I retired in 2015 after 42 years in Wafer fab manufacturing, I recall this statement by an economist (I think) about this industry, “I’ve never seen so many smart people working so hard, for so little”. I couldn’t agree more! It’s a damn cutthroat business, you’re always on call, no private life and the reward are “ok lah”. I was employed in a European, later American owned wafer fab but I heard even more horrific stories from the Asian owned one.
The semiconductor industry experienced a demand fluctuation and spike that was outside of their operating tolerances, should have installed a nice demand capacitor tied to ground, the problem is the whole legacy system acts like one big inductor.
Some wafer technology doesn’t scale to larger wafers very well. Particularly non-silicon wafers that can only be produced in 150mm and 200mm sizes before inherent wafer defects create significant yield problems. 300mm equipment may double output, but it is often 5x+ regarding capital expenditure and maintenance cost.
Also, half the tools I work with on a daily basis are older than I am. And I’m over 40. Compound semiconductor makers almost exclusively inherited equipment from the 1970s-early 1990s silicon world because that’s all that’s available in many cases. What’s left are often 200mm tools down-converted to 150mm. With those kind of capital restrictions, an industry that charges cents per unit for its product and can fit 5,000-15,000 chips on one 6” wafer is never going to transition to 200 or 300mm.
Just to the last part: Any (not only multiples of five) "regarding capital expenditure and maintenance cost" is irrelevant when the net income(here the predicted value is sufficient) is above the previous one, the old production process provided. You just have to find an investor. That is no problem for such biggies like TSMC oder Siemens ... there are extra institutions there, invented just for that case which are happy to "assist", like banks and shareholders. But that is basic capitalism 101 and I don't want to bore you, Alexander:) In a nutshell: If it's worth it, then anything will be implemented as quickly as possible. Lemme close with that picture: Haven't you ever seen a cartoon rabbit with dollar signs in its eyes?:))) Don't get me wrong: Physical restrictions exist and no human, no Solar Roadways, no SpaceX and no Elon can avoid them. If @Grak70, for example, cannot even produce the raw material in good quality, then the discussion is actually superfluous. What doesn't work, doesn't work!
One note. It's basically impossible to upgrade a 200mm fab to a 300mm one. Every single piece of machinery in there is designed for the specific wafer size so you'd have to replace everything. The only "Advantage" would be you wouldn't have to build a new building but honestly the empty building is the easy/cheap part. It would probably cost way more to take out all the 200mm tools than it would be to just build a new building next door for 300mm tools.
depends and also that statement is often false if we are talking about places outside of Asia where land doesn't sell for pennies and building permits are a huge thing. (Even in Asia it's sometimes false) A fab is not even remotely cheap or easy. Even if you ignore the millions you'd have to spend for the land then permits and then a building, you also need a lot of time to build one, and in this industry every hour is basically millions lost that could have been earned. Besides that getting the whole facility upto a working level and have all the systems- from dust and air filtration to EMF protection to fire supression to ensuring water supply and whatnot is a major task Anyone who has worked in datacenters knows that the most expensive part of a datacenter is the building itself- aka why so much focus is put on compact racks and fitting as much power in 1U and such. Similarly a fab definitely has more expensive equipment in there but the building is definitely not the cheap part- that'd probably be electricity
@@KryoNaut Even with land costs and how complicated the empty fab building is trust me the building is the easy/cheap part. I can't get into exacts but the single cheapest machine in a modern 300mm fab is easily several million dollars and takes months to install and get it working on wafers. Which goes back to my previous point it's almost always a better idea to build a new fab than to try and convert a 200mm to a 300mm. Unless you had a 200mm fab in downtown Manhattan or something.
@@Kevin-xw1eo Considering a photolithography machine for single digit/close to single digit nm fabrication processes is around 170-200million USD something as large as 200nm with the whole thing should be into several millions for sure but not enough to call the building cheap part. You are just exaggerating at that point though I'd agree building a new 200nm fab is often cheaper than 300nm.
Respectfully, shutting down a 200mm fab to retool and upgrade to 300mm would be a grave mistake relative to building a new 300 mm clean room on the same fab site. Building a new fab would nearly always be economically advantageous especially if the 200mm fab is at max utilization
Yeah I agree. Those companies need to produce WHILE a refit would be happening? Not easy to do, so in general fabs get built and then it stays the same with few exceptions. TSMC can afford to retool fabs, such as taking a 5nm fab to 4 or 3nm because most the equipment is the same. One thing about the fabs being built in AZ by TSMC is, it has the space and will probably become a full fledged giga-fab like they have in Taiwan. It gives them the ability to have 6 fabs in one place. I'm guessing they'll expand that plant with N4 which is a retooled N5, and also N3 which is a new line. When TSMC comes out on N3 the demand is going to be CRAZY. I hope Intel succeeds with their Intel 20A. If they do they'll have the most advanced node in the world at the time (late 2024 early 2025). That would take a lot of pressure off high performance nodes and then nodes like 12 - 7nm should drop in price.
@@Conservator. per transistor cost bottoms out at 28nm, so maybe there's a way to regain some of the cost advantage vs. 130nm? It's just that there wasn't enough pressure to build more, not enough to offset the risks. You do have a problem that IO area probably won't shrinkify much at all though.
@@cnordegren The fab would lose money during the shutdown period but that money could be regained in the period thereafter when they would be producing double the output at roughly the same costs.
Voltage regulators are also tough to buy. As an EE, I won’t cut a PCB design loose until reels of parts are sitting on my desk, and then I over-order which doesn’t help the shortages. I’m buying parts for life of product, knowing many of those parts will be eventually excessed.
Make sure the part you choose has many alternates from different manufacturers with the same footprint - jellybean parts like voltage regulators are the easiest ones to swap, even if it's a PIA mid production. A lot simpler than porting a bunch of firmware from one processor architecture to another.
Yup. I tend to finalize designs only once I find supplies for at least 2 alternatives. I seldom have the ability to source everything in advance of final design. Often this means extra pads for different pinouts or footprints. This also helps when sourcing parts for repair if your device is one that may see service.
I so FEEL you! Nowadays I first order enough chips to build out at least a couple of prototypes and then I start with the new design, otherwise it doesn't even make sense to even start (for the major chips at least) :P With voltage regulators I haven't had a problem tbh, but in terms of power electronics what I have had a problem sourcing is power inductors and transformers.
I understand your position and appreciate your frankness. By your own admission you’re making the problem worse. (That sounds more accusative than I mean to say, I’m not a native speaker) How long is the life of product that you’re talking about? Im just curious if it’s something like a year or several years. If you would stock for, let’s say, 6 months, wouldn’t that give you enough time to resupply? Again, I’m just curious and interested in your views as someone who works in the industry. Thanks for your comment.
@@Conservator. as of now, most microcontrollers have a lead time of over 12 months and it's changing all the time. I have an order for an ATSAMD51 that is already 1 year old, they increased the price twice during this time and I'm still waiting for them to deliver. Who knows what the lead time will be in 12 months, I think Louis Brown is right, better to have all the components in stock when launching a product.
Only a site like this would provide the 'missing information link' about what is actually going on in the industry and what is causing what to go forward or not going forward .. Bravo ! for such unique knowledge and smooth delivery that are so lacking in this world, when everyone is good at making noises, but so very few that can actually make sense ..
I worked in the industry about 25 years ago. While much has changed since then, some things have not. As you stated there are basically 2 manufacturing processes, memory and logic. Within each process, they are optimized for the highest yield. This means designs are "tweaked" to work to a specific process and the process may be "tweaked" to a specific type of chip. A fab making DRAM may NOT quickly be able to switch over and make NAND memory. Worse, combining memory and logic will never be "optimal". Today's CPU chips have large amounts of cache memory. SOC (system on a chip) have an even worse combination; logic, memory, both cache and system SRAM, and NAND flash. Many have analog portion for AtoD or DtoA. Automotive chips do not want to drop below a 5 Vref so now you have multiple voltages.
You can buy passives, though perhaps not the one you want. Processors and flash, it's hard to make similar substitutions. Right now, most microcontrollers I've been looking at have 52 week backlogs that are staying that way from week to week. I've heard that it will be this way well into 2023.
It can't be TOO bad because new electronics are coming out hot and heavy in the world of computing, but those companies wouldn't have been stupid enough to cancel orders. Graphics cards, motherboards or mainboards, even CPUs have those components especially those little tiny resistors. So, what might have happened is some companies cut their orders because they felt demand would drop because of Covid, and the compute companies like AMD, Intel, Nvidia and all the companies that work with them (long list) bought up the supply that became available. Don't a lot of those components come out of Japan?
@@jaymacpherson8167 And I'm just reporting what I see right now as I follow the computer industry. The world of GPUs has gone through a big shortage for instance because of over-demand due to crypto mining. Now the supply of GPUs from multiple companies is excellent. AMD was stretched really thin because their products have been in high demand, from CPUs for PCs and servers, APUs for game consoles, GPUs for desktop graphics and GPUs for server compute. They now have good stock of most products so they've been cranking out products. Same thing with Intel and Nvidia. So the difference is where these companies were sitting before Covid hit. Back in late 2019 the world of compute KNEW they were going to have a problem meeting demand for about 2 years so if supply opened up they bought it. Same thing with Tesla. They do preorders. They KNEW Covid wasn't going to slow their sales and now they're pumping out vehicles, no problem. Also in China, companies are pumping out different products while others took a hit because they projected a drop in sales. That's what you get with just-in-time manufacturing. And once again almost every PCB needs those tiny components you're talking about ESPECIALLY mainboards and GPUs.
There has been a refusal by many companies to migrate 0.25, 0.18, 0.13um parts to advanced nodes on 300mm wafers. They have been focused on reducing costs by fractions of cents per part and driving ASPs down. They are not used to scaling. Almost impossible to get their designers to move to a new technology.
"to get their designers to move", why, no reason! If you get 5$ for a before 1$ part there is NO reason to port. They can milk the market up to the time a greater work around get substance. SEA need to strenght there own design capacity, the capacity is in SEA. Atmel bought a small uC C compiler design group in Norway (AVR), assemble some flash memory and some medium performance AD and fill an old fab. They will never port.
Analog and mixed signal designs can be moved to more advanced nodes, but their design techniques need to change. This means they need to hire different designers who have such experience. Also there is benefit of including additional capabilities in the chips.
Those talents are extremely difficult to come by and it takes 1/3 of a lifetime to train.. meanwhile the work is too hard for most people to (even) imagine.
@@westerncivilization Plenty of the designers in Korea and China from what I see being published. Much more advanced than what I saw at a large semiconductor manufacturer in the US.
If these designs are only moved to 28nm and larger nodes it won't be that big of a deal. 22nm and lower the gates are no longer planar and this complicates things.
My VLSI professor told us 3-4 years ago this is going to hallen. His reasoning was basically that it takes a very long time to build a fab until it is fully operational so companies need to predict demand 10-20 years ahead. Of course they factored in growth but they have either underestimated that growth or they were too risk averse (building a fab you won‘t actually use fully is a very bad idea if you want to continue to exist). And roughly 5-10 years ago you were already able to tell that the new capacity never quite fully matches the increased demand. He reckons it is going to take a decade to fix this supply issue fully, so I am sorry fellow EEs, you will likely have to do logistics for a looong time to come. (he doesn‘t actually hold a professorship at my university and always tells me I shouldn‘t call him that but it is just easier to understand that way)
the biggest joke of this world is this: We build useless empty cities and condos for billions, it wrecks our economies, makes us print money, causese inflation... yet we hesitate to build chip plants because "oh my god they might not be used"....
So true!! It's very much the case of "go big or go home." Companies are very conservative because cutting edge new fab lines now cost so much that just one wrong decision can sink the entire company. If they play conservative and don't expand production aggressively, they may miss out on some business but also, because there are increasingly fewer big players, the decision not to expand can actually float the prices higher due to reduced availability. One just have to look at the crazy ups and downs of the DRAM sector over the years/decades.
Good video as usual. Some thing you don't mention, though, is how inflation will affect the "inflection point" where a cutting edge plant can now price down it's product. In essence, that plant was built with usually debt, and that debt is not inflation indexed. So when prices rise, the relative cost of that debt is lower. If the profits necessary in nominal terms to recover the sunk cost of a 8 nm plant was $1BN, and due to inflation, the chips now sells for twice as much, that plant needs to put into the market about half the chips (not exactly half, because you still have some variable costs of labor, changes to taxes - going the other was as governments move to subsidze chips) for that sunk cost to be recovered and the technology to become "mature", or, as you say, comoditized. There is a lot of complexity to that, as inflation isn't universal in the global economy. Some currencies are losing more purchaseing power than others. And large share of chips are sold across borders. But rather than looking back to building the old mature tech, what you are just as likely to see are faster-than-anticipated decline in price premium for the more advvanced chips. This is good for consumers, and, depending on how each industry adapts - e.g., most automotive being much slower and less skilled at adapting to rapid change than, say, mobile handset makers or (allegedly) Tesla -- you may see divergence and market valuations of companies will diverge even more by sector. If, for example, a previously dismissed as mature industry may be able to step up to more rapidly commoditization chip architecture than some of those already pushing the envelope, and they should get a competitive edge and should see a boost to their valuation.
In Japan a 28nm fab is being built by TSMC. I'd call that legacy. Plus, once TSMC has a lot more production on N3, N4, N5, it will rebrand what it calls trailing edge, so there will be more availability of nodes like 12 - 7nm. Now, for some electronics you can't drop down too much because the IC would end up being so small that it's too hard to cut the wafer for the hundreds of die that would come from it. So, simple circuits that still warrant an IC but only have a few hundred transistors need to stay on those older nodes.
I imagine shrinking the node size makes them less hardy as well, with respect to voltage fluctuations and static. Sometimes handling abuse is part of the requirements. Also a factor in rad hardened chips.
@@ricardokowalski1579 Yeah it's an important issue in considering how small of a node you can put a circuit on. In fact that's why Japan wanted a 28nm fab from TSMC built in Japan, for general electronics and not HPC. Even with 7nm you're at 90 million transistors/sq. mm. What happens as you try to make a die smaller and smaller, you end up getting a bad ratio between edge area to have a border for the cut, and the area of the actual transistors. It's a REALLY hard answer to find but I saw a comparison with 2 different node processes, 90nm and 65nm and with the 90nm it said the smallest die could be 0.683mm x 0.683nm and for the 60nm node it was 1.533mm x 1.533mm. It gets to, how fine can you make the border between each die, and I have to think getting below a die size of 0.5mm x 0.5mm is near impossible. So, with a 7nm node you can fit 22 million transistors in that small of a space. But then how do you make the connections between the die and a PCB? It's not only the fact that you need an edge area for making cuts between the die, but you also need wires to come out of the die to create some kind of package. Can you fit 1000 connections in a die that's only .25 sq. mm?? I think not but I could be wrong. Price becomes more expensive as the node becomes smaller, so this is a REAL issue. I'm sure there's variation in pricing so at some point a smaller node might actually be less expensive, but once you get down to the state of the art nodes, there's pretty much always a rise in cost for smaller nodes. At some point you can't make a die any smaller, so if you hit that point with a 90nm node, it's simply wasting money making that die on a 7nm node. On top of that you have other issues come up as the node gets smaller as other comments point out.
I'm not a computer guru but after the next Carrington event y'all will be talking about the most anti electrostatic chips out there. Screw the size. I'll take functionality and reliability over the latest greatest ideas out there.
@@kirkthiets2771 There are effectively 0 ICs that can handle that. (MAYBE some satellite ics that can, but basically nothing consumer that is used on the surface of earth.) I think even the very first transistors will burn out instantly, not even mentioning ICs. Tubes are the only things that can reasonably stand those kinds of voltages. Russia is still using tubes in certain military applications btw ;)
We started pulling components off recycled PSUs in our workshop and have started selling on eBay to get low volume amounts of used product to people who need them :S
These older node chips aren't just in consumer devices. They're critical in industrial manufacturing components which can't be easily swapped out or upgraded. It's weird how gas can triple in price in a couple months, but these components can be sold out for years in advance and up until they ran out of them they were selling at the same price as always. If the PS5 is still out of stock for the foreseeable future, maybe $500 just isn't a reasonable price to stick to. If Atmel and the like would double or triple the prices, we could at least be making sure they're getting to places they're really needed while also allowing them to expand capacity profitably. I don't understand why technology prices all seem so immovable.
