The chip shortage keeps getting worse – why can't we just make more?
bloomberg.com
bloomberg.com
Thus, if it had not been the pandemic, it would eventually have been something else. Sooner or later, there is always a shock to the system. If you have made sure that every step in a long complex chain is optimized to have very little slack, the thing you have optimized for is fragility.
Also, is this "fragility" really that big of a problem? Sure, prices are going to move up 20-30% for a year or two until the supply chain issues ease, but is that worth driving up costs for decades preparing for potentially the wrong issue?
If we did, more robust models might be financially attractive too.
I have never ever in my whole life seen prices going down after they have gone up. Ever.
But GP is correct, products typically don’t come down in pricing. Consumable goods are more elastic in a free market absent collusion, but finished honest-to-god products have a well-known price stickiness problem. (The price will come down, but rarely back to what it was.) Typically the only way that cycle is broken is when the product itself is obviated/supplanted by a replacement (eg an iPhone will never get cheaper but some new phone may come out that is cheaper and everyone moves to it).
These are all obviously generalities though.
[0] https://venturebeat.com/2019/09/10/apple-cuts-iphone-8-and-i...
It’s supply and demand. Prices fall if there is a surplus.
I've seen every computer part come down in cost over the years even after briefly becoming expensive through my life that recent trends have been odd to me.
C’mon meow. Hyperbole helps no one.
It's impractical to keep 12 months worth of supplies for manufacturing, even when you don't try to do lean manufacturing - just plain old manufacturing. It's something you need to finance, parts getting obsolete and product designs in some categories iterate every year or two.
We just need to deal with this new reality. You can do your best to avoid the worst case scenario but it is after all a statistical inevitability, so you just ride through it.
As I said on another comment here, though, this is going to have long standing effects on the other end of the disruption. We're going to be left with huge over capacity, unclaimed stocks and companies that couldn't produce parts will have their customers design out their parts for whatever else they can get their hands on. This is a golden opportunity for the smaller players in the industry. How giants like ST, TI and NXP will deal with this long term - I'm not quite sure. Many companies in 2021 and possibly well into 2022 who would normally design their products with TI parts will not do that because they can't even get samples, let along ramp up production.
0: How Toyota Steered Clear of the Chip Shortage Mess
Most of automotive does not need AMSL 10nm and below fabs. They need robust 200nm to 300nm chips.
This may yield car redesigns with less touchscreen dependence, simpler electronics in the essential systems, and an "infotainment" system that's less integrated with the vehicle and can be added or replaced later. Already, some new cars have shipped with a blank plate in place of the entertainment system.
There's a mindset that electric cars have to be more complicated. This is strange, because managing the batteries and motors is far simpler than managing an IC engine.
[1] https://www.programbusiness.com/news/toyota-succumbs-chip-sh...
I'd personally see that as a major improvement.
The highly integrated infotainment systems are going to make a lot of these vehicles feel obsolete before their time.
Most of the pieces are already there. The only problem is that they're proprietary.
I'm currently designing with STM32 parts and bought the last hundred available anywhere in the world, but that's only enough for a field trial, not for production. The only ones left in any quantity at at all are the insanely high pin count BGAs and chip-scale packages... no way.
I looked at TI, NXP and Microchip parts too. There's plenty of older 8-bit microcontrollers available, so if you want an AVR or similar I think you'll be fine because they're on older processes that aren't in as much demand.
If there really is an alternative from a "smaller player", I'm all ears.
Definitely shortages on regulators as well, although that's outside my area.
They didn't need to. The initial disruption was only a few weeks. The problem is that once companies saw that there would be a shortage, they started buying up all the supplies they could to rapidly build up buffers, which increases the demand for those supplies worsening the shortage, which in turn means even larger buffers are required. It's a vicious cycle that all could have been avoided.
* * *
>"Just-in-time" means "no buffer"
No, it doesn't. It just means the buffer is somewhere other than where you assume it's necessary. That may in fact be a net gain. Computers and cars and whatever don't wear out instantly. It will be a long time until all the newest equipment that already exists is old enough it absolutely has to be scrapped. In some other thread on HN, someone said that everything depends on manufacturing, and services don't mean anything without manufacturing capacity. Services! Like maintaining and repairing what's already been manufactured! Like, for an extreme example, what happened in Cuba when they couldn't get new cars for a long time. Or during WWII when civilians couldn't buy new cars. Day in and day out, people talk about how terrible a disposable society is, and then we have a crisis that requires like 1% less disposal and it's the end of society.
There was a time when every part of a computer - including, literally, an individual "bit" of ram, could be hand-repaired, craftsman-style (and even visually assessed for failure), but we've been on a long, long trajectory towards none of this stuff being user-serviceable.
We're now watching the rise of SoC, wherein my graphics card and my ram crawl into my CPU and disappear as discrete components. Strange times.
