Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/serious-brain-drain-at-tw-fabless.21776/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Serious brain drain at TW fabless

Paul2

Well-known member
The laurels of nVidia are having detrimental effect on TW companies.

I wrote than around 2007-2009, you could have had 3-4 times lower net salary as a process engineer at a TW fab vs. Western first tier fab. But the gap for software, RTL, physical design people have gone to truly ridiculous size.

Some poached RTL designers with a few years of experience get $200k+ in nVidia now, while the started salary for a fresh grad was as low as $20k in top tier TW fabless like Realtek before a big jump a few years ago.

Many, many people in design space are jumping jobs now. Huge huge turnovers everywhere.

Many career trajectories I've seen were people jumping from top-tier TW fabless to no-name RTL shops on US. And most cadres lost were some of the most motivated, and driven people who were pulling their entire units.

I know Realtek has basically lost their USB-Ethernet business to poaching. Silicon Motion has lost so many know-how holders that their median retention is only like 2 years, most new hires only manage to finish a single chip design before leaving.
 
The laurels of nVidia are having detrimental effect on TW companies.

I wrote than around 2007-2009, you could have had 3-4 times lower net salary as a process engineer at a TW fab vs. Western first tier fab. But the gap for software, RTL, physical design people have gone to truly ridiculous size.

Some poached RTL designers with a few years of experience get $200k+ in nVidia now, while the started salary for a fresh grad was as low as $20k in top tier TW fabless like Realtek before a big jump a few years ago.

Many, many people in design space are jumping jobs now. Huge huge turnovers everywhere.

Many career trajectories I've seen were people jumping from top-tier TW fabless to no-name RTL shops on US. And most cadres lost were some of the most motivated, and driven people who were pulling their entire units.

I know Realtek has basically lost their USB-Ethernet business to poaching. Silicon Motion has lost so many know-how holders that their median retention is only like 2 years, most new hires only manage to finish a single chip design before leaving.

I have heard this as well. MediaTek is also getting hit. Nvidia is a VERY competitive company and compensation plans are one of their weapons. Take a look at the Nvidia parking lot here in Silicon Valley. The cars are outrageous! Reminds me of the DotCom boom days. There were Ferraris everywhere including Jensen's.
 
That's actually the biggest problem and threat for TSMC in the long term.
Even 2nd tier fabless companies get better talent than TSMC.

With nvidia coming, TSMC will get less talent young graduates in Taiwan.
 
That's actually the biggest problem and threat for TSMC in the long term.
Even 2nd tier fabless companies get better talent than TSMC.

With nvidia coming, TSMC will get less talent young graduates in Taiwan.
TSMC will just have to cough up no?
 
Isnt this the way!

If you are supposed to be the best in the world , shouldnt you be paying as such.

It's much more likely that the best are ones who stay on the workplace the longest. Their career is tied to some successful design they carry on their back alone, they don't have options to move, and continue it.

I've seen the situation when junior RTL developers leapfrogged their superiors salary-wise simply because they had more options to move around before their career took off, and they became unmovable themselves.

Salaries by far are not the only motivator. Taiwanese semi industry have largely evolved at the time when microchips were not a super duper profitable thing.

Most consider fabless design and RTL/HDL coding a very dull, boring, grind machinistic job where people go over million lines long netlists over and over for months.

Physical process was considered "interesting" and much more "kinetic" 20 years ago since you deal with exciting exotic substances, and scary looking machines costing a few times of your lifetime income.
 
Last edited:
They have to raise the salaries but I think it's difficult to compete with the package Nvidia Provides.
TheTW R&D Centre for major companies like AMD/Nvidia will result in brain drain from smaller companies
 
They have to raise the salaries but I think it's difficult to compete with the package Nvidia Provides.
TheTW R&D Centre for major companies like AMD/Nvidia will result in brain drain from smaller companies

Will the raising of salaries in TSMC (Taiwan) reduce the "cost disparity" between the Taiwan and US operations?
 
Will the raising of salaries in TSMC (Taiwan) reduce the "cost disparity" between the Taiwan and US operations?
Definitely not but it can certainly help with brain drain problem in Taiwan.
A good compensation package is one of the best way to acquire talent and retain talent
I never liked US cost disparity but there is nothing that matches it anywhere else
 
Does TW have an equivalent item to US H1 and other worker visas? That is another way many US firms try to deal with a lack of acceptable candidates (at the salary level the firms wish to pay). And, well, the Visa program has 'worked' in the US for quite a long time now.
 