I agree with your goal, except I think the approach of relying on supply and demand is not explicit. It's forceful and does not serve the customers which in turn will not be good for the market as a whole. I studied marketing in college. Personally, I don't know the exacts is this market. For example: Are opportunity costs incentivizing large corps into buying up all the stock ? Are contracts established first, before chips are made ? My point in those questions is that I personally don't know enough details to know exactly if there are problems which need to be solved. However, f the goal is for "supply" to exist for specific market segments then a more explicit approach would yield better results. The solution is to allocate a certain quantity of the goods for each market. This would solve the problem. The approach of relying on supply and demand is only a viable solution IF that market segment is unwilling to pay the higher costs. That is a hope and a gamble. Also, with that in mind, how sure are we that the end consumers would buy these goods at elevated costs ? >> If the companies are unwilling to buy at high costs then why would consumers ?? ((Just to be clear, my assumption is that large corporations bought out all the supply and prevented supply for end users. The chips went into industry rather than sbc's / tinkerer's / schools / hobbyists. )) Again, I think a more explicit approach is better. If half the stock was allocated for end consumers/small companies and the other half for large corporations then we would be sure of a fair chance. I think a time frame should also be set so stock doesn't sit unused. >>> 50/50 split for 5 or 10 years sounds like a good plan.
Chips are increasing in price and still being sold out. Moreover it's much easier to increase prices in consumer devices rather than industrial devices. A consumer will buy a laptop or a gaming machine without thinking too much of minute price fluctuations. Even a 50 dollar increase in a laptop would not be noticed. In an industrial product the prices and volumes have been negotiated way before the product is even designed. Many times also your customers are pretty fixed and you have long running relationships with them, thus a small increase in your sales price would be noticed immediately. Nobody in the industry would pay double the price for a microcontroller unless they have really no other option. Moreover big companies like Microchip or Atmel don't operate like a retail store, where prices can fluctuate daily, the prices there are pretty fixed and customers expect that as well. Nobody expects that a microcontroller is gonna cost 50 dollars and noone is gonna pay that price as well.
@@InTimeTraveller Totally agree. I think this whole chip shortage thing is new. It's so new the market hasn't reacted yet. This seems like the initial the shock. Hopefully, it gets sorted out in a good way. Personally, I hope supply is evened out by making sure the stuff goes to the right people. Like >> prevent scalper from hoarding ; don't let companies buy up for crap IOT like alexa thermostats and wifi microwaves. I feel like some of those crap IOT devices are not adding much value to our lives, instead they are adding money to the corporations. There "are* cool IOT devices, but just adding Alexa to already existing tech should have common sense applied. I think the arduino & SBC types of devices should get priority caz tinkerers have a good chance at making contraptions which will improve our quality of life. For example >> the raspberry pi tinkerers were really into car mods b4 it was standardized as it is now >> rear camera's + integrating the phones into the stereo.
The leading/bleeding edge may be sexy, but so many industries run on jelly bean parts that you just can't get -- analog parts, mixed signal, power supply/regulators, and so many more. Quoting me an 18 month lead time on some of those jelly bean parts? Kiss of death!
we buy plcs and other automation equipment . we don't get down to the chips. we just need plc's . everything is backordered for months/ years. we go online and try to order . suppliers has 2500 or so on order from the manufacturer but are only getting like 100 every few months. none of these chips are leading edge. so we are fucked i guess.
Great video! I like the details, where a lot of other videos on the topic offer vague assertions without backing them up. I also loved the Arizona water joke. Seriously, I am amazed the problem isn't brought up more often, with the worsening draught and all.
It is not only microcontrollers. A lot of other analog and mixed signal chips too. ADC, DACS, motor drivers, voltage regulators/references e.t.c. Usually they are in stock at speculative resellers (for abysmal prices), but not in a normal one's like Digikey or Mouser.
ADCs are sold out, luckily there are still some available. Couple weeks ago I went with MCP3421 while considering ADS1115 as backup when more channels are needed.
@@raul0ca Well the process of making ethanol from corn takes more energy than you get in ethanol. So its a totaly wasted efford. Only sugar cane and palm oil can make ethanol with an energy surplus. Its just the lobby from big AG that gets the politicians to do such things, cause it will send the corn price skyhigh. The corruption... I was hoping US would remove the ethanol, to help with the hunger crisis that is happening in the world atm. But they did the opposite, and increased ethanol in gas... Making world hunger worse cause price of corn will go up.. Greedy mother fuckers, is all i have to say.
That's pretty curious. Looks like the market doesn't need that much the newer technology, but still needs LOTS of the old stuff. Really curious indeed.
I don't know why it had never occurred to me that there is a brisk market for 25+ year old semiconductor device fabrication equipment, and that they are continuing on in revenue service cranking out in-demand chips. So there is no ongoing production of new equipment for old nodes? What is the oldest node (and largest feature size) still being used to make chips?
Some laughably old nodes are still in production for niche markets - HV CMOS (up to 300V) and CMOS image sensors are particular examples where "cutting edge" is in the 300nm range
I'm not that surprised, to be quite honest. Some chips are now very old but have simply settled in the industry. I don't think there are many embedded engineers who have never touched an ATmega328. And you don't need a 4nm node to make those. But u can easily tell that a lot of places have them either put of stock or at increased prices. They've always been abundant, and very popular among hobbyists and R&D engineers. But they've become so much harder to come by.
An enormous amount of III-V production has never scaled beyond 0.5um. There’s simply no need in many cases, and the tools and processes used to create them aren’t capable of doing much better.
You can see a list of fabs here: en.wikipedia.org/wiki/List_of_semiconductor_fabrication_plants. 350, 180, 90, and 45 are all still pretty commonly used for chips where smaller process nodes are not needed.
An amazing analysis, as always. Thanks!!! So many good ebservations in the comments, too. The video was looking at the main source of the current problems, however, it's also worth mentioning that chip designs are tuned to a specific implementation of technology at a certain node, so production can not be easily shifted to another fab line, even at the same node. The more advanced the technology node, the more this is the case. This creates inflexibility, so even when there is wafer capacity available somewhere else, it's no help in expanding production of a specific chip. Older lines often produce many different chips, so production scheduling becomes a delicate balancing act, trying to satisfy ongoing demand without gaps in the supply and to minimize the frequency of retooling. In addition, while the main issue is silicon, there are material and capacity shortages in a lot of areas, affecting current semi production. At the moment, lead times are crazy and often previously confirmed orders are getting partially or fully "deallocated", just before delivery. Then of course, greed kicks in. The worst I come across was a supplier trying to charge 210(!) times the price of the normal supply price of a chip. Counterfeiting is also rife, with buyers forced to do extreme measures like x-raying chips before accepting them. Leadtimes? I've seen 99 weeks. Increasingly, manufacturers just throw their hands up and simply EOL part of their product line, causing mayhem for equipment builders. Some companies producing and selling hardware with thin margains can often make more money reselling the surplus parts they actually have (maybe becuase of over-ordering, maybe because of missing a crucial part to complete the BoM) than what they make from building new gear which may be at negative margains, due to high component prices. Of course, selling surplus stock is just a sugar hit and the reduced production may kill their business in the long run. ...but I'm sure you've all heard similar stories! ☹
The number of designs i had to convert from one MCU to another (and various other parts) boggles the mind often the same design many times for any quantity of roughly equivalent processors we can get
It's not only you. The largest (or second large, depending on the year) car manufacturer had to work around a network switch in a device that will sell in millions.
There are a *lot* more than 200 fabs in the world. I visited one in the smallish town where I live, and it isn't even the only one there. Most fabs make very low density discrete semiconductors, eg transistors, particularly power transistors (big & simple but also highly precises). Op-amps and 555 timers would also fall into this category.
Great topic but I feel you missed out another big factor that is influencing the industry at the moment. We are a small electronic production company that build some products that ship worldwide. So we see these issues on day to day basis. At this moment brokers are making millions by buying in large amounts of chips. Even new batches that come right from the production floor. Just to sell it at 4-10 times the normal price. They buying all stock world wide to create a shortage too so they can sell at a high price. For example we got a notice of chip back in stock from a manufacturer. Batch of 10k chips. Direct rushed in to buy 100 chips to keep production going. But the full batch was already gone. ( within seconds ) A few days later a batch of 9600 chips are showing up at a broker inventory for 8x the going rate of the manufacturer. Also almost all 3,3v regulators that are pretty common to use for all small logic. Are almost all gone in the world. But huge deposits are in the broker hands trying to sell them for 10-14 dollars a piece that cost normally around 1.5-2 dollars. Luckily a few chip manufacturers are now putting in the 100 chips per company rule. Meaning you can buy only 100 pieces for a set periode. With a batch size of 50k chips this gives a bit more breath room that there will be stock without brokers buying it all. As it seams that putting money into chips inventory might be the best investment at the moment as the interest and payout seems to be much higher then invest in housing market at the moment. Check a few chips at octopart . Com and see how many parts are only available trough brokers 😉
Tokyo electron is starting to make new tools of their previous generations etching tools due to customers demanding them. It's been over 10 years since a new one came off the line, but they're coming back
Just try buying a 2000V 2000A optically fired thyristor if you want to try long lead. The 300mm chip diameter has a 1% manufacturing yield… 18 month delivery time precrunch…
Preface: I've worked 25 years for semi companies, Microchip the most recent (out of 40+ years designing products). This video, and all the other ones I've seen, "blame it on the fab capacity" It's not the fab capacity. It's packaging capacity. Testing capacity. Mask costs (especially 28nm and 22nm. At
I want to thank you for an excellent summary of a much neglected topic. I retired last year from Microchip, working mostly on Reliability / Quality issues with mature EEPROMs for Auto and high Reliability customers. When you have to get long term performance out of non-Volatle memories even 250 or 180nm nodes can be touchy. Steve Sangi did a great job building up a 'trailing edge' business. Predicting the overall industry can be pretty difficult though. When I was at IBM Microelectronics I recall a senior engineer commenting (in 1991) that in ten years 2 or 3 Fabs would support all the chip needs of the workld's computers (ie, no phones, smart systems, etc...).
Essentially all STM32 parts are completely out of stock around the world. There are a few places with stocks of micros with factory price of ~$15, selling for $60-100 per part.
I don't see that. LCSC/JLCPCB assemly has thousands of different STM32s in stock. For my current design I'm choosing from three different low power parts capable of driving segment LCD.
I am glad that as a hobbyist I tend to buy components in much bigger bulk quantities than what I actually need. Some of the voltage regulators I bought back in 2019 are still being put into new hobby designs today since I can't find anything equivalent today! This is a great reminder to engineers (both hardware and software) that we should be focused on making our designs as modular as possible, so that you can easily swap out voltage regulators or even entire SoC's. PS: Wow, the ESP32 is far more powerful, much cheaper, and more available than the ATSAMD21G18A chip that you covered in the intro to this video. Then again, companies probably find it cheaper to spend exorbitant amounts on chips like this than to completely re-design the microcontrollers and firmware for their controller boards.
It's not redesigning that is the problem, it's the supplier. A company like ST meets all norms, while tensilica, the maker of ESP32 does not - because they intend to produce a hobbyists tool.
@@LMB222 yeah, I did not mean to imply the ESP32 could replace that chip for many use cases especially industrial/professional ones. But it is interesting just how much more powerful AND cheaper it is
@@LMB222 It's not that ESP's chips aren't used at scale, a lot of iot power products like colour changing lightbulbs and remote switchable power sockets used them at scale; then again these are products with... shall we say no particular expectation of reliability, often just borderline compliance to basic safety norms, and many of those have even since switched to even cheaper and wonkier chips, they plain don't care. You simply can't sustain a semiconductor company like that on hobbyists, but you can sustain it on kilotons of future e-waste.
I’ve been wondering about the supply chain of microcontrollers for a while - it doesn’t make sense that car parts would use cutting edge processors In my experience, embedded systems engineers have gotten used to overspecifying chips vs optimising code better - this can and should change
That has nothing to do with the shortage. Many micro processors are in short supply. Squeezing your code into a smaller chip isn’t going to help this problem.
Automotive chip needs to comply with automotive standard(safety, stability). That mean they go through more test process than other commercial chip. lead time can be up to 2x compare to ordinary chip.
All of this talk is interesting, but kinda missing the point anyway. Most consumers do not want or need a car with computer chips anyway. Car COMPANIES would rather put chips in their products to enable things like tracking of customer's behaviour, and to justify a higher price tag, but this is just extra expense with very little utility to the end user.
@@andreaslind6338 then just buy yourself a 70's car. Give up the fuel economy and the horse power. Give up the 6 gear transmission that adapts to the road. Give up ABS and traction control. Go back to cars that are dead at 100k miles. Enjoy exhaust pollution that was bad enough that people k*lled themselves with it. These things are there because people want them.
1. An advantage to using trailing-edge technology is that it's easier to find substitutes. As you say, these are commodity parts with multiple suppliers, so you could perhaps find a software-compatible part with a different package or extra features. Compare that with finding a substitute for a bleeding-edge Nvidia GPU. 2. You may not need to construct new buildings to convert a fab from 200 mm to 300 mm wafers, but you do have to acquire all new equipment, as 300 mm wafers won't fit in 200 mm equipment. There are half-measures that can help, like re-optimizing the current processing procedures, jamming in more 200 mm equipment in places where it would help, and eliminating special-case processing steps as much as possible. These are some of the work-arounds used in the initial 200 mm to 300 mm transition in the 1990s.
I wish i had a sliver of knowledge of the info of your videos!!! I do enjoy them and appreciate the work, explanation and work you have to do to get to my level of understanding! Blows mt mind! THANKS for all your work and ROCK ON!!!
The SAM is of course an inherited Atmel chip. Indeed all Atmel, Microchip, NXP, Nuvoton, STM low-end micros are sold out a year in advance and have been for a year and something, everything under $3, most things under $7 as well. And it's been a lot tougher to work around than shortages of higher end chips, which are still kinda available. Also special function analogue/linear semiconductors of particular kinds are hit hard. Chinese domestic chips have better availability. Maybe more 45-28nm utilisation? I do think maxing 28nm and node shrinking designs needs to be an intermediate goal, i feel, not adding another 90nm fab. But i think more 28nm fabs won't go to waste.
If the lead times surpasses the design time heavy disruptions is the design ecosystem emerge. Before no one consider to porting a 2$ chip. If you already port why not give up the old silo system of compiler + squeezed old fab + broking distribution
Not all, there are several STM32L stocked for low power with LCD control for example- And then there's the RP2040, $1 chip with $1 flash+accessories, good performance and excellent support. People should be using it more.
Atsam is a 32bit chip and will migrate easily to risc-v chips which will be made on old equipment in China for pennies. The trailing edge chipmakers understand these risks and are willing to let the world burn to avoid getting crushed.
@@jan.tichavsky rp2040 has external firmware and no provision for ROM encryption, which makes it nominally unsuitable for most commercial products. And open source projects often get fouled by its glitchy ADC, which is not entirely surprising of a company with no such design experience, they'll get it right eventually.
So short term speculation lead to short term decisions which impacted long term supply. Who would have thought of that?? This is what happens when single minded immediate profit companies thinking about the next quarter make rash unfounded decisions on short term speculation.
There is a fortune to be made for the person who sets up a proper e-recycling shop that focuses on chip reclamation with efficient acquisition, processing, and quality control. There is no shortage of e-waste and it's current use is messy at best.
Heh, china already does that - where do you think all the "counterfit" parts on ebay, aliexpress, and amazon come from? They're all re-marked recycled chips
Ok for analog or simple CMOS ICs but not for MCUs or low-power processors because you also need the software and tools to reprogram them. Despite many retroengineering projects, it's quite usual to not be able to repurpose the 90's and 00's generations. If there is a way, devs often occur challenges like assembler-only compiler, old C or C++ standard, DOS program, programmer with parrallel port interface or (in the case of processors) supported only by NetBSD.