But that doesnt mean all our existing tech will emit the magic smoke in the next few months. The endless need for more computing power is mostly driven by software bloat. We could extend the current smartphone and gadget usefulness by a significant amount of time if we stopped to fall for constant featurism and did lets say, a year or two of cleanup. Sure, "the future" like AR and such will have to wait, but heck, this "we will sell you 3d" thing is around since the 80s and didnt take of in the mainstream yet, so what?
Really, what nukes the repairability of a lot of products these days (ignoring the repair-hostile practices of many companies) is the cost of the item vs the time needed to disassemble the unit, track down where something went wrong, repair it, reassemble, and hope that component didn't blow because something else upstream is faulty that you didn't find. Cuba and Africa the time is cheap while replacement units are somewhere between expensive and unobtainium, USA/Japan/Europe not so much.
As a fun example, I’ve been simulating core memory at the level of electric/magnetic fields lately. It takes literally trillions of floating point operations to simulate just 1 nanosecond of a single bit of core memory. In doing this, I can identify things like “if we were to arrange the cores like <this>, we could reduce the peak magnitude of the radiated fields when a core flips, thus we could get by with using slightly softer ferrites (which use less energy when switching state) without encountering noise problems.”
But of course, a trillion fp operations was a lot more back then! Having more advanced tech made it substantially easier to optimize the earlier tech. Optimizing compilers are maybe a more direct, but limited, example of this. The faster a CPU can run your code, the more cycles you can have your compiler spend making that code run even faster. I’m not sure there’s a good term for these “mutual feedback loops”, but they seem to pop up a lot.
I hear that all the time. Usually, in the same paragraph, they'll also ask for Feature X, Y and Z.
I've just replaced a Blackmagic Atem Mini: sadly, I tinkered with my previous one trying to keep it alive and couldn't return it. I burned it up trying to do too much with it: ran the internal processing at super HD from a Pocket Cinema 4K, with two separate other inputs coming in and having to be rescaled, color corrections, and then the output downscaled after chromakeying to HD.
The unit basically melted itself until all it could do was flash bright colorful lines. Total loss.
So the replacement (dropping right into the previous workflow, though I'm making a point to not run as many rescaled inputs by default) will sit on a homebrewed heatsink with thermal pads and fan. That way it probably won't burn up, but if it did I won't have taken it apart, and I can most likely return it for a replacement.
The inability to repair is a concern, but on the other hand building the larger environment into the cooling support the unit should have already had, is the DIY aspect. Should work.
https://blind.guru//blog/2016-06-12-brlpi.html
https://hackaday.com/2019/03/26/not-all-raspberry-pi-laptops...
I sold my stockpile of laptops (anything i5 or newer) for triple it's value last year. People would buy anything at that point.
I imagine the downvotes were for implying that everyone else is in a similar situation as you. What about people who were being efficient before the shortage, and now can't keep their aged equipment running any longer? What about businesses who suddenly need much more equipment than they did previously, due to pandemic restrictions?
> How is the market for used equipment doing btw?
I've found it much harder to find available supplies at any given time, but I haven't actually seen prices going up as a result.
Creative outlets for junk are going to continue to become more and more fashionable as we struggle to meet our values and practical prices.
One of the issues is that repairability and/or durability have also been decreasing for many categories of goods.
> It just means the buffer is somewhere other than where you assume it's necessary.
The rest was an example of one place you might see such a buffer. In practice, there are many such buffers, of varying capacity and time duration. For transportation, there’s the households who have 3 cars but could get by with only 2: they’ll sell their extra car once the price makes it worth their while. There’s the cyclist who’d like to have a car for out-of-city trips, but decides to hold off until inventory catches up (a demand-side buffer, if you will). Many more “buffers” could be imagined.
And, well, the market libertarian might feel really satisfied over the last year of commodity prices. Sure, supply constricted and prices rose, but at no point did we see any mass crises where people became stranded somewhere because they had no transportation option. For the vast majority of people, they just had to make convenience/cost tradeoffs that were a little bit different than before.
You are just not paying enough attention. As prices rise the poor become poorer. How can you possibly claim that nobody is stranded by the increase of prices?
But, no. I haven’t heard stories of people being stranded in mass. I meant that word literally — because of the context of transportation — if that wasn’t clear? Even in the days of mass lockdown: I saw huge lines at food banks in suburban and even rural areas, but even those who are in a position to wait four hours for a few days of essential goods seem to have still found ways to get there. That observation is subject to some obvious biases, and yes I’m sure some people did fall through the cracks, fatally. But no: I truly haven’t witnessed what I would consider mass crises caused by people losing access to transportation. Despite massive shocks to a system of JIT manufacturing, there wasn’t a collapse of the supply side in the sense that you would have expected had you looked at inventory, divided by the rate of consumption, and thought “that’s how many weeks it’ll be until anyone who doesn’t own some form of transportation will be physically stranded”. Whatever services and goods existed two years ago, equivalents can by and large still be found today, albeit at a possibly-increased price or a possibly-decreased utility. They may have dimmed, but “the lights stayed on”. Is the world free of problems? No.
