TSMC To Customers: It's Time To Stop Using Older Nodes and Move to 28nm (anandtech.com) 77
AnandTech: We tend to discuss leading-edge nodes and the most advanced chips made using them, but there are thousands of chip designs developed years ago that are made using what are now mature process technologies that are still widely employed by the industry. On the execution side of matters, those chips still do their jobs as perfectly as the day the first chip was fabbed which is why product manufacturers keep building more and more using them. But on the manufacturing side of matters there's a hard bottleneck to further growth: all of the capacity for old nodes that will ever be built has been built -- and they won't be building any more. As a result, TSMC has recently begun strongly encouraging its customers on its oldest (and least dense) nodes to migrate some of their mature designs to its 28 nm-class process technologies.
Nowadays TSMC earns around 25% of its revenue by making hundreds of millions of chips using 40 nm and larger nodes. For other foundries, the share of revenue earned on mature process technologies is higher: UMC gets 80% of its revenue on 40 nm higher nodes, whereas 81.4% of SMIC's revenue come from outdated processes. Mature nodes are cheap, have high yields, and offer sufficient performance for simplistic devices like power management ICs (PMICs). But the cheap wafer prices for these nodes comes from the fact that they were once, long ago, leading-edge nodes themselves, and that their construction costs were paid off by the high prices that a cutting-edge process can fetch. Which is to say that there isn't the profitability (or even the equipment) to build new capacity for such old nodes.
Nowadays TSMC earns around 25% of its revenue by making hundreds of millions of chips using 40 nm and larger nodes. For other foundries, the share of revenue earned on mature process technologies is higher: UMC gets 80% of its revenue on 40 nm higher nodes, whereas 81.4% of SMIC's revenue come from outdated processes. Mature nodes are cheap, have high yields, and offer sufficient performance for simplistic devices like power management ICs (PMICs). But the cheap wafer prices for these nodes comes from the fact that they were once, long ago, leading-edge nodes themselves, and that their construction costs were paid off by the high prices that a cutting-edge process can fetch. Which is to say that there isn't the profitability (or even the equipment) to build new capacity for such old nodes.
My repair guy can't get parts (Score:5, Interesting)
My repair guy can't get parts for a $2500 refrigerator that's only 8 years old. Why? It's obsolete so I can't replace a broken piece of plastic. Here comes the JB Weld then, a piece of aluminum bar and pop rivets.
It sounds like the OEMs building systems based on older chip technology need to upgrade as well but unlike me, it'll cost them a few million each in re-engineering and certification.
Re:My repair guy can't get parts (Score:5, Insightful)
Re: (Score:2)
The decade old tech inside new cars are simple by today's standards, but they use dozens of them in each car. Many manufacturers use modules supplied by common vendors so that's even more of the same chips shipped. You end up needing hundreds of millions, thus the semiconductor manufacturers wanting them to get off old processes so they can make more of the chips on the same wafer to bring production speeds up.
Re:My repair guy can't get parts (Score:5, Insightful)
I admit, I'm skeptical when it comes to miniaturizing automotive electronics. Cars' electrical systems are DIRTY. Operating voltages in normal conditions vary the better part of 6V, and abnormal conditions happen to cars often enough that the electronics need to be durable in order to not leave the owner with an expensive problem due to a fleeting condition with no obvious visible cause.
If miniaturizing for cost-savings or to enable a fab to be upgraded for non-automotive applications results in expensive repairs to cars then I don't want that miniaturization to be performed. The demand for ICs isn't going down, stand up a new fab, develop new parts, prove their reliability, and then after that reliability for decades of use is shown, be in a position to stand-down older processes.
A car is not a short-term fad-purchase, it's a long-term durable good that needs to function for the better part of 20 years and 300,000 miles. If moving-on to denser processes jeopardizes this then that sort of change shouldn't happen.
Re: (Score:2)
They're talking about "upgrading" to 28nm... that's upgrading from a decades-old fab tech to another decade-and-a-half old fab tech...
Re: (Score:2)
28nm chips have far less voltage fluctuation tolerance than those done on 100+ nm techs which is the parents' point. It's nice that it's smaller, but it's going to blow up a hell of a lot more when it gets hit by "expected voltage + 6V" he mentions.