For the people going "why retool not build new when the building is the cheap part" your technically correct, but lacking context. A building might only be a couple of million of a several hundred million dollar line, but the earthworks, foundations, and fitout of the shell itself make up roughly half the total time from design to commissioning, and the vast majority of the external requirements regarding permits and legislation. And thats assuming you have the space, power, water, workers, administrative capacity to not only run 2 lines, but run one line while the others being constructed, which can entail far greater disruption than "retooling" even when it means essentially complete replacement of the actual machines in situ.
Lets you know how fragile the economy really is, and how just one bad decision can totally wreck it in many ways. The Economy isn't something that can be managed, it's literally billions of people making billions of individual decisions every single day, you can not predict what is going to happen. It's next to impossible. Business assumed the economy would slow down dramatically, but consumers kept consuming.
"...one bad decision can totally wreck..." That is exactly the operational strategy of LIM (leftist invasive management)... IE; the idea that it takes only a relative handful of people to destroy what it took billions of people create. Now coming to a capitalist based societly near you.
chip shortage should be a lesson to engineers to make their designs cpu architecture agnostic. i started doing it with my designs as standard practice around 10 years ago. little did i know how it would pay off later.
CPU or MCU? The time spent on the latter is getting all the peripherals to work, and those are not agnostic. Then an even more amount of time is clearing up the corner cases of power-on reset, oscillator start-up, UART timing, all over temperature. Those issues are typically specific to a signal part family.
for MCUs i've started writing my firmware in rust since the community made a lot of efforts to create hardware abstractions for a lot of middle-end MCUs. but that barely solves the problem of acquiring said chips since they are either sold out or have dramatically increased prices. I for example replaced the STM32F103xx with a GD32VF103xx since they got an identical pinout (infact it was meant as a drop-in replacement) but it required changing the whole compiler infrastructure since the GD32VF103 is a RISC-V processor and the STM32F103 is ARM. And even though the GD32VF103 seems to be relatively unknown it was still a nightmare finding a reseller that actually got stock of it.
@@minespeed2009 ha ha, all that HAL is what scares me; more abstraction means more bugs, and it’s tougher to find them. I’m not even an embedded guy (rather RF Microwave) but I won’t touch anything but bare C when I need to implement an MCU. Don’t get me started on using Circuit Python, Arduino, etc for shipping products. IMHO it’s a nightmare waiting to happen. As I mentioned above it seems 90% of the effort is spent fixing 10% of the corner case stuff. I actually was messing around with an Arduino compatible MCU module, and it’s got a known bug of not coming out of a low-power sleep when the UART is low, of course it’s beyond me to fix it through layers of software abstraction. But it’s this kind of shit that just drives me nuts.
@@rfengr00 it definitely makes it a bit harder for some cases, but in the end it takes me about as much time to implement something as in c since the rust compiler is VERY strict when it comes to unsafe programming. like refusing to compile when you access the same peripherals in two different code sections without transferring ownership. (ownership checks are done at compile time so no binary size or speed penality). But im also only an EE student and therefore haven't worked on larger commercial products.
So you’re saying engineers should strive to do something that’s literally impossible? You have PC boards will you can just swap in any brand of CPU you want? And you take the time to make low level drivers for each one? Tell me more?
Has been for the last 62 years and 5 months… it’s the very definition of the electronics industry, since the inception of electronics … Nobody is buying worse-slower stuff or the same stuff they bought last year for that matter… Unless that someone is the government…
@@robertw1871 I'd agree with you from the customers perspective...from the perspective of someone who had to maintain legacy designs that kept selling for decades, not so much.
@@gregoryhall9276 Yep that’s our nemesis… 17 years in medical diagnostics engineering… thankfully I’ve been on the bleeding edge since that nightmare… Not sure it’s more fun spinning a product that’s not even released to production and already 6 months obsolete… Never a dull moment though, sales weasels selling things and taking orders for things that are physically impossible to make…
@@rockapartie That’s why almost no industry is anywhere near the bleeding edge, well except For SpaceX, Tesla, Alphabet and Amazon’s ever expanding giant empire…. It’s also why the dinosaurs are going extinct, they don’t adapt and costs for maintenance on ancient technology that doesn’t perform well enough to be competitive takes them out… It’s why those few companies that are able to adapt run nearly everything already…
@@rockapartie Farm tractors are mostly software driven these days and very high tech, pretty sure they get updated fairly often… Manufacturing and Logistics are where the very rapid change is actually driven from… Medical hasn’t changed much in 30 years though, mostly because Medical companies are owned by holdings companies that are only interested in milking what’s there not investing in the future….
Well Renesas mcu are in very heavy short supply. They have caused lots of damage to Japan's car industry in the last 2 years. They may have single-handedly reduced Japan's car output by 500,000+ units over last 2 years. Renesas are very slow to invest even though they are making big profit. Earthquake and fires also don't help there Fabs.
Why did the US give up so much of it's fabs? I sorta remember when AMD did it. I think they claimed they would save money and could focus on chip design. Intel kept theirs till recently? I kinda agree with the people saying giving up so much manufacturing and industrial capabilities to make the rich richer and maybe tie everyone together so supposedly war is impossible was a mistake.
It's due to globalization. If you have a chip factory in the USA but the factory making phones or laptops is in Korea you have a problem. The solution is to raise tariffs on all imported electronics and force local production in the USA. When China goes to war, trade from Korea and Japan will probably be lost so the USA needs to be self sufficient in terms of electronics and manufacturing in general.
@@drscopeify have you considered how large the tariffs would have to be to make it more cost efficient to have fabs in the USA instead of outsourcing it to Asia? any basic economics course would teach you that these tariffs would create significant damage to trade balances. Also, outsourcing the fabs to asia is a way for these corporations to mitigate the seasonality risks that comes with the industry. The idea that a country should stride for self-sufficiency due to risks of war in the future completely trashes the idea of competitive advantage and the reason why countries like the US can be so wealthy in the first place.
@@KrDavidLee Yes you need very very large tariffs and the USA will suffer massive inflation but this is going to happen anyway if China goes to war and it is not investing trillions in a modern army to look pretty I don't think... China isn't building aircraft carriers to just pose for photos. So why not force local manufacturing in the USA today rather than wait for China to start eating up all of Asia and then being left with the pants down. Imagine inflation in the USA without exports from Japan, Taiwan, China, South Korea....
This reminds me of a project I have been working on every once in a while. It is an ASCII video converter and player, should really get back to working on that soon :D
We have the exact problem right now. Parts that used to cost 6 Eur are out of stock for up to 92 Weeks. And broker offer them for 400-500 Eur right now. Its just insane.
@@robertadsett5273 I have a board or two that God would have started working on already, if he had been listening to some choice words I had about them.
It's not like they can just start chopping down more trees. Sorry, I had to. My son and I just watched The Lorax for the 100th time and I still love it.
Brilliant channel I've learned a lot thanks. You'll gain a patron after xmas. Thanks for explaining the business layers of chip manufacturing. I have been caught out buying PICs. The chips that I prefer only need a K or so and they're over a year lead times. Thanks Deer 🙂 NEAL
Just like the demographics and climate problems, the pain hits long after the deed was done. Notice how a lot of this shortage is military industry related.
I wonder if this is where we see some kind of disruptive innovation. FABs, to me, are identical to integrated steel plants, which were disrupted over decades by minimills using a more efficient production model. I wonder if in a similar way, the lowest-grade silicon will see some new method, and slowly this method will move up the quality chain?
Chips can be like nuts and bolts. A match box car needs a wire for an axle, but a mining truck needs....... I know that supply chain disruption has been critical and legacy equipment give fabs an advantage especially in niches.
Sony's old fabrication plants in Japan proved that (at least with entertainment videogame system processors) they could take the early 1999 finalized silicon engineering sample Emotion Engine CPU and Graphics Synthesizer graphics chipset from (I forget at the moment) whatever the manufacturing process node was in early 1999 or late 1998... because the prototype PlayStation 2 engineering sample motherboard was demonstrated by Ken Kutaragi in March 1999 at a special Sony industry presentation not really made for consumers iirc but was still reported by the videogame magazines of the time back when internet news was just not organized or reliable. Anyway Sony went from whatever that initial process node was in early March 1999 to the mass production process node for launch in March 2000 (because the Sony press release said the die shrink would allow for higher yields, lower wattage, lower thermals... stuff that was common before Microsoft Xbox changed that in 2001 and later Sony followed suit at great cost and risk to them) all the way to the 2006 Sony PlayStation 3 launch which featured iirc a 90nm die package of the EE and GS processors which used a heat spreader. However due to the difficulties of that time and customer fears and confusion as well as some bad marketing campaigns on both Pro and con sides... the PS2 chips were removed from the PS3 yet PS2 chip production continued and made some major improvements to 65nm production node by 2008 to 2009 which included the expensive RDRAM memory chips inside the die package and basically no need for a heat spreader and definitely much lower wattage consumption and thermals. So for nearly ten years of the PS2 CPU and graphics processor production, Sony definitely and evidently proceeded to provide historical evidence that processor die shrinks are indeed possible if a company wants to do them. Now the EE and GS processors were proprietary to Sony and as such they had more engineering control. With the PS3's CellBE CPU and the Nv47 based RSX GPU, both of which had to be reactionary produced at 90nm in 2006 as opposed to 65nm which for Cell CPU was Sony's original target node... both of those processors continued to be fabbed until die shrinks made possible around 2011 to 2012 to either 40nm or possibly maybe 28nm. Then all that changed and had to be dropped because Sony had been sustaining losses and Microsoft had called for next generation hardware around 2011 and we ended up with AMD becoming the designer of the CPU+GPU package and TSMC being the only fabrication plant to provide processors on a 28nm process node for 2013 mass production. I don't believe Sony has much control like they had when fabbed their processors at their proprietary fab plants so I can't recall what die shrink the PS4 Slim reached around 2020 to 2021 but this is just an example of what is and was possible. However TSMC pioneered and reached the cutting edge production nodes... YEARS AGO... bearing in mind the harsh reality of the year 2020 problems the least we could expect from other chip manufacturing plants is to have improved their processes. As pointed out in this video, a 90nm fabrication plant may be limited to 90nm production unless there was some way possible to replace and reformat the manufacturing tools inside the plant which sounds easy but is just insanely expensive in reality. Still Intel owns a vast array of fabrication plants... if they wanted to or if it was plausible, they could provide services to AMD and Nvidia for GPU fabrication nodes... however iirc Intel does not have the same technology as TSMC for that aspect because their main products are CPUs Outside of that, fabrication plants that reached 28nm should in theory be able to handle die shrinks for many processors to provide increased yields in theory... however there is a cost there and maybe some manufacturers don't see the point in paying for it if their mass production processors are already being fabbed at 130nm, 90nm, 150nm, etc. Therefore if Intel wanted to produce Itanium CPUs on 28nm and lower as an example or even have AMD make 28nm Phenom II Hexacore CPUs... all those things are possible but those are not things that their industry are about... thus other processors for industrial use and car electronics are a different industry but are still affected by the year 2020 problems.
I wonder how long a lead time for a die shrink is. It seems likely to me that the PS2 logic was designed in parallel for the stable node and the future cutting edge node, but of course the future node couldn't be manufactured... until the future came about. It was likely too expensive to make systems on a node which would be out of date by release - beefier PSU needed, more silicon consumed, reliability compromise due to higher temperature, all that. And it was likely too expensive to wait for new process to make early devkits - the cost of opportunity is the highest possible cost.
I mean intel's whole "tick tock" thing was one year they do a new design in node X, then the next year shrink that design to new node Y, then do the next new design in node Y, and shrink it to node Z... all until they couldn't keep that cadence up anymore. It takes a lot of people and resources to do that - probably something the companies that make microcontrollers don't have, and don't have a benefit to do. Microchip isn't trying to increase the clock speed and reduce the power consumption of the same chip from year to year - once they do a design, they just want to keep cranking out the same thing forever without any further engineering / design cost.
Microprocessor Report did an in-depth examination of the PS2 chip set. The intent was always to do mass production at the 180nm node but this was a bit ambitious as Intel, then the global leader in process node advancement, was still ramping up for 180nm product. The 250nm version of the PS2 chip set had been intended solely for engineering samples and early developer kits. Due to delays in getting the 180nm line ready, the initial batch of PS2s for the Japanese launch used the 250nm chips, which made for a lot of red ink to keep their schedule. Another oddity in those early units was rear bay for adding a hard drive used the PCMCIA Type III form factor, mainly seen in laptops, rather the common 3.5" form factor used widely in desktop PCs. Sony offered owners of those early units and adapter to plug into the slot and use a 3.5" drive externally. The PS4 APU, along with the very similar Xbox One APU, had just the one die shrink, to 16nm. This enabled the PS4 Slim and Xbox One S models to be produce and replace the original models. The additional ceiling also allowed the creation of the PS4 Pro and Xbox One X. That was also where companies hit a wall on the value of die shrinks as opposed to new designs. A casual observer of the game console business would reasonably expect at least another die shrink, with perhaps the Pro and One X becoming the mainstream models as the generation wound down and a new generation launched. But that would be ignoring how much the economics had changed at the newer process nodes. This was one of major reasons Microsoft made the Series S. It made more sense to have a low end version of the latest architecture than continuing with the existing product in that role, as another die shrink would not reduce cost as well as it would have in past node generations..
@@SianaGearz That is correct. The launch PS2 was always intended to be mass produced at 180nm, while the 250nm version of the chip set was produced for engineering and early dev kits. The first production run for the Japanese launch had to use the 250nm chips due to delays on the 180nm line. This was a painful expense for Sony but they kept the launch schedule and had a very successful product in the long term.
Lukashenko in Belarus two years after lockdowns helped cause the current economic crisis: "Don't you guys all feel like Covtards?" PS in Russia we still a use a lot of Soviet-era processors since our microelectronics collapsed in the 90s and never really recovered. For example, a 32-bit K-1839 processor created in 1989 STILL flies aboard the GLONASS satellites.
I really hope all that extra capacity hits the market right when demand slows down. Would lead to very cheap hardware which could lead to new innovations.
great video. One thing I would take a bit of issue with is die shrinking old designs is not "always" a good thing. Less die size does mean cheaper and less on die parasitics, which means faster ICs. This should be a "win/win" as who doesn't like faster ICs? As it happens, the FCC may not. Sometimes a die shrink will make a processor too fast for it's own good, and where a design used to pass radiated emissions in a larger package, it will not in the smaller one.
No, a die shrink NEVER makes an IC too fast. An IC is a clocked circuit. The clocks that get stuck into the IC determine the speed. No that's not the issue. The issue is a node can be too small for the number of transistors needed to go into a single die (IC), so you end up with a TINY die that has to be cut from the wafer and cutting a die ever smaller I'm sure becomes very hard. For instance TSMC's N7 (7nm) has a transistor density of about 90 MILLION transistors/ sq. mm , N5 is about 170 million/sq. mm and N3 is almost 300 million/sq. mm. So, at the low end of these advanced nodes, 90 million transistors/sq. mm So you design a circuit that has a few hundred transistors. You still need it to be an IC, but you can't put it on N7 because you couldn't cut the die small enough. The other issue is power. Internally I can have transistors running at tiny voltages or current, but to drive signals off the IC to another circuit requires more power. You can end up with having to make a larger circuit on a small node to drive signals and this can be a waste. For instance AMD is making Zen 4 with two different TSMC nodes. They use TSMC N5 for cores and use N6 for the I/O functions to drive signals out of the CPU or into the CPU. Some controllers can be in the cores but most are in the I/O die. For Zen 2 and Zen 3 AMD used GloFo 12nm for the I/O die and TSMC N7 for the cores. Part of that though was having to fulfill contractual agreements with GloFo, for which I think AMD is finally done with them. So THEY could have capacity for different chipmakers as AMD finishes off with those contracts, but Zen 3 is still going to be made for a long time. So those are the reasons smaller nodes can be no good. You have a circuit that doesn't have enough transistors in it to warrant moving to something below 50 - 28nm, and that's probably the case with most, and then you have the issue with having to drive signals off the IC and need more power to push that signal. The other things aren't a problem. Clock speeds are controlled by the chipmaker. Other issues I imagine could be more sensitive to noise, like you're in an industrial area and machines are producing a bit of magnetic waves or strong frequency transmissions and smaller nodes are more susceptible to external noise. One thing that node shrinks are supposed to do is drop power consumption by the IC. You shrink a transistor and you reduce the amount of power needed to drive it.