What is being missed are the larger societal needs for new hardware. If you build infrastructure (new public transport line, factory, power plant, communications network, etc) you need new components, lots of them.
The impact of shortages gets projects delayed or cancelled due to lack of availability of components or ballooning costs. It's also not just electronics, it's also shipping and raw materials costs that are impacted.
It's just companies anticipating or seeing the same thing (an increase in consumer PC demand) and all releasing their own shitbox/AIO/desktop/laptop.
I bet most of them are sitting in warehouses unsold.
The pandemic allowed tens of millions of well-off western office workers to move out of tiny urban flats into more spacious housing. While having to spend a lot of time in their new homes, which was previously in ”public” places (eg restaurants, bars, what have you)
Now, what to do with all that space? A giant tv, washer and dryer, a nice stereo set, new induction stove, the list goes on and on…
PC demand also increased obviously, but probably bitcoin miners caused the most demand for PC hardware components
Just in time doesn't necessarily mean no buffer, the goal is to minimize excessive stockpiling and keep enough for continuous production, increasing stockpile when required.
Toyota, some might say THE pioneer of Just-in-time manufacturing, was one of the car manufacturers least impacted by the component shortage precisely because they started stockpiling very early on after they saw the upcoming issue.
Here's a quote from another Bloomberg article specifically on Toyota[1]:
"Toyota asks its Tier 1 suppliers to input detailed information about their most obscure parts and materials providers in a complex database that it maintains. Using this system to glean information about, say, a single headlight Toyota purchases for one of its cars, it can get information as granular as the names and locations of the companies that make the materials that go into surface treatments used on those headlights’ lenses and even the producers of the lubricants used on the rubber pieces in the assembly, Toyota spokeswoman Shiori Hashimoto says.
These lines of communication alerted the company early on that it needed to stockpile chips."
[1]: How Toyota Steered Clear of the Chip Shortage Mess - https://www.bloomberg.com/news/articles/2021-04-07/how-toyot...
The average American manufacturing company is run by a CFO operating with reports out of SAP or Peoplesoft. Their performance is measured by fiscal performance, so running to the penny and having no inventory benefits them more than making the company resilient. Wall St rewards quarterly performance, not resilience.
Private profit, socialized risk.
Maintaining resilience and designing antifragile systems usually pays off in missed disasters and unique opportunities.
Alternately, to save money without buffers is to continuously gamble.
For example COVID style shutdowns could be reasonably modeled at X% per year after talking to a disease specialist. But, not all of them are going to be from diseases.
Look at companies like Boeing with absurdly complex supply chains. The inventory numbers look good, but the factories are idle when a truckload of magic bolts is stuck in a blizzard in South Dakota.
It also increases actual cost. I supported a GE business unit as a supplier for awhile. We hosed them for stupid last minute orders due to this sort of thing. They would pay more for expedited shipping, overtime, waste money on leasing stuff to avoid capex, etc.
According to this, Toyota cut production by 40%, though it does acknowledge that Toyota took less of a hit vs industry:
“New cars often include dozens of microchips but Toyota benefited from having built a larger stockpile of chips - also called semiconductors - as part of a revamp to its business continuity plan, developed in the wake of the Fukushima earthquake and tsunami a decade ago.”
Also, Toyota is now cutting 40 percent of global production due to chip shortage.
https://qz.com/2004569/the-global-chip-shortage-can-be-expla...
He says Toyota is one of the few examples where just-in-time was not distorted and warped. So Toyota implements true just-in-time, which can be robust, with buffers. But most other companies implement a half-assed version that is very fragile.
My favorite was tower advising them they’d be fined for a departure during local curfew. “Roger, we better get our money’s worth then” and did a max performance takeoff and turn on course with whatever parts they were carrying.
Also, my Dad once had a late-ish flight to Long Beach that got diverted to LAX because delays pushed it past the curfew.
https://crankyflier.com/2020/10/08/jetblues-two-decade-run-i...
I’m not sure it does much to keep the airport quiet as I’ve been in a holding orbit of three low-altitude airplanes from 6:45 AM to 7:00 with all of us waiting to avoid the curfew fee.
So, instead of slipping relatively quietly into the field from a random route to a straight-in, neighbors got to hear me orbit for three rectangular patterns over the same ground points plus two other airplanes doing the same low-altitude holding. If I were a neighbor, I’d much rather the simple straight-in to be encouraged.
The sports teams all just pay the fee; I never had to break the 11PM side while I was based there, but I would have as well. (It was less than an hour’s worth of gas, so it wouldn’t make sense to divert and reposition later.) I suspect the curfew makes people feel good while making the actual experience worse.
https://www.infrastructure.gov.au/aviation/environmental/cur...