Does TW have an equivalent item to US H1 and other worker visas? That is another way many US firms try to deal with a lack of acceptable candidates (at the salary level the firms wish to pay). And, well, the Visa program has 'worked' in the US for quite a long time now.
It's resulted in thousands of people who are barely capable and only doing the jobs for the money compared to back home, meanwhile skilled, experienced, talented people who just wanted to be paid appropriately for their talents are out of work or underpaid. A broad temporary working visa employment increases the likely hood of people staying after their visa has expired, as unlike a holiday these people have financial incentive to do so (note: I'm not saying that everyone on a visa is trying to be "an illegal", I do not think people in a country without a visa is strictly an issue - however these have been very relevant to the world and more specifically the given example of US of recent decades). Furthermore, if you're doing this kind of work, do you really want people that you're not sure you can trust doing it? Especially Taiwan where the Silicon Shield is a very, very real thing. For these main points, I do not believe a working visa plan would be a great way to primarily compete here for anyone involved.

TSMC will just have to cough up no?
This will result in higher wafer costs (already going up near exponentially) for their clients, and I do not expect any major chip company to just eat that against their profit - they'll pass it on to end users. Similarly, your comment about "the best company should have the best pay"[sic] - again, this will all come from an end purchaser. There is insufficient market elasticity for this to go any other way noticeably.

As for my thoughts on the matter - clearly, TSMCs renumeration needs to be addressed. I don't think a straight up pay increase would be viable - the sheer cost of it would risk bankrupting someone major - worker campuses, free lunches, etc and a long term plan to bring the pay up to relative parity ($200,000USD in the USA != $200,000USD in TW, not even close), and I think the retention increase needs to be a year or two in front of the hiring increase. Purpose here is twofold:
1) Stop the drain, better conditions NOW, and prove to your irreplaceable skill holders that it is worth them staying. Leaving your home and family for a 10x increase is very different to a 3-4x increase, especially if you have a partner or children.
2) Keep a trickle of new hires coming in who see value in staying - this is for your standard turn over (retirement / expansion based rates). If you're not hiring at this rate at least, you're effectively losing staff, and see point 1

As for what this would cost? I'm napkin mathing ~3x labour cost over a 5-10 year period. I don't reliably know how much of a wafer cost is labour value at the end of the day, a 3x increase on 10% is only a 20% total increase, so a 30k wafer is effectively 36k, but a 3x increase on 20% is 40%, leading to a 42k wafer. Break it down (5y @10% labour, 10y @20%) and you have a ~1.2k, or ~4% wafer cost increase PA due to labour, this leaves 6% for material, transport, cap-ex repay, and inflation increases, assuming they continue their 10% PI on a yearly basis (iirc this was actually for a 4nm node, but it's "the most relevant" data I'm bringing to mind), which I do not believe was the expectation, but I do believe would be absorbed graciously enough.



That being said, TSMC are certainly not "TW Fabless" - so what's happening there?

I could see Realtek being purchased by Broadcom - just to kill them to be honest. Marvell could have a go at it, but ehh, they have Aquantia for that market. NVidia could well snap them up, integrate Mellanox + RT into a networking umbrella, with RT for consumer / the poors, and mellanox for corporates. Intel... I'm surprised they didn't sell their NIC business off with the rest. Glad, but surprised. None of these will help TW, but Realtek doesn't help much either so what's the real loss.

And to be frank - Realtek dying would just make my life easier. Supporting products with their chips is just painful. Waste of time and resources usually when an intel card's only 3-4x the cost, but will just work forever (and save you on tech labour).

Silicon motion are in a really weird spot where... I don't think they have anything to be worried about in demand, by sheer virtue of regularly being the first to their market in volume. Their products don't have critical issues, aren't releasing outside of scope, and are highly affordable. That being said, if all their designers get poached, the first I'll know about their issues is when I have a thousand devices fail, so there's a real, real concern here. Frankly, the flash market is like Milk in Australia - being sold notably below what it should ever have been, but because it has been no one wants to pay any more than what it was. I cross my fingers and hope because their controllers are decent enough for me to want them to keep developing. I would love to see Silicon Motions share holder meetings, because they really could do with paying their staff better but would be better off not increasing prices (the USA should be begging for a reason to fund Micron) - unlike the overall silicon market, the DRAM / PCIe controller market has more elasticity.