@@johndoh5182 You are correct about clock speed. However noise is generated by signal transition rise/fall time. If you decrease the rise/fall time, while maintaining the same clock speed, more RF noise will be generated. This is in general why, when designing a product, it's a good idea to select a technology that customarily runs a the speed you intend to run it, and not a technology that's intended for 10x faster, for example. Also why MCUs and SoCs have selectable output current levels on their IO pins -- so you can limit the rise/fall rates.
@@Graham_Wideman I'm going to say BS because it's the same process nodes that run in laptops where the clock frequencies can drop to VERY low levels to save energy. Link something that shows Samsung or TSMC has an issue with slow clock speeds on N7 - N3 and I'll believe it. You get issues with noise typically with HIGHER frequency because current increases. Maybe you're talking about something external to the IC. In fact CPUs are constantly changing frequencies in cores and they can drop to very low levels, like a regular CPU that goes into a PC. It's how a CPU saves power. But I do believe that in laptops the clock speeds drop lower than in a PC. But I also did say in that comment that smaller nodes can be more susceptable to external noise. But, that's not like a tiny bit of noise. After all there's a lot of magnetism going on in a PC with power supplies that can push 1600W along with having banks of mechanical drives. Fan motors. So link something or give search info and tell me the name of the web page so I can read it. Maybe you're thinking that there is some general rule about what happens as nodes shrink. And I'm going to say engineers solve those problems which is why you HAVE those smaller nodes. I means there's a LOT of Samsung die that doesn't run at high speed because it doesn't need to. They have to clock a lot of ICs slower to save battery power.
Let not forget that the tooling need in the fabs also use microchips and the microchip shortage is hindering supply of the material for that tooling. So, even if the fabs are built, the fabs won't have the tools to put into those fabs. Generators, power supplies, voltage regulators, etc. are all needed to build the tooling needed in a fab, and those need microchips, which, there isn't any.
The cost problems that the executives are talking about in those docs could be eliminated with a simple solution: eliminate the executives, and redirect their grossly inflated salaries into manufacturing.
Yeah, no. That’s orders of magnitude less money than what is required here, you’re talking about a few people making half a million a year vs a billion dollars per quarter to rapidly scale upwards like that. I’m sorry but the real world isn’t so easily dissected and solved, and those figureheads play a very important role for the structure of successful businesses.
In classical economics there is never any shortage or scarcity of any commodity, because price mediates between supply and demand so that they match each other. For example, if chip buyers want to buy a hundred million units of a chip, and the manufacturers are only able to make fifty million units, the price of the chip should rise to a level where the buyers only want to buy the fifty million units that the manufacturers can make. And, if a chip manufacturer runs out of its product, that means that they have simply set the price too low.
We may live in the Sonoran desert her in Phoenix but... there's a reason the metro valley is reaching a population of 5mil. We know how to handle water. The new FAB plant is being built a couple miles north of the Central Arizona Project. The CAP is a canal that runs for over 150miles from the Colorado River by Parker AZ, down to Phoenix and then Tucson. It's primarily use is agricultural. Less than 15 miles to the north west of the FAB is Lake Pleasent. That's a reservoir of over 1 mil acre feet of water. No river runs freely in Arizona. We have multiple reservoirs in the mountains where we collect winter snow runoff to supply drinking water. Much like California, only we do it better without all the Democrat corruption and mismanagement. I've been here over 40 years, watch the Valley quadruple and we've never had to ration water. So... pretty sure the FAB will be just fine.
Maybe there is mismanagement and corruption nearly everywhere, and rather than enabling it by doing the corrupt politicians' campaigning for them we can pull our heads from our tribalistic asses & acknowledge that fact? Or I suppose we could not do that and let ourselves fall deeply under their sway, to the point we can't talk about things as basic as water without signaling our loyalty by shoehorning their talking points verbatim into every comment conversation and quip? Nothing screams individualism like millions of ppl using another person's words to express another person's ideas every time they open their mouths. Nothing exudes freedom like feeling it necessary to constantly reassure others that you're one of them. But go on, tell me all about how you came to believe what you believe of your own volition, putting 2 and 2 together after you got a hankering to audit the financing of water works in a small town in California. How having such a blast on that audit lead you to do the same for the next town over, and the next then the next until before you knew it 2 and a half years had passed and you'd performed an audit of every municipality in the state.
Also, might want to look at the trends re; the flow of that river over those 40 years. Definitely take an gander at the rate of replenishment vs usage. Might also be a good idea to brush up on the demographic and economic changes up stream and maybe also the state of water rights and what outcomes that might incentivize. Oh, and take a look at the situation down stream too, in Mexico. Or believe whatever makes you feel secure and comfortable regardless. Can't see that ever becoming an issue. Never has before.
Water Wars Next stage of water restrictions in Arizona could come as soon as August Officials with Central Arizona Project and the Dept. of Water Resources warn the state needs to do more to conserve water to help stave off even more restrictions. Author: William Pitts Published: 6:17 PM MST May 6, 2022 Updated: 7:40 AM MST May 10, 2022 ARIZONA, USA - Arizona water officials now predict the West will enter the next stage of drought restrictions as soon as August. A Tier 2 shortage declaration would mean even more cuts to the amount of water Arizona gets from the Colorado River, but not nearly as much as the Tier 1 shortage declared in 2021. Still, officials with the Central Arizona Project and the Department of Water Resources warn the state needs to do more to conserve water to help stave off even more restrictive cuts. “Most of the water use is outside the home,” CAP Deputy Director Ted Cooke said. “It’s in the yard, it’s pools, it’s plants, it’s lawns.” Cooke said there’s a lot of room for conservation just by having people cut their home water use. The Federal Bureau of Reclamation declares water shortages, which trigger water cuts based on previously approved agreements. In 2021, the Bureau of Reclamation declared its first-ever water shortage, cutting more than 500,000 acre-feet of water going to Arizona. An acre-foot of water would generally supply three average Phoenix households with water for a year. A Tier 2 shortage, if it’s declared, cuts an additional 80,000 acre-feet of water to Arizona. “How that’s going to impact the homeowner is…they’re probably not going to see much of an impact at all,” Arizona Department of Water Resources director Tom Buschatzke said. Lake Mead and Lake Powell are at record low levels, thanks to a 22-year-long drought in the West. RELATED: Lake Mead water levels hit lowest point in 51 years The Bureau of Reclamation announced this week that it would keep another 500,000 acre-feet of water in Lake Powell just to ensure that Glen Canyon Dam can stay functional.
I can barely grasp pound per feet to express torque, instead of the far more practical Newton per meter units. But acre-feet makes it even harder to understand 😅
Manufacturer: Oh man, we really need to order some more of these chips. Chipmaker: We don't have any of those in stock. Manufacturer: But we need them for our products. Chipmaker: This chip is old tech, that's getting more expensive. Manufacturer: How much is it going to cost to make them? Chipmaker: Nah, you don't want to pay that much. Manufacturer: If the choice is between selling fewer products because we had to raise the price, or selling no products, we'll take the more expensive chips. Just make the damn chips and we'll buy them! Chipmaker: Ehhhhhhh we don't wanna make these old chips. It'd take a lot of work and we'd have to raise the price. How about we just don't sell you those chips, and we'll make less money? Manufacturer: _*#%& @^#% #$&)!_
Note that more modern machines have also way higher throughput. An older ASML 200mm machine can handle 100 wafers per minute, while a newer 300mm can handle more than 300 per minute. So, that 6 times more dies per minute. Although the floor space is larger. Finally on 300mm you can better utilise the edges than on 200mm.
So I am from the Netherlands and what I don't understand is that if ASML is making a large portion of the world's leading chip making machines, why aren't we making those chips ourselves? What does Taiwan have that we don't have that make them the largest producers of chips?
during the 2020 pandemic it was not that paper products were necessarily in short supply, it was more of an issue of hoarding. A bulk of the chip "shortages" in 2021 and 2022 are caused by the CCP hoarding semiconductors...The CCP builds with impunity.
That has negligible effect on legacy chips, where the real shortage is. It will lessen the pressure on the cutting edge chips though, but it will not decrease actual manufacturing volume of these chips, just more of these products will be available to regular consumers. To that end, it will not affect the shortage as an industry wide problem affecting all dependent industries at all, or rather it will, but only after those lucrative markets are saturated, so years later!
There is a severe water shortage in Arizona right now that is leading to water rationing for human use. The Colorado river, the main water supply for Arizona and the Southwest, is at historic low levels. If water levels drop any further, the hydroelectric generating capacity at the Hoover Dam will have to cut back production to maintain water levels. Why on Earth are FABS that EACH require 9M gallons of fresh water daily being built in Arizona?!? Only about 60% of that water used by the FAB is recyclable. Building multiple FABs in a desert during a drought is just plain asinine. The people of Arizona are not like the people of Taiwan and will not tolerate severe water rationing just so a chip FAB can operate. I cannot fathom this abject stupidity.
Yup, both Intel and TSMC are building many fabs which needs the Colorado river. Republicans are just straight stupid anything for money forget the people and the environment.
@Unknown Alien There is not enough electricity to run desalinization plants in the American southwest and the ultra high cost of electricity in California makes desalinization uneconomical. Nuclear electricity generating plants are being shut down and replaced with solar panels and windmills which don’t work at night or if there is no wind. If FABs are to be built, they should be built where there is an abundant fresh water supply (e.g. Pacific Northwest or Great Lakes areas). Building FABs in a desert during a drought is absurd in the extreme, courts ruinous financial failure, and is just plain stupid.
Both Intel and TSMC are building fabs there AND they are going to be LARGE fabs. In the case of TSMC they bought up the space to create a giga-fab, with 6 smaller fabs in a single location, just like they have in Taiwan in a few places. Closed loop. The water gets recycled and used again. I read about what it is they're doing because I said the same thing to myself, which is that's about the most insane place to build ANY manufacturing that requires a lot of water, because that area is running out and that's not going to change. I don't have a good answer for why this is and I'm not going to search for an answer but companies seem to like to build fabs where it's hot and dry. Maybe it's about what can survive in the air such as viruses and bacteria, I don't know.
@@jstasiak2262 They are going to recycle the water so they HAVE to be running some type of filtration. Go read about these plants that are being built. As far as electricity, if stupid ### Biden would lift tariffs on ALL solar panels coming into the US it would be dirt cheap for a fab in AZ where it's sunny 95% of the daytime hours to have all the electricity they need, INCLUDING the cost of grid storage. The company ESS produces iron redux flow batteries that use water (closed loop) and iron as the main ingredients and would be very suitable for this situation. For large companies that are buying thousands of panels, the cost of panels is very small to the point that it out-competes natural gas. It's an upfront investment though. A company like Intel or TSMC will have the ability to pay that upfront cost, just like what Tesla does with their giga-factories.
The fab capacity that was freed by the Automotive industry cancelling most their semiconductor pipeline was not re-tooled, but instead used for MEMS and mixed signal chips?
I think the most used ancient chip might actually be a TL431. Obviously i'm not the first to voice such an opinion, but i have seen maybe two 555s in a product so far in my hands, but i have had dozens or low hundreds of TL431s in my hands.
The TL431, introduced around 1977, is certainly used in some current designs. But the LM555 predates thatm arriving in 1972. Perhaps modern-day designs would not use the exact same model, mostly adopting SMD format. As for being the most ancient still used in new designs, I'll suggest as candidates some 4000-series CMOS logic chips introduced around 1968. Around the mid 70's, there was a transition from the "A" series to the "B" suffixes, (I believe the B series added a buffer stage to the output, and possibly better ESD protection). But again, these days you would choose SMD variants.
I think we need graphene copper boards and transistors . i dont know what introducing such an alloy would do to lithography but the lower resistance and higher overall conductivity will certainly help something. If not just power delivery and heat dissipation.
I'm glad you made a slide pointing out that TSMC is building a new fab in a region of the US that is RUNNING OUT of water. I don't understand why no one is sounding the alarm about this. Lake Powell (located in Northern Arizona) and Lake Meade (Arizona-Nevada border) are RUNNING OUT of water. Not only that, the electricity that Lake Powell generates is drying up as well. Add to that an increasing population that is escaping the nightmare that California has become, and you get a disaster. When did we become so insane?
I retired in 2015 after 42 years in Wafer fab manufacturing, I recall this statement by an economist (I think) about this industry, “I’ve never seen so many smart people working so hard, for so little”. I couldn’t agree more! It’s a damn cutthroat business, you’re always on call, no private life and the reward are “ok lah”. I was employed in a European, later American owned wafer fab but I heard even more horrific stories from the Asian owned one.
How much did u make per year
"Ok lah" recalling my time in Singapore and Malaysia working in OSAT industry, I agree.
People that step out of line..die...its good for you... that you were a boot licker.
The semiconductor industry experienced a demand fluctuation and spike that was outside of their operating tolerances, should have installed a nice demand capacitor tied to ground, the problem is the whole legacy system acts like one big inductor.
If only we all throught like electrical engineers
cute! our "inflation" is not going away. too bad the pol shysters decided on a broad racism strategy to localize production. they decide you know.
Wow that is so well written. Well done sir! Well done! I hope you will let me quote you on that, it’s truly very clever.
Some wafer technology doesn’t scale to larger wafers very well. Particularly non-silicon wafers that can only be produced in 150mm and 200mm sizes before inherent wafer defects create significant yield problems. 300mm equipment may double output, but it is often 5x+ regarding capital expenditure and maintenance cost.
Yup. I work in GaAs. Hard to make a Cz ingot when half the atoms in your melt sublime into vapor at atmospheric pressure.
Also, half the tools I work with on a daily basis are older than I am. And I’m over 40. Compound semiconductor makers almost exclusively inherited equipment from the 1970s-early 1990s silicon world because that’s all that’s available in many cases. What’s left are often 200mm tools down-converted to 150mm. With those kind of capital restrictions, an industry that charges cents per unit for its product and can fit 5,000-15,000 chips on one 6” wafer is never going to transition to 200 or 300mm.
Thanks for the perspective. Helps somebody coming from years of digital logic chips where increasing transistor counts have been the key essential.
Just to the last part: Any (not only multiples of five) "regarding capital expenditure and maintenance cost" is irrelevant when the net income(here the predicted value is sufficient) is above the previous one, the old production process provided. You just have to find an investor. That is no problem for such biggies like TSMC oder Siemens ... there are extra institutions there, invented just for that case which are happy to "assist", like banks and shareholders. But that is basic capitalism 101 and I don't want to bore you, Alexander:)
In a nutshell: If it's worth it, then anything will be implemented as quickly as possible. Lemme close with that picture: Haven't you ever seen a cartoon rabbit with dollar signs in its eyes?:)))
Don't get me wrong: Physical restrictions exist and no human, no Solar Roadways, no SpaceX and no Elon can avoid them. If @Grak70, for example, cannot even produce the raw material in good quality, then the discussion is actually superfluous. What doesn't work, doesn't work!
@@Grak70 how close are you guys to offering the coating of prescription lenses with the gallium arsenide. i really want night vision next pair .
One note. It's basically impossible to upgrade a 200mm fab to a 300mm one. Every single piece of machinery in there is designed for the specific wafer size so you'd have to replace everything. The only "Advantage" would be you wouldn't have to build a new building but honestly the empty building is the easy/cheap part. It would probably cost way more to take out all the 200mm tools than it would be to just build a new building next door for 300mm tools.
depends and also that statement is often false if we are talking about places outside of Asia where land doesn't sell for pennies and building permits are a huge thing. (Even in Asia it's sometimes false)
A fab is not even remotely cheap or easy. Even if you ignore the millions you'd have to spend for the land then permits and then a building, you also need a lot of time to build one, and in this industry every hour is basically millions lost that could have been earned. Besides that getting the whole facility upto a working level and have all the systems- from dust and air filtration to EMF protection to fire supression to ensuring water supply and whatnot is a major task
Anyone who has worked in datacenters knows that the most expensive part of a datacenter is the building itself- aka why so much focus is put on compact racks and fitting as much power in 1U and such. Similarly a fab definitely has more expensive equipment in there but the building is definitely not the cheap part- that'd probably be electricity
@@KryoNaut Even with land costs and how complicated the empty fab building is trust me the building is the easy/cheap part. I can't get into exacts but the single cheapest machine in a modern 300mm fab is easily several million dollars and takes months to install and get it working on wafers. Which goes back to my previous point it's almost always a better idea to build a new fab than to try and convert a 200mm to a 300mm. Unless you had a 200mm fab in downtown Manhattan or something.