Arguably lean manufacturing is an anti-pattern.
When taken to the extremes that modern business has taken it to, it is 100% an anti-pattern.
One has to optimize to reduce fragility, but when massive chip (or anything else) shortages start to appear for extended periods it is obviously already way to late. And way too downstream if your strategy was just a punctual stockpile decided by a single company. Because that probably would have been impossible for everybody to apply this punctual strategy at the same time...
This was in part because the domestic Japanese market was so small. Of course, it needs support from quality development and manufacturing. But it's also efficienct because it's much cheaper to sell something to someone who trusts you than to get it out by cold calling.
(I have no idea if it is common or not, just that this doesn’t contradict that claim)
https://businesswales.gov.wales/sites/business-wales/files/S...
https://www.theverge.com/2021/8/31/22648372/willy-shih-chip-...
TLDR: it essentially was the pandemic: cancelling orders -> redirecting supply -> too late to uncancel + competitive hoarding.
https://www.theverge.com/2021/8/31/22648372/willy-shih-chip-...
TLDR: it essentially was the pandemic: cancelling orders -> redirecting supply -> too late to uncancel -> low supply + competitive hoarding. Plus a 10 year $150B head start.
Personally I have been ranting about the Modern use of Just in time manufacturing for well over a decade. Just in Time manufacturing as it originate in Japan doesn't actually means no buffering. Somewhere a long the line in typical Chinese whisper fashion the true meaning got loss when it go to the west. If you are software developer, just like at Agile and Kanban board.
The problem make worst when CEO and COO dont actually understand this, and force those in supply chain to comply with their view of JIT. It works in a sense when everything is normal. It doesn't work when there is a shock. Look at Apple, the only company that took JIT to a level beyond Toyota. They are doing just fine while others are fighting for parts.
The other thing worth pointing to is that this isn't just chips, but also every other commodity. The reason behind all these different industry are exactly the same. You could have swap the title for Beef, Poultry, Steel, Toilet Paper, Mask, Chips, Milk, Pencil, paper etc... I really do wish people learn a little about how supply chain works. It is the fabric of our daily life and yet very little attention has been paid to it.
This is a balance point, not an optimization problem where you're supposed to go to the extreme.
In fact, one of the books that popularized Lean manufacturing, Eli Goldblatt's "The Goal", explicitly makes the point throughout the book that making work centers within a plant locally optimal will cause your overall operation to fail to deliver efficient outputs.
To make global efficiency improvements you have to accept that individual work centers may be inefficient. Rather, just like a developer optimizing their software program, you need to look at your entire organization and find where the actual bottlenecks are, and improve those specific bottlenecks, potentially by running "inefficient" (but non-bottlenecked) work centers to assist the teams that are the current bottleneck.
And then as you change things you'll have to be willing to address a different bottleneck.
Moreover, unavoidable variances (which the book describes in a story about a Boy Scout camping trip) means it will be potentially disastrous to 'optimize' even your critical path, just like you wouldn't want a network router to be consistently near max capacity.
To be honest I would recommend a separate book for software practitioners, Reinertsen's Principles of Product Development Flow, because it gives you methods to look at which are more applicable to creative work where you can't easily turn your work into "production lines" to then use Lean manufacturing/JIT principles on.
Either way none of this works if you let every local work center just produce produce produce up to its peak output or most efficient point. At that point it becomes impossible to find the bottleneck because there are so many bottlenecks (including bottlenecks not on the critical path). That's a key reason why all the methods (Goldblatt, Reinertsen) focus on reducing the amount of work-in-progress and reducing batch sizes of each individual work package or work item.
And what exactly is a big stockpile of raw materials and components that aren't intended to be delivered to the customer under normal circumstances?
It's important to remember that in traditional manufacturing at the time, each work center wasn't graded on its ability to contribute to customer sales, they were graded on things like local efficiency on a per sub-assembly basis. The "customer" wasn't the end user, the "customer" was the plant manager or the PMO or some other stand-in.
The work centers were structured around functional tasks like "heat treat" or "machining" rather than an end user product, so your "high priority" customer product had to make it through multiple independent backlogs before it could be shipped, even in theory. God forbid rework would be needed! Quality issues were the job of customer service to address, not the work center that introduced the defect (as after all, it might be years before the sub assembly finally ends up in a user's hands).
Turns out it's Goldratt. Very difficult to read as it is in the form of a melodrama. I suppose the point was to make it light reading, but to me it's a difficult slog through the protagonist's personal life to get to the data I want.
What is important to discuss is that no one was paying for a extra idling factory before, and now suddenly it is needed for one year to satisfy every customer wanting to keep a one year local buffer.
One also have to take the effect of a one year buffer for the whole supply chain into account. The demand for the manufactured good also increased due to their customers wanting to keep the same one year of buffer, so the one year buffer (based of your previous demand) would be consumed solely by your customers buffer and your orders would have to be larger than one year of inventory for you to keep one year of inventory after the demand spike.