How do you float companies like that generally next to NVidia? Well, more trained staff overall really. You make sure you have enough people to staff NVidia, AND a reasonable enough chunk left over that even with bleed off to the USA / Europe, you'll have enough people stay for passion. When that's not enough, what you need next is the government to step in and say "How do we make our people want to stay in their home country? What can we do for them to keep them?" Because losing workers to the USA IS a national issue, and it will need government oversight at the very least to improve the situation.

TL;DR:
Supply : Demand must always be considered with market conditions, wages between nations are rarely comparable, and paying more costs more. Fabless companies losing staff to NVidia should fall under monopoly concerns, as that's what will happen in the long run; fabless companies losing staff to overseas should fall under a national economic stability concern; both of these happening at once is what we have. And thanks, now I've put all my thoughts down, trimmed out some weird relics... I'm more worried. dang.
 
Last edited:
And to be frank - Realtek dying would just make my life easier. Supporting products with their chips is just painful. Waste of time and resources usually when an intel card's only 3-4x the cost, but will just work forever (and save you on tech labour).

I can't tell of anything more than so broken, and so profitable as RTL8153. It's a very very old chip, more than a decade old. Yet, it's the most popular gigabit ethernet chip on the market. It goes everywhere, from ethernet dongles by millions, to laptops, and TVs.

It has a few pages long bug list, it had many revisions over the last decade, yet some of their oldest items on errata lay uncorrected. Why? Because Windows still reign supreme. All issues they had, were previously fixed with driver workarounds, which were easy when Windows was the sole thing you had to care about, and Apple was forced to copy Realtek's driver workarounds simply because RTL8153 is everywhere.

Realtek owes almost everything to the PC era, and it is still retains a grand reputation among TW semi companies. Back then, it was possible to sell a single chip without changes for 10 years, now it is not. Talent was both unlimited, and not moving around in the nineties, and noughties, now it is not. Running a semi-business was unbelievably easy in comparison to today.

The bane of Taiwanese tech industry is very aged leadership. They are all made by their formative experience during PC industry's heydays.
 
That's actually the biggest problem and threat for TSMC in the long term.
Even 2nd tier fabless companies get better talent than TSMC.

With nvidia coming, TSMC will get less talent young graduates in Taiwan.

I think MediaTek and the other TW fabless companies are more worried than TSMC. There is definitely a divide between design and manufacture. Working at TSMC holds a lot of prestige in Taiwan. TSMC also has a good bonus system that is paying off in the boom years. The best is yet to come for TSMC employees, absolutely.

 
I think MediaTek and the other TW fabless companies are more worried than TSMC. There is definitely a divide between design and manufacture. Working at TSMC holds a lot of prestige in Taiwan. TSMC also has a good bonus system that is paying off in the boom years. The best is yet to come for TSMC employees, absolutely.

100% agree. TSMC will be fine. even if they increased wages 30%, that would have trivial impact to wafer cost (labor fraction of wafer cost is small and dropping)

the real challenges are to smaller companies, design houses etc. Taiwan has plenty of smart hardworking people to support large companies.
 
100% agree. TSMC will be fine. even if they increased wages 30%, that would have trivial impact to wafer cost (labor fraction of wafer cost is small and dropping)

the real challenges are to smaller companies, design houses etc. Taiwan has plenty of smart hardworking people to support large companies.
TSMC can simply pass that cost to Apple Nvidia and everyone they have the power over them others not so much unless they have some sort of non compete clause
 
Silicon motion are in a really weird spot where... I don't think they have anything to be worried about in demand, by sheer virtue of regularly being the first to their market in volume.

Update on SMI

They are not selling well these days. Their PCIE gen 4 controller are not selling well at all despite otherwise very good performance, and consumer response.

MaxIO fell upon the flash controller market like a brick out of the sky with both cheap, and always available in stock MAP1602. N12 MAP1602 is still beating competitor chips fabbed on nodes as recent as TSMC N6.
 
Back
Top