@@Kevin-xw1eo Considering a photolithography machine for single digit/close to single digit nm fabrication processes is around 170-200million USD
something as large as 200nm with the whole thing should be into several millions for sure but not enough to call the building cheap part. You are just exaggerating at that point though I'd agree building a new 200nm fab is often cheaper than 300nm.
@@KryoNaut mm not nm. We're talking about the wafer diameter not transistor size.
Its not impossible Intel did it in Ireland with Fab 10/14/24
Respectfully, shutting down a 200mm fab to retool and upgrade to 300mm would be a grave mistake relative to building a new 300 mm clean room on the same fab site. Building a new fab would nearly always be economically advantageous especially if the 200mm fab is at max utilization
Yeah I agree. Those companies need to produce WHILE a refit would be happening? Not easy to do, so in general fabs get built and then it stays the same with few exceptions. TSMC can afford to retool fabs, such as taking a 5nm fab to 4 or 3nm because most the equipment is the same. One thing about the fabs being built in AZ by TSMC is, it has the space and will probably become a full fledged giga-fab like they have in Taiwan. It gives them the ability to have 6 fabs in one place. I'm guessing they'll expand that plant with N4 which is a retooled N5, and also N3 which is a new line. When TSMC comes out on N3 the demand is going to be CRAZY.
I hope Intel succeeds with their Intel 20A. If they do they'll have the most advanced node in the world at the time (late 2024 early 2025). That would take a lot of pressure off high performance nodes and then nodes like 12 - 7nm should drop in price.
Building a new fab for low margin products doesn’t seem feasible to me. I think this was also addressed in the video.
@@Conservator. per transistor cost bottoms out at 28nm, so maybe there's a way to regain some of the cost advantage vs. 130nm? It's just that there wasn't enough pressure to build more, not enough to offset the risks. You do have a problem that IO area probably won't shrinkify much at all though.
Correct!
You simply cannot shutdown an operating fab, even during a chip glut. The fab literally loses money during any shurdown.
@@cnordegren The fab would lose money during the shutdown period but that money could be regained in the period thereafter when they would be producing double the output at roughly the same costs.
Voltage regulators are also tough to buy. As an EE, I won’t cut a PCB design loose until reels of parts are sitting on my desk, and then I over-order which doesn’t help the shortages. I’m buying parts for life of product, knowing many of those parts will be eventually excessed.
Make sure the part you choose has many alternates from different manufacturers with the same footprint - jellybean parts like voltage regulators are the easiest ones to swap, even if it's a PIA mid production. A lot simpler than porting a bunch of firmware from one processor architecture to another.
Yup. I tend to finalize designs only once I find supplies for at least 2 alternatives. I seldom have the ability to source everything in advance of final design. Often this means extra pads for different pinouts or footprints. This also helps when sourcing parts for repair if your device is one that may see service.
I so FEEL you! Nowadays I first order enough chips to build out at least a couple of prototypes and then I start with the new design, otherwise it doesn't even make sense to even start (for the major chips at least) :P With voltage regulators I haven't had a problem tbh, but in terms of power electronics what I have had a problem sourcing is power inductors and transformers.
I understand your position and appreciate your frankness. By your own admission you’re making the problem worse. (That sounds more accusative than I mean to say, I’m not a native speaker)
How long is the life of product that you’re talking about? Im just curious if it’s something like a year or several years.
If you would stock for, let’s say, 6 months, wouldn’t that give you enough time to resupply? Again, I’m just curious and interested in your views as someone who works in the industry. Thanks for your comment.
@@Conservator. as of now, most microcontrollers have a lead time of over 12 months and it's changing all the time. I have an order for an ATSAMD51 that is already 1 year old, they increased the price twice during this time and I'm still waiting for them to deliver. Who knows what the lead time will be in 12 months, I think Louis Brown is right, better to have all the components in stock when launching a product.
Only a site like this would provide the 'missing information link' about what is actually going on in the industry and what is causing what to go forward or not going forward .. Bravo ! for such unique knowledge and smooth delivery that are so lacking in this world, when everyone is good at making noises, but so very few that can actually make sense ..
I worked in the industry about 25 years ago. While much has changed since then, some things have not.
As you stated there are basically 2 manufacturing processes, memory and logic. Within each process, they are optimized for the highest yield. This means designs are "tweaked" to work to a specific process and the process may be "tweaked" to a specific type of chip. A fab making DRAM may NOT quickly be able to switch over and make NAND memory.
Worse, combining memory and logic will never be "optimal". Today's CPU chips have large amounts of cache memory. SOC (system on a chip) have an even worse combination; logic, memory, both cache and system SRAM, and NAND flash. Many have analog portion for AtoD or DtoA. Automotive chips do not want to drop below a 5 Vref so now you have multiple voltages.
Great point. And the tiny resistors, capacitors, diodes, etc that go into the boards are also in short supply with backlogs of >12+ months.
You can buy passives, though perhaps not the one you want. Processors and flash, it's hard to make similar substitutions. Right now, most microcontrollers I've been looking at have 52 week backlogs that are staying that way from week to week. I've heard that it will be this way well into 2023.
Especially power J leaded parts
It can't be TOO bad because new electronics are coming out hot and heavy in the world of computing, but those companies wouldn't have been stupid enough to cancel orders. Graphics cards, motherboards or mainboards, even CPUs have those components especially those little tiny resistors.
So, what might have happened is some companies cut their orders because they felt demand would drop because of Covid, and the compute companies like AMD, Intel, Nvidia and all the companies that work with them (long list) bought up the supply that became available.
Don't a lot of those components come out of Japan?
@@johndoh5182 I am simply relaying what a close friend who works as a purchasing agent in one of the relevant processor companies told me.
@@jaymacpherson8167 And I'm just reporting what I see right now as I follow the computer industry.
The world of GPUs has gone through a big shortage for instance because of over-demand due to crypto mining. Now the supply of GPUs from multiple companies is excellent. AMD was stretched really thin because their products have been in high demand, from CPUs for PCs and servers, APUs for game consoles, GPUs for desktop graphics and GPUs for server compute. They now have good stock of most products so they've been cranking out products. Same thing with Intel and Nvidia.
So the difference is where these companies were sitting before Covid hit. Back in late 2019 the world of compute KNEW they were going to have a problem meeting demand for about 2 years so if supply opened up they bought it. Same thing with Tesla. They do preorders. They KNEW Covid wasn't going to slow their sales and now they're pumping out vehicles, no problem.
Also in China, companies are pumping out different products while others took a hit because they projected a drop in sales.
That's what you get with just-in-time manufacturing. And once again almost every PCB needs those tiny components you're talking about ESPECIALLY mainboards and GPUs.
There has been a refusal by many companies to migrate 0.25, 0.18, 0.13um parts to advanced nodes on 300mm wafers. They have been focused on reducing costs by fractions of cents per part and driving ASPs down. They are not used to scaling. Almost impossible to get their designers to move to a new technology.
yes, you hit the NAIL on the head with this....its a mindset thing...due to years of low cost MATURE tech...
Right, but they've competed on low cost and refitting a fab means costs go up, and then your customers run to Chinese fabs.
"to get their designers to move", why, no reason! If you get 5$ for a before 1$ part there is NO reason to port. They can milk the market up to the time a greater work around get substance. SEA need to strenght there own design capacity, the capacity is in SEA. Atmel bought a small uC C compiler design group in Norway (AVR), assemble some flash memory and some medium performance AD and fill an old fab. They will never port.
From the video: 0.28um parts have the lowest price per junction and this is a low margin business.
@Conservator 0.028um, not .28 - Mass market is far from that
For chips that work with high voltages (20 volts being "high"), they need larger nodes so that there is enough insulation.
Analog and mixed signal designs can be moved to more advanced nodes, but their design techniques need to change. This means they need to hire different designers who have such experience. Also there is benefit of including additional capabilities in the chips.
Those talents are extremely difficult to come by and it takes 1/3 of a lifetime to train.. meanwhile the work is too hard for most people to (even) imagine.
You mean the people that are retiring?
@@westerncivilization Plenty of the designers in Korea and China from what I see being published. Much more advanced than what I saw at a large semiconductor manufacturer in the US.
How come scaling down problems worked around in analog ics? Can you explain it like giving an example?
If these designs are only moved to 28nm and larger nodes it won't be that big of a deal. 22nm and lower the gates are no longer planar and this complicates things.
as usual, another timely, informative and well produced video. thank you for all the work, and sharing.
My VLSI professor told us 3-4 years ago this is going to hallen. His reasoning was basically that it takes a very long time to build a fab until it is fully operational so companies need to predict demand 10-20 years ahead. Of course they factored in growth but they have either underestimated that growth or they were too risk averse (building a fab you won‘t actually use fully is a very bad idea if you want to continue to exist). And roughly 5-10 years ago you were already able to tell that the new capacity never quite fully matches the increased demand.
He reckons it is going to take a decade to fix this supply issue fully, so I am sorry fellow EEs, you will likely have to do logistics for a looong time to come.
(he doesn‘t actually hold a professorship at my university and always tells me I shouldn‘t call him that but it is just easier to understand that way)
the biggest joke of this world is this: We build useless empty cities and condos for billions, it wrecks our economies, makes us print money, causese inflation... yet we hesitate to build chip plants because "oh my god they might not be used"....
what is "hallen?"
@@dominicfong6341 happen*
@@collynha6952 Thank you very much for your kind explanation dear Collyn ha.
So true!! It's very much the case of "go big or go home." Companies are very conservative because cutting edge new fab lines now cost so much that just one wrong decision can sink the entire company. If they play conservative and don't expand production aggressively, they may miss out on some business but also, because there are increasingly fewer big players, the decision not to expand can actually float the prices higher due to reduced availability. One just have to look at the crazy ups and downs of the DRAM sector over the years/decades.
Good video as usual. Some thing you don't mention, though, is how inflation will affect the "inflection point" where a cutting edge plant can now price down it's product. In essence, that plant was built with usually debt, and that debt is not inflation indexed. So when prices rise, the relative cost of that debt is lower. If the profits necessary in nominal terms to recover the sunk cost of a 8 nm plant was $1BN, and due to inflation, the chips now sells for twice as much, that plant needs to put into the market about half the chips (not exactly half, because you still have some variable costs of labor, changes to taxes - going the other was as governments move to subsidze chips) for that sunk cost to be recovered and the technology to become "mature", or, as you say, comoditized. There is a lot of complexity to that, as inflation isn't universal in the global economy. Some currencies are losing more purchaseing power than others. And large share of chips are sold across borders. But rather than looking back to building the old mature tech, what you are just as likely to see are faster-than-anticipated decline in price premium for the more advvanced chips. This is good for consumers, and, depending on how each industry adapts - e.g., most automotive being much slower and less skilled at adapting to rapid change than, say, mobile handset makers or (allegedly) Tesla -- you may see divergence and market valuations of companies will diverge even more by sector. If, for example, a previously dismissed as mature industry may be able to step up to more rapidly commoditization chip architecture than some of those already pushing the envelope, and they should get a competitive edge and should see a boost to their valuation.
In Japan a 28nm fab is being built by TSMC. I'd call that legacy.
Plus, once TSMC has a lot more production on N3, N4, N5, it will rebrand what it calls trailing edge, so there will be more availability of nodes like 12 - 7nm. Now, for some electronics you can't drop down too much because the IC would end up being so small that it's too hard to cut the wafer for the hundreds of die that would come from it. So, simple circuits that still warrant an IC but only have a few hundred transistors need to stay on those older nodes.
I had not even considered the physical IC size. Great comment. Thanks for the perspective.👍
I imagine shrinking the node size makes them less hardy as well, with respect to voltage fluctuations and static. Sometimes handling abuse is part of the requirements. Also a factor in rad hardened chips.
@@ricardokowalski1579 Yeah it's an important issue in considering how small of a node you can put a circuit on. In fact that's why Japan wanted a 28nm fab from TSMC built in Japan, for general electronics and not HPC.
Even with 7nm you're at 90 million transistors/sq. mm. What happens as you try to make a die smaller and smaller, you end up getting a bad ratio between edge area to have a border for the cut, and the area of the actual transistors.
It's a REALLY hard answer to find but I saw a comparison with 2 different node processes, 90nm and 65nm and with the 90nm it said the smallest die could be 0.683mm x 0.683nm and for the 60nm node it was 1.533mm x 1.533mm.
It gets to, how fine can you make the border between each die, and I have to think getting below a die size of 0.5mm x 0.5mm is near impossible. So, with a 7nm node you can fit 22 million transistors in that small of a space.
But then how do you make the connections between the die and a PCB? It's not only the fact that you need an edge area for making cuts between the die, but you also need wires to come out of the die to create some kind of package. Can you fit 1000 connections in a die that's only .25 sq. mm?? I think not but I could be wrong.
Price becomes more expensive as the node becomes smaller, so this is a REAL issue. I'm sure there's variation in pricing so at some point a smaller node might actually be less expensive, but once you get down to the state of the art nodes, there's pretty much always a rise in cost for smaller nodes.
At some point you can't make a die any smaller, so if you hit that point with a 90nm node, it's simply wasting money making that die on a 7nm node. On top of that you have other issues come up as the node gets smaller as other comments point out.
I'm not a computer guru but after the next Carrington event y'all will be talking about the most anti electrostatic chips out there. Screw the size. I'll take functionality and reliability over the latest greatest ideas out there.
@@kirkthiets2771 There are effectively 0 ICs that can handle that. (MAYBE some satellite ics that can, but basically nothing consumer that is used on the surface of earth.) I think even the very first transistors will burn out instantly, not even mentioning ICs. Tubes are the only things that can reasonably stand those kinds of voltages. Russia is still using tubes in certain military applications btw ;)
We started pulling components off recycled PSUs in our workshop and have started selling on eBay to get low volume amounts of used product to people who need them :S
Thank you!
You’re doing holy work. man. Never stop 🍻
*to make money
@@dieselgeezer18 one does not exclude the other, and usually only the combination of both is sustainable
This is absolutely incredible. These are some of the best and most informative bits of content anywhere.
These older node chips aren't just in consumer devices. They're critical in industrial manufacturing components which can't be easily swapped out or upgraded. It's weird how gas can triple in price in a couple months, but these components can be sold out for years in advance and up until they ran out of them they were selling at the same price as always. If the PS5 is still out of stock for the foreseeable future, maybe $500 just isn't a reasonable price to stick to. If Atmel and the like would double or triple the prices, we could at least be making sure they're getting to places they're really needed while also allowing them to expand capacity profitably. I don't understand why technology prices all seem so immovable.
I agree with your goal, except I think the approach of relying on supply and demand is not explicit. It's forceful and does not serve the customers which in turn will not be good for the market as a whole.
I studied marketing in college. Personally, I don't know the exacts is this market. For example: Are opportunity costs incentivizing large corps into buying up all the stock ? Are contracts established first, before chips are made ? My point in those questions is that I personally don't know enough details to know exactly if there are problems which need to be solved. However, f the goal is for "supply" to exist for specific market segments then a more explicit approach would yield better results.
The solution is to allocate a certain quantity of the goods for each market. This would solve the problem. The approach of relying on supply and demand is only a viable solution IF that market segment is unwilling to pay the higher costs. That is a hope and a gamble. Also, with that in mind, how sure are we that the end consumers would buy these goods at elevated costs ? >> If the companies are unwilling to buy at high costs then why would consumers ?? ((Just to be clear, my assumption is that large corporations bought out all the supply and prevented supply for end users. The chips went into industry rather than sbc's / tinkerer's / schools / hobbyists. ))
Again, I think a more explicit approach is better. If half the stock was allocated for end consumers/small companies and the other half for large corporations then we would be sure of a fair chance. I think a time frame should also be set so stock doesn't sit unused. >>> 50/50 split for 5 or 10 years sounds like a good plan.