In the end it is easy to imagine that new buffers would lead to a huge scarcity, especially for the longer supply chains.
One of the reasons is they hide feedback about downstream capacity problems. (See "bufferbloat" for this in internet terms, Principles of Product Development Flow for it in terms of creative work, and any book on lean manufacturing for it in terms of repetitive work.
I mean would we all like the base model Ford Focus to cost $45,000 simply so that they'd be available on the lot during a once in a century pandemic?
By raising prices on the high demand stuff, you send a signal to the people that can get by with alternative stuff that now is a good time to do so.
So, like, highly hypothetically, if I couldn't find a calculator for a good price, I'm in a good seat to whip out my slide rule. This gives me an edge over the people who have to pay high prices for calculators, and simultaneously smooths out demand. What's good for me is also good for the people who really need calculators.
Optimally distributing fewer goods isn't a solution to the problem. The problem is a shortage.
Prices go up on the food most people prefer to eat, first. People who cannot afford the foods they're accustomed to eating will switch to lower-priced foods they would never have considered otherwise.
That's largely how tomatoes, potatoes, and more came into use... Desperate, hungry peasants.
When food suppliers see the increase in price, they will work hard to squeeze more yield out of their existing supplies, and as soon as possible will invest in producing more (typically for next year). Others who are capable of producing food, but normally produce something else due to economics, will see the equation change in favor of food and switch their output.
Famines are not caused or encouraged by supply/demand economics. It is vastly more common for communist governments to cause them.
That number isn't a constant, it is a function of supply and demand. Therefore there is no such thing as a chip shortage. We'd maybe like prices to be higher or lower depending on our perspective.
> What happens when the good in question is food?
If there are 10 mouths and 9 meals then there is a food shortage. But that is a biological observation, not an economic one.
It does take some time for prices to adjust, and if the supply and/or demand curves shift very quickly, there absolutely can be a temporary shortage. I think that's what we're going through now.
It isn't like COVID is a surprise any more, it has been with us for nearly 2 years.
Apart from the Toyota approach, discussed, Dell pioneered a different system: they forced "SLA"s on suppliers for whom Dell was a major customer. This was back in the mail order build-to-order business of the 90s; I don't really know what Dell does these days.
It worked like this: Dell took a truckload of parts. They hit the loading dock; were (in the finance sense "received"); Dell accepted the invoice (to be paid 30 days later or whenever) and the goods appeared on their balance sheet. They used them in computers they built, sent out, and then charged the customer's credit card for, getting paid immediately. When they ran low on part P they called up the vendor and said "deliver us some more right away". What was "right away"? Why pretty much immediately; the vendor had them in a truck outside Dell's facility. If they didn't, well, they stopped being Dell's vendor.
Pretty clever: the contents of those trucks were still on the vendor's balance sheet until Dell asked for them; Dell got to use them and get revenue for them long before they needed to be paid for, but basically not before they were needed. They made so much money on the float, and all at their vendors' expense.
They did the opposite too: if they had inventory ready to go but without a customer (prebuilt configs) they put it in trucks and drove the trucks away from the docs. I don't know how they got their auditors to accept it as GAAP but they did and were lauded by Business Week, the WSJ etc for their cleverness.
No, that's a degenerate fragile microoptimization inspired by seeing the short-term cost savings of just-in-time and going all in on those savings without understanding just-in-time.
What “just-in-time” actually means is dynamically exactly enough buffering as is cost effective given current expectations of forward conditions considering the likelihood of supply irregularities, cost of production interruption, costs of inventory, etc.
I can definitively confirm that just-in-time is about trying to optimize the amount and strategic positioning of buffer needed to support the "right amount of production" to meet anticipated and actual demand, exactly as you describe.
Now, doing that can be rather difficult in the case of black swan events. But the goal is as you describe.
"Exactly enough" means "no buffer". You are simply mislabelling inventory as buffer. What people needed was buffer beyond what is cost effective given current expectations because something unexpected happened causing improbable supply irregularities.
“Exactly enough” means nothing by itself.
“Exactly enough for X” means different things based on X.
“Exactly enough for current production” means no buffer.
“Exactly enough buffering as is cost effective given current expectations of forward conditions considering the likelihood of supply irregularities, cost of production interruption, costs of inventory, etc.” does not generally mean “no buffering” unless conditions are expected to be extremely stable, or response time to changing conditions from identification to new deliveries are near instant and there is negligible risk of that changing, or there is negligible cost (including opportunity cost) to production interruption.
If you expect some supply irregularities and you don't have enough to compensate for them, that is less than enough - a shortage. Having enough for normal operations (which includes forseeable irregularities) is no buffer. A buffer is what you have in excess of that.
HBS is even realizing too much optimization/efficiency is a bad thing. The slack/margin is squeezed out and with that, an ability to change vectors quickly. It is the large company/startup agility difference with the added weight of physical/expensive manufacturing.