Re: (Score:2)
Exactly. When one starts a combustion-powered car, the nominal battery voltage of around 13V drops far enough in the course of spinning the starter-motor that many electronics brown-out and shut off. When an alternator charges, it's running upwards of 15V sometimes, all for what's dubbed a 12V system. And if a car's alternator is getting weak then it might not output anything and the car runs until its voltage drops below that which will sustain generating spark or controlling the engine. The other day
Re: (Score:2)
I will say that I currently run an older car that has a premium kit for cold weather that consists of both the heater and the battery charger. And battery charger part means that control deck comes with an option to constantly display current system voltage, which I actually keep always on because I'm a weird engineering nerd that likes to have additional mostly irrelevant data points for that one time when it might be relevant.
And I can tell you from long term observation over many years that my specific c
Re: (Score:2)
Also, I'm sleepy so I confused alternator with inverter. Sorry about that.
Re: (Score:3)
Cars' electrical systems are DIRTY.
That's what conformal coating is for. Bigger = durable, is the kind of answer a first year engineer comes up with before they learn about how to build reliable and durable electronics.
There's no inherent reliability problem that isn't solvable through basic design related to size (providing we don't talk about extremes, but then TSMC isn't telling Ford that it's next car needs to be an on experimental 3nm node either).
Re: My repair guy can't get parts (Score:2)
Its electrically dirty not physically jeez ok so I design a circuit and 5 years later a critical part changes to a newer Fab process and can't meet the original specifications I'm a noob?
That's not how it works
Re: (Score:2)
Its electrically dirty
Oh you're talking about the voltage problem. A problem simply solved through electrical design. A car isn't anywhere near as electrically "dirty" as many other applications. Again, "just make it bigger" is the kind of answer you get when don't understand how to solve the problem. Let the engineers engineer, and if they can't solve the problem fire them and hire someone competent.
Re: (Score:3)
I am an engineer for automotive electronics, and you have no clue what you are talking about so please stop (and no, its not "just a voltage problem")
Re: (Score:2)
Wait, no.
There's a reason why we invented things like caps and voltage regulators. These electrical problems are long solved, and while I agree that the power may be somewhat dirty, it's quite clean by the time it reaches the main cpu...
Re: (Score:2)
Automotive isn't "small quantities".
Smaller is relative. Honda total worldwide car sales 2019: 5.3M [statista.com]. Samsung Galaxy S10 series alone: 36M [gizmochina.com]
The decade old tech inside new cars are simple by today's standards, but they use dozens of them in each car. Many manufacturers use modules supplied by common vendors so that's even more of the same chips shipped. You end up needing hundreds of millions, thus the semiconductor manufacturers wanting them to get off old processes so they can make more of the chips on the same wafer to bring production speeds up.
How many chips do you think Samsung requires for their phones much less their entire line of businesses?
Re: (Score:2)
While I understand what you mean, perhaps I wasn't clear on what I said. When I said "hundreds of millions of chips" I was talking about a singular chip, not one of dozens different designs used for different purposes. While each one of Samsung's phones has a single large CPU and perhaps one or two trivial ones for other uses, a typical car has dozens and while Honda may make 5M or so cars a year that's a lot of weak-ass CPUs to sour
Re: (Score:2)
'Automotive isn't "small quantities".'
iPhone sales alone are 3x automotive sales. Who are you going to add capacity for: high-price high-volume customers, or marginal-cost cheapskates?
Re: (Score:2)
As another poster said, cars aren't low volume for chips. Car design involves lots of simple chips placed all over the car. Cars have decent volume, and many chips per car means they need a ton of chips.
But more importantly, the car manufacturers all expected demand for cars to drop significantly, so they cut their chip orders. The chip manufacturers responded by shutting down old factories that were no longer needed. Once that was done, increasing production became a massive task.
Re: (Score:2)
As another poster said, cars aren't low volume for chips. Car design involves lots of simple chips placed all over the car. Cars have decent volume, and many chips per car means they need a ton of chips.
