Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/testing-the-bounds-of-loyalty-for-tsmcs-inner-circle.17401/page-2
        )

    [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] => 2021370
            [XFI] => 1050270
        )

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

Testing the bounds of loyalty for TSMC's inner circle

The Motley Fool
Mon, February 6, 2023 at 8:14 AM PST

Shares of Taiwan Semiconductor Manufacturing (NYSE: TSM) were falling 4% in morning trading Monday at 10:37 a.m. on no company-specific news, but the stock has been on a tear so far this year, up 22% year to date. In a notable mention last week, Apple (NASDAQ: AAPL) CEO Tim Cook told analysts during the tech giant's fiscal first quarter earnings conference call that the company was "all in" on being Taiwan Semiconductor's largest customer at its new Arizona facilities. Although Taiwan Semiconductor has been able to sidestep many of the supply chain snags that have embroiled other leading chipmakers, its own latest earnings report had the world's leading pure-play foundry saying weakening consumer demand could result in first-quarter revenue dropping as much as 5%, leading to cuts in this year's capital expenditures compared to a year ago.

 
Isn't Intel competing with Apple's Mac processors, and Samsung with Apple's phones? That's the issue with foundries who are also IDMs.

Intel does not make Macs, they make CPUs for IBM compatible PCs, and there is close to no crossover, or migration in between the two.

Intel has left mobile market as too low margin on its own volition long time ago.
 
If you want to push the same device through multiple foundries (e.g. TSMC and IFS or Samsung) at advanced nodes, you have no choice except to have separate development teams, because the foundry contracts require that there must be zero chance of any technology leak -- engineers who work on a TSMC process have to be totally firewalled from employees who work on a competitor's process, to the extent of having completely separated teams, tools and licenses.

This pretty much doubles the back-end development effort and cost, so very few companies can justify doing it due to the cost and resources needed -- only truly massive projects with huge revenue would even consider it.

I never heard of anything like that
 
IanD said:
If you want to push the same device through multiple foundries (e.g. TSMC and IFS or Samsung) at advanced nodes, you have no choice except to have separate development teams, because the foundry contracts require that there must be zero chance of any technology leak -- engineers who work on a TSMC process have to be totally firewalled from employees who work on a competitor's process, to the extent of having completely separated teams, tools and licenses.

This pretty much doubles the back-end development effort and cost, so very few companies can justify doing it due to the cost and resources needed -- only truly massive projects with huge revenue would even consider it.

Paul2

I never heard of anything like that

Absolutely true. That is in the NDA / wafer agreement for all FinFET processes at TSMC. And with good reason.
 
Absolutely true. That is in the NDA / wafer agreement for all FinFET processes at TSMC. And with good reason.
I’ve always wondered about that. Seeing dimensions doesn’t really tell you very much. Obviously all IP is worth protecting, and you can gleam some information on what TSMC is doing based on dimensions and DRs based on what is physically possible with the current technology (eg DTCO or what materials might be used). But overall I can’t imagine this is very useful. You could have given me all of the CDs and told me every DTCO feature N3 has, but that wouldn’t be enough to reverse engineer it because the real magic is in the process and integration. I would need process recipes, which tools they used for each layer, and a year or two for process tuning.

Long story short I am unconvinced that chip designers could steal any highly sensitive IP. Maybe the exception could be folks like Apple. Although I suspect their collaboration with TSMC is more like asking something like “Hey we would like a tweaked voltage curve” with TSMC worrying about all the details of how you go about tweaking the voltage curve.
 
I’ve always wondered about that. Seeing dimensions doesn’t really tell you very much. Obviously all IP is worth protecting, and you can gleam some information on what TSMC is doing based on dimensions and DRs based on what is physically possible with the current technology (eg DTCO or what materials might be used). But overall I can’t imagine this is very useful. You could have given me all of the CDs and told me every DTCO feature N3 has, but that wouldn’t be enough to reverse engineer it because the real magic is in the process and integration. I would need process recipes, which tools they used for each layer, and a year or two for process tuning.

Long story short I am unconvinced that chip designers could steal any highly sensitive IP. Maybe the exception could be folks like Apple. Although I suspect their collaboration with TSMC is more like asking something like “Hey we would like a tweaked voltage curve” with TSMC worrying about all the details of how you go about tweaking the voltage curve.
It doesn't matter whether you're convinced about whether this is pointless or not, it's how the foundries work -- if you don't do this, you don't get access to the advanced process, so go somewhere else.

Oh dear, there isn't anywhere else... :-(
 
It doesn't matter whether you're convinced about whether this is pointless or not, it's how the foundries work -- if you don't do this, you don't get access to the advanced process, so go somewhere else.