The High Price of Efficiency, Our Obsession with Efficiency Is Destroying Our Resilience [1]
> Superefficient businesses create the potential for social disorder.
> A superefficient dominant model elevates the risk of catastrophic failure.
> *If a system is highly efficient, odds are that efficient players will game it.*
Over efficiency and supply chain concentration single point(s) of failure (China/Asia with chips especially across Hong Kong and Taiwan that make most of the chips) caused supply chain disruptions, some hoarding is also going on with chips. [2]
The HBA MBA-itis and Chicago style thinking of excessive efficiency made the players that can compete larger and less of them, that takes away resilience and leads the market to gaming. The market has thus been gamed as the players that control it are more powerful in terms of market leverage. When you give up diversification and flexibility you get leverage and added weight to any needed quick changes as the large players attempt to control the market. [3]
US is now building up silicon supply chains again [4][5], it was much bigger in the 90s/early 00s with Intel/Motorola/etc additionally, but business leaders allowed a concentration to happen and it led to market leverage.
Hopefully that same mistake is not made in the future. It will take time to build up diversification of market leverage in terms of chips for availability.
This chip shortage, and all the supply chain problems during the pandemic as well, will hopefully introduce more wisdom and knowledge into business institutions that just because things are ok while being overly super efficient, that is almost a bigger risk than higher prices/costs. Competition is a leverage reducer. Margin is a softer ride even if the profit margins aren't as big.
[1] https://hbr.org/2019/01/the-high-price-of-efficiency
[2] https://finance.yahoo.com/news/china-stockpiles-chips-chip-m...
[3] https://www.bloomberg.com/news/articles/2019-12-03/china-sto...
[4] https://www.cnbc.com/2021/03/23/intel-is-spending-20-billion...
[5] https://www.cnbc.com/2020/05/15/tsmc-to-build-us-chip-factor...
I was talking to someone who was rebuilding a home, and the price of plywood went from $8/sheet to > $100/sheet
Another person said that the company that makes 75% of the electrical panels had stopped production.
those are just two pieces of the bigger puzzle, but since many if not all pieces have similar shocks, it would really surprise me if capacity could ever increase to meet demand.
This is true, but I don't see an antidote. The government couldn't just tell businesses to "be inefficient"!
Many of the major semi companies are running at fairly high capacity, but supporting elements like IC package build capacity, sometimes materials can't keep up as well.
We looked at some boards to redesign, but when you are short hundreds of parts it gets nearly insurmountable to manage a hardware redesign. Interesting and challenging allocation period, definitely worse than 2003-2005.
What are these immense amounts of passives used for?
Googling aroubd I see other stories of Apple buying all available capacity of 5nm manufacturing or glass or sapphire for smartwatch screens etc, here's one source for RAM tho (2009): https://appleinsider.com/articles/09/02/18/apple_buying_up_a...
Buying something in sufficient quantities that results in others not having it, but still selling out of it is not nefarious.
In principle, it seems to me that this shouldn't be causing component shortages — on the contrary, it should reduce planning uncertainty for component manufacturers.
If Apple has a contract to buy 10% of a manufacturers capacitors. But then material shortages drops the manufacturers capacity in half. Apple's share is now 20% and anyone without a standing order is SOL.
I don't have insight into Apple ,justthis is how contracts work in general
Imagine you sell widgets for $200 a pop. Widgets are hard to manufacture, production is limited. Someone else has 10k widgets that they want to sell for $75 per.
If you buy all the stock, people will be forced to buy yours, because of supply constraint.
200-75 = 125 is more income than if you price-matched and sold at 75.
Now why destroy them, and not just sell them back? Well, they're not necessarily identical to the widgets you make. Maybe they have the other brand's logo stamped on them. Maybe the datasheet doesn't quite match.
Either way, your customers might notice. If you just buy and burn, no one downstream of you sees anything, except for higher prices
Note also that the above sounds immoral enough to be illegal in a whole lot of countries, so I would advise you not do that, though I'm neither lawyer nor spiritual advisor.
E.g. if you can make 100k widgets for all of which you have customers at a dollar each, then in a supply shortage with limited alternatives you may find that if you credibly cut supply to 80k widgets, so that some customers will actually be left out, then they might be willing to pay two dollars per widget (so, 160k instead of 100k for that toy example), so you get more money for less goods.
Longer term this shutting out of companies not on the leading edge will result in other changes. Components fabricated integral to the PCB being one such change.
I hadn't heard about this, and it sounds interesting. Do you have any references?
I get the impression looking at the electronics industry from the outside over decades, that techniques seem to be moving from "level to level".
That is, you now see single silicon chips made like entire clusters of computers, with on-board network connections between surprisingly independent cores and modules. It's almost like a bunch of silicon chips all within one monoilithic silicon circuit board, with no external wiring.