I did not say low volume. I said "smaller quantities" as compared to other industries. I would guarantee you Apple uses way more chips than Ford. But in the case of a production issue, which customer's orders would you fulfill first. Apple who wants more chips but they are also newer and more expensive and did not cancel their order or Ford who uses an older line with cheaper products and fewer chips (and also cancelled their order). Ideally I want to fulfill both orders but if I had to choose priority, I w
Re: (Score:2)
I would guarantee you Apple uses way more chips than Ford
My quick back of a napkin calculations suggest Apple uses about twice as many chips as Ford, but if Ford made all EVs, they would use about twice as many chips as Apple. I'm assuming only 1,000 chips in a typical ICEV (it's probably more — a modern ICEV has a whole bunch of modules, sensors with digital interfaces, and so on) and 3,000 in a typical EV. Automakers alone currently use about 10% of the ICs on the global market.
Re: (Score:2)
The chip manufacturers responded by shutting down old factories that were no longer needed.
That's not what happened at all.
The automakers did cancel a bunch or orders, but that line time was immediately resold to other customers. You can look at the book to bill ratios and see them selling a lot of time when the automakers did a dumb. When they realized that demand was still high and tried to un-cancel, they were put at the back of the line. In the end you are totally right, they did it to themselves.
Re: (Score:2)
Automakers, defense contractors, aircraft manufacturers, EMS and SCADA controls suppliers, security system suppliers, and many other industries all need stuff that works, not stuff that's bleeding-edge. In many cases bleeding-edge is bad.
Re: (Score:2)
Re:what fucking difference does that make (Score:5, Informative)
I agree with you, but when the automakers cancelled the orders, the equipment did not go to the scrapyard.
That equipment hardly ever goes to the scrap yard. Is kept in the fab floor churnig chips for other customers if possible, and when the clean room floor space they occupy is needed, they are mothballed, put in storage, and sold to "lesser" fabs in china, SE asia, europe and the USoA.
Any of the readers wanna start their road to silicon mogul? Buy a small fab for pennies on the Dollar (for the floor space and clean rooms), remove the equipment, buy better used equipment, sell the equipment yo removed to even lesser fabs, start churning more modern chips, profit!
Here is one of the sites where you can buy used equipment:
https://www.globalfoundriesuse... [globalfoun...ipment.com]
Re: (Score:2)
Restarting a fab facility isn't trivial.
- Utilities restoration
- HVAC restoration
- Testing of every piece of tech (potentially over a thousand line items)
- Replacement of dead equipment (just like a complete datacenter outage, not everything will fire back up)
- Recertification of clean rooms
- Rehiring of staff
- Reestablishment of chains of command and management structures
- Supply chain reconnection
If you're going to restart an idled plant there has to be a really good financial incentive to do so. It's COS
Re: (Score:3)
Multiple major automotive manufacturers canceled their chip orders at the start of the pandemic. Then they wanted their place in line restored. TSMC and others shocked everyone when they sent all of them to the back of the line. The chip supply problem would have not been a problem for automotive if they staffed a small warehouse and kept their orders going. Other companies like Apple, Samsung, Sony, Nintendo, etc know better it seems.
In the end the auto industry paid foundries extra money to expedite order
Re: (Score:2)
TSMC and others shocked everyone when they sent all of them to the back of the line.
How did that shock everyone? If you cancel your order, you get put at the end of the line in priority.
The chip supply problem would have not been a problem for automotive if they staffed a small warehouse and kept their orders going. Other companies like Apple, Samsung, Sony, Nintendo, etc know better it seems.
For those companies, chips are essential to their business and they place a priority on their supply. Automotive did not place as high a value as I am sure they were more focused on things like steel, plastics, glass, rubber, etc.. Also to note, just-in-time is the current model for many businesses including the automotive which is part of the problem. Holding inventory would have been a major change to the
Re: (Score:3)
"How did that shock everyone?"
You underestimate the sense of entitlement that the auto industry feels.
Re: (Score:2)
ding ding ding. you are correct sir.
The auto industry is used to dealing with vendors that depend on their continued business to hold up profits. But it's not like they can take their business elsewhere. If TSMC treats them badly, they can go to Samsung and get treated the same way.
