who will get the money and how will they be held accountable for delivery? What, even, are they expected to deliver - no use if the plant is more expensive than TSMC so everyone chooses them anyway.
The US is heavily investing in that, there are recent DARPA programs (POSH, IDEA, SHIELD, SAHARA, ...) addressing this and continuous government investment in semiconductor technologies.
having skilled peoples at every org level capable to do the job is the most difficult part. Any gov could spend Trillions and still get little in return
You can read around the parallax propeller 2 forums for an idea of what it takes on the low end - they were very open about the process. IIRC they have one guru who does software and Verilog, one who does RTL layout (term?) and the rest are support staff who do other things at the rest of the company most of the time (PCB designer, assembly, sales, HR, etc.)
The raspberry pi pico team also did a good podcast episode talking about this - I think it was a podcast called innovation coffee or something like that. Hosted by a guy from ARM Europe.
> How much would it cost to start a design semiconductor business? On the low end of the budget.
A decade ago, it would've taken you $1m in China for a 50/50 shot at it.
If your first tapeout works, you will make money, if not... you wasted $1m.
$1m looks like pocket change to most American tech people now, but the fabless trains has long since departed.
You will need a sum with few more zeroes, to get a jumpstart in the industry now.
There was 4-3 years long extreme consolidation push in the industry, with big swallowing dozens producers of commodity products. It is very reminiscent how the demise of American oil industry happened when the industry gone from thousands of oil producers, to less than 10 in one decade. Everybody bets now that there will be 10-15 or so mega-fabless running the industry in coming decade.
There are plenty of designs where just $10k will get you an ASIC built in some university... It won't be anything like leading edge silicon though, but for some products that's all you need.
A bunch of talented people with experience who can use the equipment productively = priceless. Probably can't be bought, otherwise some countries would already have done it.
It depends pretty heavily on the semiconductors you want to produce, if you want to replicate the technology of the early 60s people have done that in their kitchen (see cooking with Jerry for an example.)