Then, several of these chips are in turn mounted on a physical board that is "just" another silicon wafer, possibly made on a very old node size, something still measured in microns. But nonetheless, it's a full silicon chip in its own right, with high-tech chips riding on top of it like circuit elements.
And then the motherboards seem to be miniaturising too. I remember when designers hesitated to use chips with much more than 20 pins because it would need to many traces that were too fine. Now? Thousands of pins are the norm for mainstream processors. Soon? Maybe tens of thousands, or even hundreds, and I wouldn't be surprised if circuit components start getting printed directly into the motherboards, like you said you saw.
I remember going into a library and picking up an old edition of Scientific American, from the 1960s I think. It had a full-page ad from Texas Instruments advertising their record-setting integration, because they could pack ten (10!) transistors into a single chip. They had an "actual size" picture in the ad, so you could see for yourself and count all ten transistors. The metal layer wires looked thicker than I've seen on some small PCBs these days, which should blow your mind...
The most convincing piece of evidence for this IMO is that if you compare the states that ended relief funds and bonus unemployment early to try to combat this issue with those that let it keep going, there was no statistically significant difference between the two halves of the USA in terms of hiring shortages.
Though it's ending this week
Unflavored half and half single servings for coffee go early. The bulk half and half dispensers are usually empty these days. (The flavored side is often still available.) Then coffee lids. Then large coffee cups. Gatorade Zero often goes, and then regular Gatorade. The Power Aide seems to always be available. The open cooler boat with yogurt and sandwiches often runs low or empty.
Vicious cycle of staff shortages everywhere, thus unable to source, produce, deliver and stock at former rates.
EDIT: I'm not joking, the salaries for mostly any position in a bar or a restaurant in Spain tend to be a joke with long hours and shameful conditions. Paying in cash to avoid taxes is a common thing. Doing extra hours without extra compensation as law requires. You don't like it? Ok bye, you're fired, and we'll complain to the news reporter on TV, about how nobody wants to work any more nowadays.
Americans got a paltry few thousand dollars. So you certainly agree that if a single payment amounts to more than their shit job then that shit job isn't paying enough.
Many people who made 20-30k/year working full-time, were suddenly making close to 50k/year without having to work for close to 2 years.
and i know that a lot of the deaths have been in older people who may have already been retired, but they are often the childcare providers who enable others to work.
I was making a comfortable wage previously but this year most of my coworkers have left for other jobs and I'm on the way out too since the current job listings are close to double what I make currently. An eye watering amount of money is on the table right now.
But paying more is a zero sum game. To gain developers you have to be draining them from some other company and it will take years for new supply to come in.
Prices are up because you're starting to actually have to compete with major players due to all the remote work normalizing the playing field.
Paying more won't bring senior developers out of the void because there do not exist enough to fill all positions. If all companies expand at the same time they can not possibly all fill their positions no matter what they pay.
This was a problem that was forecasted by many heterodox shutdown detractors. These "shortages" are caused by our politicians and bureaucrats making terrible decisions unilaterally.
Intel doesn't make 10nm chips that you put in a car or bulldozer or telecom box that has to bake in desert heat. Some of their older process stuff, yes, but not the top.
What about COVID caused the shortage? Is it shipping speeds? Change in demand? Less efficiency down the supply chain due to health requirements? Trade changes?
The next time Bloomberg manages that will be the first time.
Completely, utterly useless for tech reporting of any kind.
- Blame Toyota with their Just In Time inventory management ideas.
- Factory closures, lockdowns, etc.
- Increased demand for certain items. Everyone's at home surfing the web, playing games on their computers, mining bitcoin?
- Monetary policies are creating excess (see bitcoin mining) demand.
- Shipping delays. Ports running at reduced capacity. Ships sitting at sea waiting to get access. Suez Canal blocked. Shipping companies have no spare capacity. Blame Toyota.
- Building new factories takes time, getting/building new machinery takes time, everything was set up to not have any over capacity (since that's loss) ... blame Toyota again.
- Big companies swallowed everyone so we don't have as much diversity of products as we used to. Probably related to easy money as well.
tl;dr Toyota, Greenspan, Covid ... Somewhat in that order.
EDIT: Should probably add the trade war between the US and China to that list.
Toyota had a stockpile [1] of semiconductors and are only now-ish running into the same supply problems everyone has been since March 2020.
[1] https://www.bloomberg.com/news/articles/2021-04-07/how-toyot...
> But the Tohoku earthquake’s aftermath pushed Toyota to increase flexibility, and the value of inventory Toyota carries has almost doubled since 2011. Speaking at a briefing in February, Toyota Chief Financial Officer Kenta Kon said as part of the company’s business continuity plans, it keeps as many as four months of stock for some crucial components such as chips. Toyota didn’t expect the semiconductor shortage to disrupt production in the near term, he said.
Saying blame Toyota is funny, but really all the execs who say a cost savings and took without consideration of supply chain problems are actually to blame.