Re: (Score:2)
For those companies, chips are essential to their business and they place a priority on their supply. Automotive did not place as high a value as I am sure they were more focused on things like steel, plastics, glass, rubber, etc.. Also to note, just-in-time is the current model for many businesses including the automotive which is part of the problem. Holding inventory would have been a major change to their operations.
ECUs have been in cars for decades. And with the massive profits pulled in by infotainment trim options the automakers ought to be keenly aware that they depend on chips just as much as anyone else.
Holding inventory is normal practice for products with highly specialized components, long service life (such as factory warranty parts), or high per unit price. Just-in-time is useful in many stages but it isn't a panacea and you can mix and match appropriately. For inventory, you often contract another company
Re: (Score:2)
Long term client + stable business + large predictable orders barring force majore events. Those go in front of the queue in most scenarios.
But now is not most scenarios. Now is a period of extreme shortage.
Re: (Score:1)
> automotive which is part of the problem. Holding inventory would have been a
> major change to their operations.
Yes! EXACTLY THIS! In an effort to drive down costs, which includes warehousing of chip safety stock, the automotive industry, among others, has discovered the disavantage of "Just In Time manufacturing".
I remember reading an article serveral years ago during the Great Recession, where the larger
Re: (Score:2)
This is one of the main reasons why automobiles were hit harder by the chip supply problem. Whereas redesigning/re-engineering an electronic device would cost an OEM like Samsung or Sony some money and time, automobile manufacturers have to re-certify as safety is somewhat more important in automobiles.
That's an issue for stuff related to driving, but some of the delays in delivery have been because of ancillary systems like (on one ford model) touch screen rear climate control units. Those don't need to be recertified for safety, and Ford's not doing any UL-style testing of those units in any case. Nor is the supplier.
Re: (Score:2)
They did not take into account that their low cost parts would be on a lower priority in the event of a supply problem.
What they did was take the idea of "just in time" manufacturing and implement it wrong. They implemented it as "no inventory, for anything" manufacturing. Which generally used to work just fine when everything was stable.
The correct way to do it is to actually think about what parts - like custom chips - can't just be bought off the shelf from anywhere, and still keep enough inventory somewhere to ride out a manufacturer switch requiring re-tooling, maybe recertification, etc. Commodity parts, yeah, you can
Re: (Score:2)
"automobile manufacturers have to re-certify as safety is somewhat more important in automobiles"
-1, offtopic
Automotive certification only applied to automotive products.
Re: (Score:2)
A Z80 such as the Z84C00 family is still in active production. I doubt they will go away in 8 years, or even in 20 years. There are several other capable microcontrollers with very long support lives and decent supply chain. Generally any traditional businesses that use a warehouse instead of just-in-time manufacturing is immune to temporary supply chain crunch due to foundry changes or shipping backups.
Once you want a wifi-enabled fridge with a graphical multimedia interface, then you need a system-on-chip
Re: (Score:2)
Don't get a Smart TV. a Smart Fridge. or Smart Oven [androidpolice.com].
Good luck getting a non-smart TV with 4K DisplayHDR1000 support... And if any readers find one, please, for all that is holly, let me know.
And remmeber:
Monitor!=TV
MonitorTV
A TV has, among other things that monitors do not: a tunner (ATSC in the USoA, ISDB-b in my country) for Digital OtA TV, A remote (with a sleep function if possible), speakers* , and many more non-smart attributes.
Industrial signage monitors in particular have, among some undesirable things, cooling fans to disipate the heat of extended u
Re: (Score:2)
Fortunately it's (usually) pretty easy to ignore the "smart" parts of a TV. Just never connect it to any network, or use any of the "smart" functions. The biggest problem is often just that the "basic" functions like volume, source selection, and misc. settings get buried behind a "smart" interface - especially bad for sets with the dreaded "one touch controls" so popular in lower-end TVs. Leave it to marketers to try to make "we were too cheap to add more than one button, so you'll have to push this one
Re: (Score:2)
Indeed. The old stuff will not go away as soon as it has some reasonable level of performance. For example, it is hard to get 4-bit MCUs these days. But they were extremely limited. On the other hand, most 8 bit CPUs/MCUs are pretty versatile. Sure, you cannot put a lot of code or data in there, but functionality-wise if you want to control some electrical appliance these are entirely fine. Just look at what people do with the 8 bit version of Arduino and what it all supports.