Chips are increasing in price and still being sold out. Moreover it's much easier to increase prices in consumer devices rather than industrial devices. A consumer will buy a laptop or a gaming machine without thinking too much of minute price fluctuations. Even a 50 dollar increase in a laptop would not be noticed. In an industrial product the prices and volumes have been negotiated way before the product is even designed. Many times also your customers are pretty fixed and you have long running relationships with them, thus a small increase in your sales price would be noticed immediately. Nobody in the industry would pay double the price for a microcontroller unless they have really no other option. Moreover big companies like Microchip or Atmel don't operate like a retail store, where prices can fluctuate daily, the prices there are pretty fixed and customers expect that as well. Nobody expects that a microcontroller is gonna cost 50 dollars and noone is gonna pay that price as well.
They are double the price on digikey.
@@InTimeTraveller Totally agree. I think this whole chip shortage thing is new. It's so new the market hasn't reacted yet. This seems like the initial the shock. Hopefully, it gets sorted out in a good way.
Personally, I hope supply is evened out by making sure the stuff goes to the right people. Like >> prevent scalper from hoarding ; don't let companies buy up for crap IOT like alexa thermostats and wifi microwaves.
I feel like some of those crap IOT devices are not adding much value to our lives, instead they are adding money to the corporations. There "are* cool IOT devices, but just adding Alexa to already existing tech should have common sense applied.
I think the arduino & SBC types of devices should get priority caz tinkerers have a good chance at making contraptions which will improve our quality of life. For example >> the raspberry pi tinkerers were really into car mods b4 it was standardized as it is now >> rear camera's + integrating the phones into the stereo.
The leading/bleeding edge may be sexy, but so many industries run on jelly bean parts that you just can't get -- analog parts, mixed signal, power supply/regulators, and so many more. Quoting me an 18 month lead time on some of those jelly bean parts? Kiss of death!
we buy plcs and other automation equipment . we don't get down to the chips. we just need plc's . everything is backordered for months/ years. we go online and try to order . suppliers has 2500 or so on order from the manufacturer but are only getting like 100 every few months. none of these chips are leading edge. so we are fucked i guess.
It’s all fun and games until that MESFET discrete your fancypants state-of-the-art module needs is backordered!
Great video! I like the details, where a lot of other videos on the topic offer vague assertions without backing them up.
I also loved the Arizona water joke. Seriously, I am amazed the problem isn't brought up more often, with the worsening draught and all.
It is not only microcontrollers. A lot of other analog and mixed signal chips too.
ADC, DACS, motor drivers, voltage regulators/references e.t.c.
Usually they are in stock at speculative resellers (for abysmal prices), but not in a normal one's like Digikey or Mouser.
speculative cough Rochester cough
ADCs are sold out, luckily there are still some available. Couple weeks ago I went with MCP3421 while considering ADS1115 as backup when more channels are needed.
I thought you were gonna explain why I couldn't find any yellow tortilla chips at the store today. That's the real problem nobody is talking about!
All the corn is going into ethanol now. It's more eco
Sorry about that I took them all
@@johnl.7754 I knew it!
@@raul0ca Well the process of making ethanol from corn takes more energy than you get in ethanol. So its a totaly wasted efford.
Only sugar cane and palm oil can make ethanol with an energy surplus.
Its just the lobby from big AG that gets the politicians to do such things, cause it will send the corn price skyhigh.
The corruption...
I was hoping US would remove the ethanol, to help with the hunger crisis that is happening in the world atm. But they did the opposite, and increased ethanol in gas... Making world hunger worse cause price of corn will go up..
Greedy mother fuckers, is all i have to say.
@@MrDanisve Haha I was being ironic
That's pretty curious. Looks like the market doesn't need that much the newer technology, but still needs LOTS of the old stuff. Really curious indeed.
I don't know why it had never occurred to me that there is a brisk market for 25+ year old semiconductor device fabrication equipment, and that they are continuing on in revenue service cranking out in-demand chips. So there is no ongoing production of new equipment for old nodes? What is the oldest node (and largest feature size) still being used to make chips?
Some laughably old nodes are still in production for niche markets - HV CMOS (up to 300V) and CMOS image sensors are particular examples where "cutting edge" is in the 300nm range
I'm not that surprised, to be quite honest. Some chips are now very old but have simply settled in the industry. I don't think there are many embedded engineers who have never touched an ATmega328. And you don't need a 4nm node to make those. But u can easily tell that a lot of places have them either put of stock or at increased prices. They've always been abundant, and very popular among hobbyists and R&D engineers. But they've become so much harder to come by.
An enormous amount of III-V production has never scaled beyond 0.5um. There’s simply no need in many cases, and the tools and processes used to create them aren’t capable of doing much better.
You can see a list of fabs here: en.wikipedia.org/wiki/List_of_semiconductor_fabrication_plants. 350, 180, 90, and 45 are all still pretty commonly used for chips where smaller process nodes are not needed.
@@Marco_Onyxheart
They still make the LM555.
And probably will for a long time.
An amazing analysis, as always. Thanks!!! So many good ebservations in the comments, too.
The video was looking at the main source of the current problems, however, it's also worth mentioning that chip designs are tuned to a specific implementation of technology at a certain node, so production can not be easily shifted to another fab line, even at the same node. The more advanced the technology node, the more this is the case. This creates inflexibility, so even when there is wafer capacity available somewhere else, it's no help in expanding production of a specific chip.
Older lines often produce many different chips, so production scheduling becomes a delicate balancing act, trying to satisfy ongoing demand without gaps in the supply and to minimize the frequency of retooling.
In addition, while the main issue is silicon, there are material and capacity shortages in a lot of areas, affecting current semi production.
At the moment, lead times are crazy and often previously confirmed orders are getting partially or fully "deallocated", just before delivery.
Then of course, greed kicks in. The worst I come across was a supplier trying to charge 210(!) times the price of the normal supply price of a chip. Counterfeiting is also rife, with buyers forced to do extreme measures like x-raying chips before accepting them. Leadtimes? I've seen 99 weeks. Increasingly, manufacturers just throw their hands up and simply EOL part of their product line, causing mayhem for equipment builders. Some companies producing and selling hardware with thin margains can often make more money reselling the surplus parts they actually have (maybe becuase of over-ordering, maybe because of missing a crucial part to complete the BoM) than what they make from building new gear which may be at negative margains, due to high component prices. Of course, selling surplus stock is just a sugar hit and the reduced production may kill their business in the long run.
...but I'm sure you've all heard similar stories! ☹
The number of designs i had to convert from one MCU to another (and various other parts) boggles the mind
often the same design many times for any quantity of roughly equivalent processors we can get
It's not only you. The largest (or second large, depending on the year) car manufacturer had to work around a network switch in a device that will sell in millions.
There are a *lot* more than 200 fabs in the world. I visited one in the smallish town where I live, and it isn't even the only one there. Most fabs make very low density discrete semiconductors, eg transistors, particularly power transistors (big & simple but also highly precises). Op-amps and 555 timers would also fall into this category.
Great topic but I feel you missed out another big factor that is influencing the industry at the moment.
We are a small electronic production company that build some products that ship worldwide. So we see these issues on day to day basis.
At this moment brokers are making millions by buying in large amounts of chips. Even new batches that come right from the production floor. Just to sell it at 4-10 times the normal price.
They buying all stock world wide to create a shortage too so they can sell at a high price.
For example we got a notice of chip back in stock from a manufacturer. Batch of 10k chips.
Direct rushed in to buy 100 chips to keep production going. But the full batch was already gone. ( within seconds )
A few days later a batch of 9600 chips are showing up at a broker inventory for 8x the going rate of the manufacturer.
Also almost all 3,3v regulators that are pretty common to use for all small logic. Are almost all gone in the world. But huge deposits are in the broker hands trying to sell them for 10-14 dollars a piece that cost normally around 1.5-2 dollars.
Luckily a few chip manufacturers are now putting in the 100 chips per company rule. Meaning you can buy only 100 pieces for a set periode. With a batch size of 50k chips this gives a bit more breath room that there will be stock without brokers buying it all.
As it seams that putting money into chips inventory might be the best investment at the moment as the interest and payout seems to be much higher then invest in housing market at the moment.
Check a few chips at octopart . Com and see how many parts are only available trough brokers 😉
Ugh. I hate advocating for regulation but this might just need the govt or better still an industry body to step in.
Tokyo electron is starting to make new tools of their previous generations etching tools due to customers demanding them. It's been over 10 years since a new one came off the line, but they're coming back
Glad I'm finally getting context to the chip shortage
Thank you for explaining this rather complex subject so well. Even I could understand most of it!
Moving to a larger wafer size is almost building a new fab with almost all equipment needing to be replaced. It is not as simple as you suggest.
It never is. Think I made that pretty clear
Just try buying a 2000V 2000A optically fired thyristor if you want to try long lead.
The 300mm chip diameter has a 1% manufacturing yield…
18 month delivery time precrunch…
The new fab Globalfoundries is building in Singapore is also a trailing edge 300mm wafer fab.
Preface: I've worked 25 years for semi companies, Microchip the most recent (out of 40+ years designing products).
This video, and all the other ones I've seen, "blame it on the fab capacity"
It's not the fab capacity.
It's packaging capacity. Testing capacity. Mask costs (especially 28nm and 22nm. At
Hi Paul, thank you for your insights!
I want to thank you for an excellent summary of a much neglected topic. I retired last year from Microchip, working mostly on Reliability / Quality issues with mature EEPROMs for Auto and high Reliability customers. When you have to get long term performance out of non-Volatle memories even 250 or 180nm nodes can be touchy. Steve Sangi did a great job building up a 'trailing edge' business. Predicting the overall industry can be pretty difficult though. When I was at IBM Microelectronics I recall a senior engineer commenting (in 1991) that in ten years 2 or 3 Fabs would support all the chip needs of the workld's computers (ie, no phones, smart systems, etc...).
Essentially all STM32 parts are completely out of stock around the world.
There are a few places with stocks of micros with factory price of ~$15, selling for $60-100 per part.
I don't see that. LCSC/JLCPCB assemly has thousands of different STM32s in stock. For my current design I'm choosing from three different low power parts capable of driving segment LCD.
11:52 did they really paint a Mondrian on the side of the factory? That's so cool.
I am glad that as a hobbyist I tend to buy components in much bigger bulk quantities than what I actually need. Some of the voltage regulators I bought back in 2019 are still being put into new hobby designs today since I can't find anything equivalent today! This is a great reminder to engineers (both hardware and software) that we should be focused on making our designs as modular as possible, so that you can easily swap out voltage regulators or even entire SoC's.
PS: Wow, the ESP32 is far more powerful, much cheaper, and more available than the ATSAMD21G18A chip that you covered in the intro to this video. Then again, companies probably find it cheaper to spend exorbitant amounts on chips like this than to completely re-design the microcontrollers and firmware for their controller boards.
It's not redesigning that is the problem, it's the supplier. A company like ST meets all norms, while tensilica, the maker of ESP32 does not - because they intend to produce a hobbyists tool.
@@LMB222 yeah, I did not mean to imply the ESP32 could replace that chip for many use cases especially industrial/professional ones. But it is interesting just how much more powerful AND cheaper it is
@@LMB222 It's not that ESP's chips aren't used at scale, a lot of iot power products like colour changing lightbulbs and remote switchable power sockets used them at scale; then again these are products with... shall we say no particular expectation of reliability, often just borderline compliance to basic safety norms, and many of those have even since switched to even cheaper and wonkier chips, they plain don't care. You simply can't sustain a semiconductor company like that on hobbyists, but you can sustain it on kilotons of future e-waste.
My friend your channel is pure gold.
I’ve been wondering about the supply chain of microcontrollers for a while - it doesn’t make sense that car parts would use cutting edge processors
In my experience, embedded systems engineers have gotten used to overspecifying chips vs optimising code better - this can and should change
That has nothing to do with the shortage. Many micro processors are in short supply. Squeezing your code into a smaller chip isn’t going to help this problem.
Automotive chip needs to comply with automotive standard(safety, stability). That mean they go through more test process than other commercial chip. lead time can be up to 2x compare to ordinary chip.
All of this talk is interesting, but kinda missing the point anyway. Most consumers do not want or need a car with computer chips anyway. Car COMPANIES would rather put chips in their products to enable things like tracking of customer's behaviour, and to justify a higher price tag, but this is just extra expense with very little utility to the end user.
@@andreaslind6338 True, anyway all the user wants is an iPad instead of the shitty car screen and OS.
@@andreaslind6338 then just buy yourself a 70's car. Give up the fuel economy and the horse power. Give up the 6 gear transmission that adapts to the road. Give up ABS and traction control. Go back to cars that are dead at 100k miles. Enjoy exhaust pollution that was bad enough that people k*lled themselves with it.
These things are there because people want them.
1. An advantage to using trailing-edge technology is that it's easier to find substitutes. As you say, these are commodity parts with multiple suppliers, so you could perhaps find a software-compatible part with a different package or extra features. Compare that with finding a substitute for a bleeding-edge Nvidia GPU.
2. You may not need to construct new buildings to convert a fab from 200 mm to 300 mm wafers, but you do have to acquire all new equipment, as 300 mm wafers won't fit in 200 mm equipment. There are half-measures that can help, like re-optimizing the current processing procedures, jamming in more 200 mm equipment in places where it would help, and eliminating special-case processing steps as much as possible. These are some of the work-arounds used in the initial 200 mm to 300 mm transition in the 1990s.
Many thanks for this information!
I wish i had a sliver of knowledge of the info of your videos!!! I do enjoy them and appreciate the work, explanation and work you have to do to get to my level of understanding! Blows mt mind! THANKS for all your work and ROCK ON!!!
Great comment, keep studying from multiple sources, and you will increase your understanding.
The SAM is of course an inherited Atmel chip.
Indeed all Atmel, Microchip, NXP, Nuvoton, STM low-end micros are sold out a year in advance and have been for a year and something, everything under $3, most things under $7 as well. And it's been a lot tougher to work around than shortages of higher end chips, which are still kinda available. Also special function analogue/linear semiconductors of particular kinds are hit hard.
Chinese domestic chips have better availability. Maybe more 45-28nm utilisation? I do think maxing 28nm and node shrinking designs needs to be an intermediate goal, i feel, not adding another 90nm fab. But i think more 28nm fabs won't go to waste.
If the lead times surpasses the design time heavy disruptions is the design ecosystem emerge. Before no one consider to porting a 2$ chip. If you already port why not give up the old silo system of compiler + squeezed old fab + broking distribution
Not all, there are several STM32L stocked for low power with LCD control for example- And then there's the RP2040, $1 chip with $1 flash+accessories, good performance and excellent support. People should be using it more.
Atsam is a 32bit chip and will migrate easily to risc-v chips which will be made on old equipment in China for pennies.
The trailing edge chipmakers understand these risks and are willing to let the world burn to avoid getting crushed.
@@jan.tichavsky rp2040 has external firmware and no provision for ROM encryption, which makes it nominally unsuitable for most commercial products. And open source projects often get fouled by its glitchy ADC, which is not entirely surprising of a company with no such design experience, they'll get it right eventually.
@@SianaGearz And last I checked, you can't buy more than 10 at a time. It's limited for home DIY and school farting around.
Always good research and reporting on this channel
So short term speculation lead to short term decisions which impacted long term supply. Who would have thought of that?? This is what happens when single minded immediate profit companies thinking about the next quarter make rash unfounded decisions on short term speculation.
Oh thank god someone finally made this (excellent) video so I don’t have to keep explaining this to people! Now I can just give them this link!
There is a fortune to be made for the person who sets up a proper e-recycling shop that focuses on chip reclamation with efficient acquisition, processing, and quality control. There is no shortage of e-waste and it's current use is messy at best.
Heh, china already does that - where do you think all the "counterfit" parts on ebay, aliexpress, and amazon come from? They're all re-marked recycled chips
I remember there is a video where someone try to build Iphone in China, and they go to this electronic market to buy the Apple processor
@@edojayakusuma8209 Probably a Stranger Parts video...
Ok for analog or simple CMOS ICs but not for MCUs or low-power processors because you also need the software and tools to reprogram them. Despite many retroengineering projects, it's quite usual to not be able to repurpose the 90's and 00's generations. If there is a way, devs often occur challenges like assembler-only compiler, old C or C++ standard, DOS program, programmer with parrallel port interface or (in the case of processors) supported only by NetBSD.