Oh dear, there isn't anywhere else... :-(
Like I said ALL IP is worth protecting that is how ALL industries work. Even big picture process flows for well understood processes are kept under NDA. Everyone and their mother knows the big picture of how MSG gets made, but that doesn't mean that every little detail about Ajinomoto's process won't be kept under NDA no matter how basic the information. My point was more so that I question the idea that chip designs could glean any "critical" information. It would be like the information getting out that DuPoint uses the Haber-Bosch process for making fertilizer. Is this under NDA? Yes. Is Dupoint scared of this getting out? No. Either way ALL IP must be protected, if you don't then when actually valuable IP gets stollen you can't argue in court that the perpetrator should be punished because you haven't protected your IP before then.
 
This pretty much doubles the back-end development effort and cost, so very few companies can justify doing it due to the cost and resources needed -- only truly massive projects with huge revenue would even consider it.
Is there a scale of project where you contract for a team at the foundry to take your design from something like validate and tested RTL model and deliver it to the foundry model on contract?

Seems like there would also be few companies who can justify building that team even once, especially if they do not have an immediate following project, or if they do but would value being able to get bids from other foundries.

Seems like there would also be some issues with labor laws if not allowing individuals to practice what they have learned in other employment.
 
Is there a scale of project where you contract for a team at the foundry to take your design from something like validate and tested RTL model and deliver it to the foundry model on contract?

Seems like there would also be few companies who can justify building that team even once, especially if they do not have an immediate following project, or if they do but would value being able to get bids from other foundries.

Seems like there would also be some issues with labor laws if not allowing individuals to practice what they have learned in other employment.
That's not how it works, you'd have to subcontract it to an ASIC house (e.g. Synopsys) not the foundry. The problem is that the time/effort/cost of doing the back-end -- especially verification -- has been rising exponentially, and this now dwarfs doing the RTL which is the quick/easy/cheap bit -- and manufacture (mask sets) and validation of the finished device are also extremely costly. If you want to put a chip through two foundries, all farming it out to an ASIC house does is move the problem -- and then you get their GM added onto the cost of the project, so doing 2 chips through an ASIC house costs at least 3x what doing one chip internally does.

If the reason for doing this is to try and allow two chip suppliers to them fight it out on price, it's probably not worth it -- you'll never make the higher NRE back from a (small) die cost saving, especially when you consider the additional product support costs (hardware and software) for two devices, because they're *never* identical. If you're doing it for security of supply, the big foundries have multiple fabs anyway.

The other problem is that it's not until the back-end design/layout is competed (timing closure on a clean design) that you know what the die size and power consumption are, and historically there's been a significant difference between foundries -- and in recent years TSMC have always been the winner.

There may be labor law issues if you move to another job, but not within one job -- if TSMC put a contract condition in place which means you have to firewall off the entire design team from those working with Samsung, then that's what you have to do. Nobody's going to try and find a backdoor way round this because if they get found out TSMC can then shut them out, which would be disastrous.
 
No. Either way ALL IP must be protected, if you don't then when actually valuable IP gets stollen you can't argue in court that the perpetrator should be punished because you haven't protected your IP before then.
You seem to confuse Trademark law with Copyright or Contract law. Wether you protected other IP or not before does not impact in any way how you can protect new unrelated IP.
 
You seem to confuse Trademark law with Copyright or Contract law. Wether you protected other IP or not before does not impact in any way how you can protect new unrelated IP.
The way legal talked about it at my firm was that all IP needs to be protected to maximize the chances that you win any IP/trade secret litigation. The same concerns were also present for the minor/obvious details when I was working with a food company on a process characterization project. The logic seemed to check out for me, matched my prior experiences, and I thought who am I to argue with a lawyer on how court cases work.
 
The way legal talked about it at my firm was that all IP needs to be protected to maximize the chances that you win any IP/trade secret litigation.
Patents for corporations are valued rather differently than patents for an individual. Your employer may have mutual IP agreements with multiple other companies in overlapping spaces. Every few years these are renegotiated. Each company brings a stack of relevant patents to the table. An evaluation team, who may be independent, conducts an assessment of the value of each stack - picking out the most important ones from each stack and then sampling from the rest to assure overall quality. At the end of negotations, one company ends up paying a net license value to the other, they agree not to sue for N years, and schedule to do it again.

That is how they get along smoothly in mature industries, including semiconductors, where it is inconceivable that you can operate without using patented technology, and they do not want to be stuck in the mud arguing all the licenses all the time. They generally have no interest in invalidating or defending specific patents, the closest they may get to that is disputing a value assessment on something.

Another pattern is to form an entity, like MPEG or JEDEC, where licenses may be pooled or where standards may be agreed where it is believed (not always successfully) that the members contributed all necessary IP and have generally acceptable (FRAND for example) policies to ensure they can do business using the standards.

It is very different for inventors outside of that system, where big companies often react poorly and agressively, rather than constructively to the presence of IP outside their boxes.
 
Back
Top