I do agree that putting a multi
Re: (Score:2)
"A Z80 such as the Z84C00 family is still in active production. "
In any volume? Can I get a million a week for the next three years?
Re: (Score:1)
Re: (Score:2)
Mental that an 8 year old fridge is "obsolete".. I wish I had the knowhow to make appliances. Dumb shit with a realistic lifespan.
Re: (Score:2)
If you design it to be serviced then it costs more. And if you design it to be worth servicing then it costs a lot more. A rooftop AC for an RV doesn't even have service ports. They just fill it up and then fold over the fill tube and seal it. If you want to service it you can, you add a service fitting that pierces a line. But those fittings are common points of leakage themselves (they depend on a rubber washer type seal against the tubing) and more importantly, the unit is made cheaply enough that it's n
Re: (Score:2)
With the changes to refrigerants due to efficiency and global warming requirements, it's not that mental to think an 8-year old fridge is obsolete. But you should still be able to get it serviced, especially for a broken piece of plastic - that's all on the manufacturer.
to editors and the fancy people at anandatech: (Score:5, Informative)
and offer sufficient performance for simplistic devices like power management ICs (PMICs)
simplistic
adjective
treating complex issues and problems as if they were much simpler than they really are.
yw
(idk, this is people who make a living out of writing, you would expect they get the most elementary aspect of writing right first.)
Re: (Score:2)
I spent about 6 months fine tuning a PMIC driver for Linux. Just the different possible power rail sequences for boot, sleep, and wake was dizzying.
Certainly (Score:2)
Can you deliver?
No?
Well...
Migrate to better process nodes during revolutions (Score:3)
Right now, the automotive industry in particular, is going through 2 revolutions at the same time.
The most publicized one is moving away from fossil fuels and into electrification, either via all electric or plug-in (series) hybrid vehicles.
But there is another revolution, the automakers are moving the low voltage electrical systems from 12V to 48V.
Instead of keep using the same old 120nm designs, they should die-shrink and re-certify them in the smaller sensible node possible, or design from scratch for the smallest sensible node.
Of course, not all chips will work well at 28nm, but many a chip is being fabbed at say 120nm today that could work equaly well if redesaigned in say 40nm...
Manufacturers are eager to get rid of older nodes for many reasons, and one of them (not mentioned in TFS) is that many of the older nodes are based on 200mm wafers. Getting rid of 200mm equipment frees floor space in the very costly clean room of the fab to retrofit it with 300mm waffer equipment.
Even if the equipment you end up deploying is not leading edge (say, a used 300mmwaffer/32 nanometer lithograpy equipment), still this will allow fabs to increase output significantly, easing the chips shortage for everyone.
But the users of the chips (automakers, in this example) simply do not care to redesing (whining about shortages in front of the press is cheaper)
Re: (Score:2)
But there is another revolution, the automakers are moving the low voltage electrical systems from 12V to 48V.
48V was 'the next big thing' in the 90's-2000's, but my understanding is that effort died when high-voltage became available with hybrids and BEV's. Are you aware of any currently produced or planned vehicles with 48V systems? I may not be up to date.
Re: (Score:2)
But there is another revolution, the automakers are moving the low voltage electrical systems from 12V to 48V.
48V was 'the next big thing' in the 90's-2000's, but my understanding is that effort died when high-voltage became available with hybrids and BEV's. Are you aware of any currently produced or planned vehicles with 48V systems? I may not be up to date.
Start from here:
https://www.designnews.com/ele... [designnews.com]
then search engines are your friends.
Old chips on new fabs? (Score:3)
I don't know a whole lot about the details of chip manufacturing, and I'm wondering - why can't you just build a 100nm chip on a 10nm fab?
Having the smallest features be 10x larger than the fab could make is obviously a waste of potential, and presumably fabs charge at least in part by chip surface area, and they'd be competing for fab capacity with newer chips that would use the smaller sizes, so chip prices would presumably be considerably higher than if they were made on an older fab. But if it's a choice between more expensive chips and no chips... it naively seems like that would at least offer a stopgap solution.