The Chinese are already doing this. Nobody wants reclaim chips. The whole process runs the risk of damage plus you don’t know how old the chip is.
For the people going "why retool not build new when the building is the cheap part" your technically correct, but lacking context. A building might only be a couple of million of a several hundred million dollar line, but the earthworks, foundations, and fitout of the shell itself make up roughly half the total time from design to commissioning, and the vast majority of the external requirements regarding permits and legislation. And thats assuming you have the space, power, water, workers, administrative capacity to not only run 2 lines, but run one line while the others being constructed, which can entail far greater disruption than "retooling" even when it means essentially complete replacement of the actual machines in situ.
Lets you know how fragile the economy really is, and how just one bad decision can totally wreck it in many ways. The Economy isn't something that can be managed, it's literally billions of people making billions of individual decisions every single day, you can not predict what is going to happen. It's next to impossible. Business assumed the economy would slow down dramatically, but consumers kept consuming.
"...one bad decision can totally wreck..."
That is exactly the operational strategy of LIM (leftist invasive management)...
IE; the idea that it takes only a relative handful of people to destroy
what it took billions of people create.
Now coming to a capitalist based societly near you.
I think I understood something you said at some point in this video but found it extremely fascinating!
" TSMC Arizona plant. Note all the plentiful water" 🤣
Lol right!?, 😂
They run the Palo Verde Nuclear Generation System with the treated wastewater from Phoenix, AZ.
@@hithere7382
Nuclear power plants do not require ultra pure water. That makes water water recycling MUCH easier.
@@jstasiak2262 Where in my statement did you construe my meaning as reactors need ultra-pure water for cooling?
Thank you. Managed the Si validation labs for I,A and Natsemi. Best synopsis yet. I have now left the industry.
chip shortage should be a lesson to engineers to make their designs cpu architecture agnostic. i started doing it with my designs as standard practice around 10 years ago. little did i know how it would pay off later.
CPU or MCU? The time spent on the latter is getting all the peripherals to work, and those are not agnostic. Then an even more amount of time is clearing up the corner cases of power-on reset, oscillator start-up, UART timing, all over temperature. Those issues are typically specific to a signal part family.
for MCUs i've started writing my firmware in rust since the community made a lot of efforts to create hardware abstractions for a lot of middle-end MCUs. but that barely solves the problem of acquiring said chips since they are either sold out or have dramatically increased prices. I for example replaced the STM32F103xx with a GD32VF103xx since they got an identical pinout (infact it was meant as a drop-in replacement) but it required changing the whole compiler infrastructure since the GD32VF103 is a RISC-V processor and the STM32F103 is ARM. And even though the GD32VF103 seems to be relatively unknown it was still a nightmare finding a reseller that actually got stock of it.
@@minespeed2009 ha ha, all that HAL is what scares me; more abstraction means more bugs, and it’s tougher to find them. I’m not even an embedded guy (rather RF Microwave) but I won’t touch anything but bare C when I need to implement an MCU. Don’t get me started on using Circuit Python, Arduino, etc for shipping products. IMHO it’s a nightmare waiting to happen. As I mentioned above it seems 90% of the effort is spent fixing 10% of the corner case stuff.
I actually was messing around with an Arduino compatible MCU module, and it’s got a known bug of not coming out of a low-power sleep when the UART is low, of course it’s beyond me to fix it through layers of software abstraction. But it’s this kind of shit that just drives me nuts.
@@rfengr00 it definitely makes it a bit harder for some cases, but in the end it takes me about as much time to implement something as in c since the rust compiler is VERY strict when it comes to unsafe programming. like refusing to compile when you access the same peripherals in two different code sections without transferring ownership. (ownership checks are done at compile time so no binary size or speed penality). But im also only an EE student and therefore haven't worked on larger commercial products.
So you’re saying engineers should strive to do something that’s literally impossible? You have PC boards will you can just swap in any brand of CPU you want? And you take the time to make low level drivers for each one?
Tell me more?
Always interesting and professional reports shown here on this channel !
A "better, faster" chip isn't always a good thing.
Has been for the last 62 years and 5 months… it’s the very definition of the electronics industry, since the inception of electronics … Nobody is buying worse-slower stuff or the same stuff they bought last year for that matter… Unless that someone is the government…
@@robertw1871 I'd agree with you from the customers perspective...from the perspective of someone who had to maintain legacy designs that kept selling for decades, not so much.
@@gregoryhall9276 Yep that’s our nemesis… 17 years in medical diagnostics engineering… thankfully I’ve been on the bleeding edge since that nightmare… Not sure it’s more fun spinning a product that’s not even released to production and already 6 months obsolete… Never a dull moment though, sales weasels selling things and taking orders for things that are physically impossible to make…
@@rockapartie That’s why almost no industry is anywhere near the bleeding edge, well except For SpaceX, Tesla, Alphabet and Amazon’s ever expanding giant empire…. It’s also why the dinosaurs are going extinct, they don’t adapt and costs for maintenance on ancient technology that doesn’t perform well enough to be competitive takes them out… It’s why those few companies that are able to adapt run nearly everything already…
@@rockapartie Farm tractors are mostly software driven these days and very high tech, pretty sure they get updated fairly often… Manufacturing and Logistics are where the very rapid change is actually driven from… Medical hasn’t changed much in 30 years though, mostly because Medical companies are owned by holdings companies that are only interested in milking what’s there not investing in the future….
Your channel is brilliant. Thank you for posting.
Well Renesas mcu are in very heavy short supply. They have caused lots of damage to Japan's car industry in the last 2 years. They may have single-handedly reduced Japan's car output by 500,000+ units over last 2 years. Renesas are very slow to invest even though they are making big profit. Earthquake and fires also don't help there Fabs.
This is by far the best explanation I've ever seen
Why did the US give up so much of it's fabs? I sorta remember when AMD did it. I think they claimed they would save money and could focus on chip design. Intel kept theirs till recently?
I kinda agree with the people saying giving up so much manufacturing and industrial capabilities to make the rich richer and maybe tie everyone together so supposedly war is impossible was a mistake.
It's due to globalization. If you have a chip factory in the USA but the factory making phones or laptops is in Korea you have a problem. The solution is to raise tariffs on all imported electronics and force local production in the USA. When China goes to war, trade from Korea and Japan will probably be lost so the USA needs to be self sufficient in terms of electronics and manufacturing in general.
@@drscopeify have you considered how large the tariffs would have to be to make it more cost efficient to have fabs in the USA instead of outsourcing it to Asia? any basic economics course would teach you that these tariffs would create significant damage to trade balances. Also, outsourcing the fabs to asia is a way for these corporations to mitigate the seasonality risks that comes with the industry. The idea that a country should stride for self-sufficiency due to risks of war in the future completely trashes the idea of competitive advantage and the reason why countries like the US can be so wealthy in the first place.
@@KrDavidLee Yes you need very very large tariffs and the USA will suffer massive inflation but this is going to happen anyway if China goes to war and it is not investing trillions in a modern army to look pretty I don't think... China isn't building aircraft carriers to just pose for photos. So why not force local manufacturing in the USA today rather than wait for China to start eating up all of Asia and then being left with the pants down. Imagine inflation in the USA without exports from Japan, Taiwan, China, South Korea....
Awesome channel, cheers from Brazil!!
This reminds me of a project I have been working on every once in a while.
It is an ASCII video converter and player, should really get back to working on that soon :D
Isn't that built into VLC already? Or do you have a breakthrough advancement in ASCII-video in mind, that you're keeping close to the chest?
@@Validole He no, I'm just dumb and want to do cool stuffs c:
* Hides secrete code *
We have the exact problem right now. Parts that used to cost 6 Eur are out of stock for up to 92 Weeks. And broker offer them for 400-500 Eur right now. Its just insane.
I look forward to times when this type of greed will be punished by law extensively.
God just spent an afternoon redesigning a system to use an alternative to the ATSAMD21G18. This chip shortage is a gigantic pain in the ass.
Are the ATSAM more affected than the STM32 ?
Those that work to switch away from a part are not coming back.
I didn’t know god was available for contract 😏
@@robertadsett5273 I have a board or two that God would have started working on already, if he had been listening to some choice words I had about them.
@@alexforget Not much difference really, i would say...
Giant pain, but good on exercising the skills, and the paycheck too.
This installment was uncommonly interesting and useful to me. Many thanks!!
I got a bad feeling that this problem will most likely carry through 2024 or 2025....
It's not like they can just start chopping down more trees. Sorry, I had to. My son and I just watched The Lorax for the 100th time and I still love it.
@@mathew00 i have the solution...
YOU MUST CONSTRUCT ADDITIONAL PYLONS.
Brilliant channel I've learned a lot thanks. You'll gain a patron after xmas. Thanks for explaining the business layers of chip manufacturing.
I have been caught out buying PICs. The chips that I prefer only need a K or so and they're over a year lead times.
Thanks Deer 🙂
NEAL
Just like the demographics and climate problems, the pain hits long after the deed was done. Notice how a lot of this shortage is military industry related.
This vid aligned really well with a rumor of Samsung looking to buy NXP.
I wonder if this is where we see some kind of disruptive innovation. FABs, to me, are identical to integrated steel plants, which were disrupted over decades by minimills using a more efficient production model. I wonder if in a similar way, the lowest-grade silicon will see some new method, and slowly this method will move up the quality chain?
Chips can be like nuts and bolts. A match box car needs a wire for an axle, but a mining truck needs....... I know that supply chain disruption has been critical and legacy equipment give fabs an advantage especially in niches.
Sony's old fabrication plants in Japan proved that (at least with entertainment videogame system processors) they could take the early 1999 finalized silicon engineering sample Emotion Engine CPU and Graphics Synthesizer graphics chipset from (I forget at the moment) whatever the manufacturing process node was in early 1999 or late 1998... because the prototype PlayStation 2 engineering sample motherboard was demonstrated by Ken Kutaragi in March 1999 at a special Sony industry presentation not really made for consumers iirc but was still reported by the videogame magazines of the time back when internet news was just not organized or reliable.
Anyway Sony went from whatever that initial process node was in early March 1999 to the mass production process node for launch in March 2000 (because the Sony press release said the die shrink would allow for higher yields, lower wattage, lower thermals... stuff that was common before Microsoft Xbox changed that in 2001 and later Sony followed suit at great cost and risk to them) all the way to the 2006 Sony PlayStation 3 launch which featured iirc a 90nm die package of the EE and GS processors which used a heat spreader.
However due to the difficulties of that time and customer fears and confusion as well as some bad marketing campaigns on both Pro and con sides... the PS2 chips were removed from the PS3 yet PS2 chip production continued and made some major improvements to 65nm production node by 2008 to 2009 which included the expensive RDRAM memory chips inside the die package and basically no need for a heat spreader and definitely much lower wattage consumption and thermals.
So for nearly ten years of the PS2 CPU and graphics processor production, Sony definitely and evidently proceeded to provide historical evidence that processor die shrinks are indeed possible if a company wants to do them.
Now the EE and GS processors were proprietary to Sony and as such they had more engineering control. With the PS3's CellBE CPU and the Nv47 based RSX GPU, both of which had to be reactionary produced at 90nm in 2006 as opposed to 65nm which for Cell CPU was Sony's original target node... both of those processors continued to be fabbed until die shrinks made possible around 2011 to 2012 to either 40nm or possibly maybe 28nm.
Then all that changed and had to be dropped because Sony had been sustaining losses and Microsoft had called for next generation hardware around 2011 and we ended up with AMD becoming the designer of the CPU+GPU package and TSMC being the only fabrication plant to provide processors on a 28nm process node for 2013 mass production.
I don't believe Sony has much control like they had when fabbed their processors at their proprietary fab plants so I can't recall what die shrink the PS4 Slim reached around 2020 to 2021 but this is just an example of what is and was possible.
However TSMC pioneered and reached the cutting edge production nodes... YEARS AGO... bearing in mind the harsh reality of the year 2020 problems the least we could expect from other chip manufacturing plants is to have improved their processes.
As pointed out in this video, a 90nm fabrication plant may be limited to 90nm production unless there was some way possible to replace and reformat the manufacturing tools inside the plant which sounds easy but is just insanely expensive in reality.
Still Intel owns a vast array of fabrication plants... if they wanted to or if it was plausible, they could provide services to AMD and Nvidia for GPU fabrication nodes... however iirc Intel does not have the same technology as TSMC for that aspect because their main products are CPUs
Outside of that, fabrication plants that reached 28nm should in theory be able to handle die shrinks for many processors to provide increased yields in theory... however there is a cost there and maybe some manufacturers don't see the point in paying for it if their mass production processors are already being fabbed at 130nm, 90nm, 150nm, etc.
Therefore if Intel wanted to produce Itanium CPUs on 28nm and lower as an example or even have AMD make 28nm Phenom II Hexacore CPUs... all those things are possible but those are not things that their industry are about... thus other processors for industrial use and car electronics are a different industry but are still affected by the year 2020 problems.
I wonder how long a lead time for a die shrink is. It seems likely to me that the PS2 logic was designed in parallel for the stable node and the future cutting edge node, but of course the future node couldn't be manufactured... until the future came about. It was likely too expensive to make systems on a node which would be out of date by release - beefier PSU needed, more silicon consumed, reliability compromise due to higher temperature, all that. And it was likely too expensive to wait for new process to make early devkits - the cost of opportunity is the highest possible cost.
I mean intel's whole "tick tock" thing was one year they do a new design in node X, then the next year shrink that design to new node Y, then do the next new design in node Y, and shrink it to node Z... all until they couldn't keep that cadence up anymore. It takes a lot of people and resources to do that - probably something the companies that make microcontrollers don't have, and don't have a benefit to do. Microchip isn't trying to increase the clock speed and reduce the power consumption of the same chip from year to year - once they do a design, they just want to keep cranking out the same thing forever without any further engineering / design cost.
Microprocessor Report did an in-depth examination of the PS2 chip set. The intent was always to do mass production at the 180nm node but this was a bit ambitious as Intel, then the global leader in process node advancement, was still ramping up for 180nm product. The 250nm version of the PS2 chip set had been intended solely for engineering samples and early developer kits. Due to delays in getting the 180nm line ready, the initial batch of PS2s for the Japanese launch used the 250nm chips, which made for a lot of red ink to keep their schedule. Another oddity in those early units was rear bay for adding a hard drive used the PCMCIA Type III form factor, mainly seen in laptops, rather the common 3.5" form factor used widely in desktop PCs. Sony offered owners of those early units and adapter to plug into the slot and use a 3.5" drive externally.
The PS4 APU, along with the very similar Xbox One APU, had just the one die shrink, to 16nm. This enabled the PS4 Slim and Xbox One S models to be produce and replace the original models. The additional ceiling also allowed the creation of the PS4 Pro and Xbox One X. That was also where companies hit a wall on the value of die shrinks as opposed to new designs. A casual observer of the game console business would reasonably expect at least another die shrink, with perhaps the Pro and One X becoming the mainstream models as the generation wound down and a new generation launched. But that would be ignoring how much the economics had changed at the newer process nodes. This was one of major reasons Microsoft made the Series S. It made more sense to have a low end version of the latest architecture than continuing with the existing product in that role, as another die shrink would not reduce cost as well as it would have in past node generations..
@@SianaGearz That is correct. The launch PS2 was always intended to be mass produced at 180nm, while the 250nm version of the chip set was produced for engineering and early dev kits. The first production run for the Japanese launch had to use the 250nm chips due to delays on the 180nm line. This was a painful expense for Sony but they kept the launch schedule and had a very successful product in the long term.
Lukashenko in Belarus two years after lockdowns helped cause the current economic crisis: "Don't you guys all feel like Covtards?"
PS in Russia we still a use a lot of Soviet-era processors since our microelectronics collapsed in the 90s and never really recovered. For example, a 32-bit K-1839 processor created in 1989 STILL flies aboard the GLONASS satellites.
Can you make a video if mainland China attacks Taiwan and the consequences for chip market
The picture you used in your thumbnail I used in a class project like 4 years ago in college 😂
I really hope all that extra capacity hits the market right when demand slows down. Would lead to very cheap hardware which could lead to new innovations.
That is exactly why companies are hesitant to invest.
great video. One thing I would take a bit of issue with is die shrinking old designs is not "always" a good thing. Less die size does mean cheaper and less on die parasitics, which means faster ICs. This should be a "win/win" as who doesn't like faster ICs? As it happens, the FCC may not. Sometimes a die shrink will make a processor too fast for it's own good, and where a design used to pass radiated emissions in a larger package, it will not in the smaller one.
No, a die shrink NEVER makes an IC too fast. An IC is a clocked circuit. The clocks that get stuck into the IC determine the speed.
No that's not the issue. The issue is a node can be too small for the number of transistors needed to go into a single die (IC), so you end up with a TINY die that has to be cut from the wafer and cutting a die ever smaller I'm sure becomes very hard.
For instance TSMC's N7 (7nm) has a transistor density of about 90 MILLION transistors/ sq. mm , N5 is about 170 million/sq. mm and N3 is almost 300 million/sq. mm.
So, at the low end of these advanced nodes, 90 million transistors/sq. mm
So you design a circuit that has a few hundred transistors. You still need it to be an IC, but you can't put it on N7 because you couldn't cut the die small enough.
The other issue is power. Internally I can have transistors running at tiny voltages or current, but to drive signals off the IC to another circuit requires more power. You can end up with having to make a larger circuit on a small node to drive signals and this can be a waste. For instance AMD is making Zen 4 with two different TSMC nodes. They use TSMC N5 for cores and use N6 for the I/O functions to drive signals out of the CPU or into the CPU. Some controllers can be in the cores but most are in the I/O die. For Zen 2 and Zen 3 AMD used GloFo 12nm for the I/O die and TSMC N7 for the cores. Part of that though was having to fulfill contractual agreements with GloFo, for which I think AMD is finally done with them. So THEY could have capacity for different chipmakers as AMD finishes off with those contracts, but Zen 3 is still going to be made for a long time.
So those are the reasons smaller nodes can be no good. You have a circuit that doesn't have enough transistors in it to warrant moving to something below 50 - 28nm, and that's probably the case with most, and then you have the issue with having to drive signals off the IC and need more power to push that signal.
The other things aren't a problem. Clock speeds are controlled by the chipmaker.
Other issues I imagine could be more sensitive to noise, like you're in an industrial area and machines are producing a bit of magnetic waves or strong frequency transmissions and smaller nodes are more susceptible to external noise.
One thing that node shrinks are supposed to do is drop power consumption by the IC. You shrink a transistor and you reduce the amount of power needed to drive it.
@@johndoh5182 You are correct about clock speed. However noise is generated by signal transition rise/fall time. If you decrease the rise/fall time, while maintaining the same clock speed, more RF noise will be generated. This is in general why, when designing a product, it's a good idea to select a technology that customarily runs a the speed you intend to run it, and not a technology that's intended for 10x faster, for example. Also why MCUs and SoCs have selectable output current levels on their IO pins -- so you can limit the rise/fall rates.
@@Graham_Wideman I'm going to say BS because it's the same process nodes that run in laptops where the clock frequencies can drop to VERY low levels to save energy. Link something that shows Samsung or TSMC has an issue with slow clock speeds on N7 - N3 and I'll believe it. You get issues with noise typically with HIGHER frequency because current increases. Maybe you're talking about something external to the IC. In fact CPUs are constantly changing frequencies in cores and they can drop to very low levels, like a regular CPU that goes into a PC. It's how a CPU saves power. But I do believe that in laptops the clock speeds drop lower than in a PC.
But I also did say in that comment that smaller nodes can be more susceptable to external noise. But, that's not like a tiny bit of noise. After all there's a lot of magnetism going on in a PC with power supplies that can push 1600W along with having banks of mechanical drives. Fan motors.
So link something or give search info and tell me the name of the web page so I can read it. Maybe you're thinking that there is some general rule about what happens as nodes shrink. And I'm going to say engineers solve those problems which is why you HAVE those smaller nodes. I means there's a LOT of Samsung die that doesn't run at high speed because it doesn't need to. They have to clock a lot of ICs slower to save battery power.
@@Graham_Wideman Here, a product that uses TSMC N7 other than HPC
anysilicon.com/wp-content/uploads/2020/06/Movellus-HPDPLL-N7-Datasheet-1.pdf
@@johndoh5182 It's not clock rate you have to worry about; it's edge rate.
Let not forget that the tooling need in the fabs also use microchips and the microchip shortage is hindering supply of the material for that tooling. So, even if the fabs are built, the fabs won't have the tools to put into those fabs. Generators, power supplies, voltage regulators, etc. are all needed to build the tooling needed in a fab, and those need microchips, which, there isn't any.
The cost problems that the executives are talking about in those docs could be eliminated with a simple solution: eliminate the executives, and redirect their grossly inflated salaries into manufacturing.
Yeah, no. That’s orders of magnitude less money than what is required here, you’re talking about a few people making half a million a year vs a billion dollars per quarter to rapidly scale upwards like that. I’m sorry but the real world isn’t so easily dissected and solved, and those figureheads play a very important role for the structure of successful businesses.
In classical economics there is never any shortage or scarcity of any commodity, because price mediates between supply and demand so that they match each other. For example, if chip buyers want to buy a hundred million units of a chip, and the manufacturers are only able to make fifty million units, the price of the chip should rise to a level where the buyers only want to buy the fifty million units that the manufacturers can make. And, if a chip manufacturer runs out of its product, that means that they have simply set the price too low.
We may live in the Sonoran desert her in Phoenix but... there's a reason the metro valley is reaching a population of 5mil. We know how to handle water.
The new FAB plant is being built a couple miles north of the Central Arizona Project. The CAP is a canal that runs for over 150miles from the Colorado River by Parker AZ, down to Phoenix and then Tucson. It's primarily use is agricultural. Less than 15 miles to the north west of the FAB is Lake Pleasent. That's a reservoir of over 1 mil acre feet of water. No river runs freely in Arizona. We have multiple reservoirs in the mountains where we collect winter snow runoff to supply drinking water. Much like California, only we do it better without all the Democrat corruption and mismanagement.
I've been here over 40 years, watch the Valley quadruple and we've never had to ration water.
So... pretty sure the FAB will be just fine.
Maybe there is mismanagement and corruption nearly everywhere, and rather than enabling it by doing the corrupt politicians' campaigning for them we can pull our heads from our tribalistic asses & acknowledge that fact? Or I suppose we could not do that and let ourselves fall deeply under their sway, to the point we can't talk about things as basic as water without signaling our loyalty by shoehorning their talking points verbatim into every comment conversation and quip? Nothing screams individualism like millions of ppl using another person's words to express another person's ideas every time they open their mouths. Nothing exudes freedom like feeling it necessary to constantly reassure others that you're one of them. But go on, tell me all about how you came to believe what you believe of your own volition, putting 2 and 2 together after you got a hankering to audit the financing of water works in a small town in California. How having such a blast on that audit lead you to do the same for the next town over, and the next then the next until before you knew it 2 and a half years had passed and you'd performed an audit of every municipality in the state.
Also, might want to look at the trends re; the flow of that river over those 40 years. Definitely take an gander at the rate of replenishment vs usage. Might also be a good idea to brush up on the demographic and economic changes up stream and maybe also the state of water rights and what outcomes that might incentivize. Oh, and take a look at the situation down stream too, in Mexico.
Or believe whatever makes you feel secure and comfortable regardless. Can't see that ever becoming an issue. Never has before.
Water Wars Next stage of water restrictions in Arizona could come as soon as August
Officials with Central Arizona Project and the Dept. of Water Resources warn the state needs to do more to conserve water to help stave off even more restrictions.
Author: William Pitts
Published: 6:17 PM MST May 6, 2022
Updated: 7:40 AM MST May 10, 2022
ARIZONA, USA - Arizona water officials now predict the West will enter the next stage of drought restrictions as soon as August.
A Tier 2 shortage declaration would mean even more cuts to the amount of water Arizona gets from the Colorado River, but not nearly as much as the Tier 1 shortage declared in 2021.
Still, officials with the Central Arizona Project and the Department of Water Resources warn the state needs to do more to conserve water to help stave off even more restrictive cuts.
“Most of the water use is outside the home,” CAP Deputy Director Ted Cooke said. “It’s in the yard, it’s pools, it’s plants, it’s lawns.”
Cooke said there’s a lot of room for conservation just by having people cut their home water use.
The Federal Bureau of Reclamation declares water shortages, which trigger water cuts based on previously approved agreements.
In 2021, the Bureau of Reclamation declared its first-ever water shortage, cutting more than 500,000 acre-feet of water going to Arizona. An acre-foot of water would generally supply three average Phoenix households with water for a year.
A Tier 2 shortage, if it’s declared, cuts an additional 80,000 acre-feet of water to Arizona.
“How that’s going to impact the homeowner is…they’re probably not going to see much of an impact at all,” Arizona Department of Water Resources director Tom Buschatzke said.
Lake Mead and Lake Powell are at record low levels, thanks to a 22-year-long drought in the West.
RELATED: Lake Mead water levels hit lowest point in 51 years
The Bureau of Reclamation announced this week that it would keep another 500,000 acre-feet of water in Lake Powell just to ensure that Glen Canyon Dam can stay functional.
I can barely grasp pound per feet to express torque, instead of the far more practical Newton per meter units. But acre-feet makes it even harder to understand 😅
@@theagentsmith torque: foot pounds, not pounds/foot. Pounds per foot is how much you pay for lumber in the UK :-)
I am subbed and I don't always watch new videos, but every time I do it's very informative. Keep up the great work!
Manufacturer: Oh man, we really need to order some more of these chips.
Chipmaker: We don't have any of those in stock.
Manufacturer: But we need them for our products.
Chipmaker: This chip is old tech, that's getting more expensive.
Manufacturer: How much is it going to cost to make them?
Chipmaker: Nah, you don't want to pay that much.
Manufacturer: If the choice is between selling fewer products because we had to raise the price, or selling no products, we'll take the more expensive chips. Just make the damn chips and we'll buy them!
Chipmaker: Ehhhhhhh we don't wanna make these old chips. It'd take a lot of work and we'd have to raise the price. How about we just don't sell you those chips, and we'll make less money?
Manufacturer: _*#%& @^#% #$&)!_
Next day...
The sun has a major solar flare and mass ejection with earth in it's direct path and f#cks all y'all over in the next Carrington event
accurate af. 😂
Note that more modern machines have also way higher throughput. An older ASML 200mm machine can handle 100 wafers per minute, while a newer 300mm can handle more than 300 per minute. So, that 6 times more dies per minute. Although the floor space is larger. Finally on 300mm you can better utilise the edges than on 200mm.
*per hour, not minute. Those kind of breakneck speeds would open a wormhole.
The real chip shortage is in my Doritos bag right now
So I am from the Netherlands and what I don't understand is that if ASML is making a large portion of the world's leading chip making machines, why aren't we making those chips ourselves? What does Taiwan have that we don't have that make them the largest producers of chips?
sounds like PERFECT storm is brewing
with Recession in the Horizon and NEW mature tech factories coming up....
Price crash incoming.
during the 2020 pandemic it was not that paper products were necessarily in short supply, it was more of an issue of hoarding. A bulk of the chip "shortages" in 2021 and 2022 are caused by the CCP hoarding semiconductors...The CCP builds with impunity.
my local coles doesn't have any lays chips i'm pretty sure that's the real chip shortage
Excellent video, very informative for your subscribers!
Maybe Bitcoin/Cypto downtown in price help in reduce the demand (of certain chips) when mining becomes no longer profitable.
That has negligible effect on legacy chips, where the real shortage is. It will lessen the pressure on the cutting edge chips though, but it will not decrease actual manufacturing volume of these chips, just more of these products will be available to regular consumers. To that end, it will not affect the shortage as an industry wide problem affecting all dependent industries at all, or rather it will, but only after those lucrative markets are saturated, so years later!
Amazing analysis, thank you!
There is a severe water shortage in Arizona right now that is leading to water rationing for human use. The Colorado river, the main water supply for Arizona and the Southwest, is at historic low levels. If water levels drop any further, the hydroelectric generating capacity at the Hoover Dam will have to cut back production to maintain water levels.
Why on Earth are FABS that EACH require 9M gallons of fresh water daily being built in Arizona?!? Only about 60% of that water used by the FAB is recyclable. Building multiple FABs in a desert during a drought is just plain asinine. The people of Arizona are not like the people of Taiwan and will not tolerate severe water rationing just so a chip FAB can operate.
I cannot fathom this abject stupidity.
Yup, both Intel and TSMC are building many fabs which needs the Colorado river. Republicans are just straight stupid anything for money forget the people and the environment.
Taiwan is a very wet country
@Unknown Alien
There is not enough electricity to run desalinization plants in the American southwest and the ultra high cost of electricity in California makes desalinization uneconomical. Nuclear electricity generating plants are being shut down and replaced with solar panels and windmills which don’t work at night or if there is no wind.
If FABs are to be built, they should be built where there is an abundant fresh water supply (e.g. Pacific Northwest or Great Lakes areas). Building FABs in a desert during a drought is absurd in the extreme, courts ruinous financial failure, and is just plain stupid.
Both Intel and TSMC are building fabs there AND they are going to be LARGE fabs. In the case of TSMC they bought up the space to create a giga-fab, with 6 smaller fabs in a single location, just like they have in Taiwan in a few places.
Closed loop. The water gets recycled and used again. I read about what it is they're doing because I said the same thing to myself, which is that's about the most insane place to build ANY manufacturing that requires a lot of water, because that area is running out and that's not going to change.
I don't have a good answer for why this is and I'm not going to search for an answer but companies seem to like to build fabs where it's hot and dry. Maybe it's about what can survive in the air such as viruses and bacteria, I don't know.
@@jstasiak2262 They are going to recycle the water so they HAVE to be running some type of filtration. Go read about these plants that are being built.
As far as electricity, if stupid ### Biden would lift tariffs on ALL solar panels coming into the US it would be dirt cheap for a fab in AZ where it's sunny 95% of the daytime hours to have all the electricity they need, INCLUDING the cost of grid storage. The company ESS produces iron redux flow batteries that use water (closed loop) and iron as the main ingredients and would be very suitable for this situation. For large companies that are buying thousands of panels, the cost of panels is very small to the point that it out-competes natural gas. It's an upfront investment though. A company like Intel or TSMC will have the ability to pay that upfront cost, just like what Tesla does with their giga-factories.
The fab capacity that was freed by the Automotive industry cancelling most their semiconductor pipeline was not re-tooled, but instead used for MEMS and mixed signal chips?
I see what you did there, saying Kaoshiung instead of Taiwan. Got hide from from the Papa Xi's TH-cam censors. 🤣
Koashiung is a city in Taiwan
Your tinfoil had is overheating. This channel frequently mentions Taiwan.
thanks.
very relevant video.
Nothing sexy but it really hurts the rest of the industry with the shortages.
I wonder how many things are still using a 555 Timer 😆
I think the most used ancient chip might actually be a TL431. Obviously i'm not the first to voice such an opinion, but i have seen maybe two 555s in a product so far in my hands, but i have had dozens or low hundreds of TL431s in my hands.
The TL431, introduced around 1977, is certainly used in some current designs. But the LM555 predates thatm arriving in 1972. Perhaps modern-day designs would not use the exact same model, mostly adopting SMD format. As for being the most ancient still used in new designs, I'll suggest as candidates some 4000-series CMOS logic chips introduced around 1968. Around the mid 70's, there was a transition from the "A" series to the "B" suffixes, (I believe the B series added a buffer stage to the output, and possibly better ESD protection). But again, these days you would choose SMD variants.
I think we need graphene copper boards and transistors . i dont know what introducing such an alloy would do to lithography but the lower resistance and higher overall conductivity will certainly help something. If not just power delivery and heat dissipation.
I'm glad you made a slide pointing out that TSMC is building a new fab in a region of the US that is RUNNING OUT of water. I don't understand why no one is sounding the alarm about this. Lake Powell (located in Northern Arizona) and Lake Meade (Arizona-Nevada border) are RUNNING OUT of water. Not only that, the electricity that Lake Powell generates is drying up as well. Add to that an increasing population that is escaping the nightmare that California has become, and you get a disaster. When did we become so insane?