Re: (Score:2)
Because 100 or so expensive 10nm chips are a lot more profitable than 1 cheap 100nm chip.
Re: (Score:2)
Sure. But this is about the "standard libraries", not elements designed by geometry. In most cases this will just be a recompile though.
Re: (Score:2)
TSMC has recently begun strongly encouraging its customers...
You have a good point! I guess charging customers the 28nm pricing to make their old 10um 4004's would 'strongly encourage' them to migrate.
I do wonder about the article's example of power management IC's, where a certain amount of real estate is needed for ampacity. Is there really much to be gained by shrinking the logic level circuitry?
...81.4% of SMIC's revenue come from outdated processes.
Hard to say a process is 'outdated' when it provides a majority of revenue for one of the world's largest fabs.
Re: (Score:2)
"Is there really much to be gained by shrinking the logic level circuitry?"
It's all about fab capacity. By shrinking the circuits they can go back to buying cheap marginal capacity on processes funded by high-volume high-performance customers. Or, they can pay the fully-allocated cost of adding fab capacity at the nodes they have already designed products. Either way, they need to pay for more capacity.
Re: (Score:2)
Re: (Score:2)
I think the big question is, what percentage of the chip is actually dedicated to logic circuitry?
If you're talking about power controllers, then depending on just how much amperage you're talking, and just how much complexity is in the logic, the logic circuitry might be only a few percent of the total chip area. If 90% of the chip *can't* be shrunk due to power handling requirements, then there's minimal gains due to be had by shrinking the remaining 10%. Especially if doing so risks reducing reliabilit
Re:Old chips on new fabs? (Score:5, Informative)
Some manufacture tricks depend on the feature size. In general one needs to reduce thicknesses consistently with the lateral dimension (thickness of metal lines in the chip, and thickness of the photoresist used for the process).
Let's assume they developed an efficient process that efficiently etches 10 nm lines into the substrate; it will need a tuning of the process conditions before it works for the 100 nm lines. For example if it's reactive ion etching (a vacuum process working with chemicals in plasma), the access of the gas into the narrower or wider lines is not the same, also the ion bombardment might be more or less efficient on the different areas. The etch rate is most certainly slower on narrower lines. The different etch rate might make the process less uniform across the wafer, which still might be fine, but has to stay within the limits agreed with the customers.
So someone needs to run tests, which means the money-making machine stops for some time, some weeks. After the batch at 100 nm is done and you come back to your precious 10 nm, there will be no guarantee it works the same, because of invisible details like the dirt did not accumulate in the machine at the same rate as previously so the machine is too clean or too dirty (but you can't see this so you don't know where the problem comes from). It will take time to recondition, maybe a batch of 1000 wafers will fail the quality inspection (all at the end after the 2 month to completion of all steps). That's lot of money at stake and lots of headache for the manufacturing team.
Ideally if you do production and you make money, you get one dedicated machine per project so you keep the projects separate and they keep working in a reproducible way. If you are into small-scale production, you have just one machine and adapt the parameters for each customer, then you spend lots of time conditioning and optimizing the process until it works again as it was before (wit the risk that it might never comes back to what it was).
Independently of manufacturability issues, if they get expensive machines guaranteed to work for the 10 nm node, they won't want to use them for less expensive 100 nm process, except if they start sell the 100 nm products at the same price (per silicon area) of the 10 nm, in which case nobody buys it.
Re: Old chips on new fabs? (Score:2)
The whole problem seems to be something like "we don't make enough money on the older nodes to keep them alive so please migrate to newer ones that add no value for you so that we can make more money". Why don't they just increase prices on the older nodes instead, until customers either decide it's worth migrating, or until the older processes are economical for the foundry? Then like in office space, "the issue will work itself out natural
Re: (Score:3)
Isn't the obvious solution just to raise prices on the older nodes?
The whole problem seems to be something like "we don't make enough money on the older nodes to keep them alive so please migrate to newer ones that add no value for you so that we can make more money". Why don't they just increase prices on the older nodes instead, until customers either decide it's worth migrating, or until the older processes are economical for the foundry? Then like in office space, "the issue will work itself out naturally". If you want to keep using old processes then pay what it costs to keep using those old processes...
Isn't this what people get fancy business degrees for? Is this supposed to be hard?
That's what they are doing, they are rising the price on older nodes.
But since they DO have bussiness degrees and it seems you don't (infering from the way you talk about said degrees), they know that they have to approach the situation with finesse, not bluntness.
For one thing, some of the customers on older nodes have a lot of political and economical clout. Not only the automotive industry. Actually, the automotive industry is the least of their worries.
For instance, RAD-HARD equipment uses (among others
Re: Old chips on new fabs? (Score:3)
I still think the industry creates problems for itself by failing to implement obvious solutions and failing to be the least bit creative sometimes. The entire "chip shortage" debacle is living proof. Only in business can you collectively fail so epically but still convince the world that some mysterious external phenomena are responsible rather than extremely basic supply and deman
Re: (Score:2)
I'm actually a semiconductor engineer, and I know that process roadmaps and EOL's are closely managed and clearly communicated.
I still think the industry creates problems for itself by failing to implement obvious solutions and failing to be the least bit creative sometimes. The entire "chip shortage" debacle is living proof. Only in business can you collectively fail so epically but still convince the world that some mysterious external phenomena are responsible rather than extremely basic supply and demand factors that are extremely easy to model and mitigate against of you just don't suck at the one thing you are supposed to be experts at.
Obligatory SMBC:
https://www.smbc-comics.com/co... [smbc-comics.com]
Re: Old chips on new fabs? (Score:2)
Re: (Score:2)
You would need to design a Xnm process that uses the equipment, materials, and steps already in place for the Ynm process. By the time you have certified that process, you have probably spent as much time and money as you would have putting up a fab for an old 100nm process that you already have certified.
For example, did the old Xnm process ever come on 12 inch wafers? Ynm never came on 8 inch wafers, so now you would need to characterize all the processing steps in your new Xnm process for consistent resu
Re: (Score:2)
Re: (Score:1)
The process is part art and part science. A 10nm process may be constantly tweaked so that the parameters come out correctly with each lot that runs through the fab. At my previous company, some processes were run in multiple fabs and the exact recipe would be slightly different in each fab, despite the fact that they might be using the exact same equipment. The upshot of this is that TSMC's customers would likely have to requalify their part, which defeats most of the purpose of keeping everything the s
Probably just a recompile (Score:2)
Well, in most cases unless somebody did something "clever". As older processes generally have more variation in electrical parameters (unless you go up to the cutting edge), this is most likely painless for most designs and may really improve them.
If it ain't broke... (Score:2)
Upgrade so that it is.
28nm va others (Score:3)
Re: (Score:2)
That is probably the reason AMD has a separate I/O die on their most recent CPUs.
It's called chiplet based design, and no it's not the reason. It may be a slight side benefit but that's about it.
Re: 28nm va others (Score:2)
Power problems make this not an easy die shrink (Score:2)
28nm is when leakage current became a serious problem. Not as bad as the catastrophe that is 20nm but still, you can't just die shrink a 40nm design and expect your power and thermal requirements to work out. In many cases, it is going to be a redesign. And that's just for the digital part. Most these legacy parts have analog sections that don't scale well at all.
If you want capacity, pay for it (Score:2)
All of these companies that need low-performance ICs rode on the coat-tails of the high paying customers, paying the marginal cost to fill in capacity as high-volume parts moved to newer processes. Eventually, they became the high-volume parts on that process, and there is no more marginal capacity. They are covering the marginal costs, but not the fully-allocated costs.
If you only want to pay marginal prices, expect to get only marginal volumes.
Unlimited growth?! (Score:1)
Chinese fabs (Score:2)
Dozens of comments in and still no one mentioned the elephant in the room.
Even with all the high-tech sanctions against China, older generation tools can still be bought, and hence China is building up lots of fabs that can produce older generation chips.
What TSMC do not want to see is its customers move the production of older chips to Chinese fabs (which will obviously offer lower price than TSMC), and thus cut into TSMC's revenue. By pushing their customers to 28nm-class or above level, they avoid compe