1. Companies cut their orders expecting less demand and lost their place in line for orders.
2. Because of the rush to work remote so suddenly people needed everything from new computers for kids, monitors and accessories for a home office and in general a lot of stuff that used chips.
3. Because of Covid there were shutdowns at semiconductor manufacturers and even disruptions due to fire and weather.
Suddenly the entire supply chain went into shock, something that had been perfectly balanced between supply and demand was full of uncertainty. Now we have a situation where everyone is begging for chips and they cannot be built fast enough. Essentially the companies are behind on production and the demand is so high they cannot meet it. Everyone keeps predicting an end but it seems that manufacturers are still falling even further behind than they were before.
Most of the companies that cut their orders were auto and industrials that projected a prolonged downturn and didn't want to carry the inventory on their books. They predominantly use the components manufactured on less advanced process nodes.
Thus when demand for new process nodes was increasing and old ones were decreasing at the peak of the pandemic it made sense for pipelines to retool to meet that demand instead.
On top of that US-China tensions means SMIC and Hua Hong Semi are now banned for many US producers which is further constraining supply. Also creating asymmetry where China is experiencing lesser a supply crunch. However there is also trouble brewing there due to Trump forcing the Dutch to block export of ASMLs new EUV lithography tech to China. This means they aren't able to start building fabs w/EUV to fill the coming supply gap there either but instead are now having to pour billions into their own lithography progam. Long term this will be good for the market (but very bad for ASML most likely) but short term its not great.
Second, the quarantines combine with unemployment that frees an abused workforce from abusive conditions. Instead of being wage slaves forced to take any job what-so-ever workers are able to demand jobs that value safety protocols, pay well enough to live in an area (which is also extra insane right now), and have reasonable hours (enough, the right times etc).
Third, (rent) given a combination of working from home, closed 'entertainment' venues, and general quality of life concerns many have decided to flee the high rents of cities for suburban and even rural areas where their money goes farther. This causes more desire to purchase and more wasteful food packaging and production at home as well. Everywhere around me prices have gone up on eating out something like 25% since the start of the pandemic; my wages have not. So instead of eating small portions of bulk shipped and prepared ingredients I'm now eating at home more, mostly frozen stuff. The price of rent didn't quite increase that much but housing is also insane right now, just like even used cars, due to the supply crunch.
Supply correction is how we as a culture, country, and world escape this problem. Increasing supply in all of the problem areas is the only way out.
Supply means buying power goes up, irrespective of wages. Those misguided attempts to raise everyone's minimum wage (tax the middle class, since the elite who own the rentable units will just raise their rates) would work far better by decreasing the cost of a better life. It could also be targeted towards desired resources, growing out of undesired patterns and standards which is also opportunity for better energy efficiency.
We could also just make the elite pay for it. Just saying.
The real solution is to densify and build way, way more. Why doesn't SF look like China with skyscrapers on every block?
Edit: it was legal or pseudo-legal until 1964 for developers to subtly (or not so subtly) advertise that their new developments would not be seeded with non-whites. It was a selling point.
If the rental rates were set correctly, none of the extra taxes can be passed on. The most likely result is the price of the buildings will go down and new renters will be able to buy in more cheaply but get less per rental.
Once again the Land Value Tax shows its brilliance.
I agree that LVT would make a huge improvement if it was set high enough though. But it will never happen when you have homeowners leveraged up 20x on govt subsidized mortgages who vote for themselves.
Take a look at "the beer game". One of the lessons from that game is that a single change in consumer demand causes a wide variety of supply chain disruptions. This lesson was learnt in the 1950s and unfortunately it looks like they stopped teaching this in mba schools, so we have to learn it the hard way all over again.
if you're not over-supplied, when a new spike in demand starts a whip through your factory, things can go nuts for quite a while. you get power surges. drones that were behaving beautifully before start flying in chaotic patterns due to under/over supply and making a bunch of dumb local decisions about storage boxes. stuff grinds to a halt for no apparent reason, only to discover that some random component has become a bottleneck for practically everything, and then it can quickly shift to something on the opposite side of your factory...
in that sense, perhaps the biters help keep your factory more resistant. you have to deal with occasional component-losses and spikes in weapon demands, and if you don't handle them well enough you can get into an awful persistent struggling state.
Yeah, it definitely shouldn't. But that's me, hacking together whatever I could to see that unsatisfying rocket launch and finally get the game out of my head. So I'd build something new and supply it from a steel conveyor that had a supply buffer, under estimating how much it would actually use. Then it would eat too much and cause a shortage, which meant other components somewhere else overflowed through lack of use, and overall progress would slow down until I ramped up steel production some more. That sort of thing.
So, ideally, everything would go smooth. And plenty of the time for me it did! But a tiny flaw could also be magnified through the system and suddenly I can't produce yellow research bulbs because I was two steel furnaces short of actual demand needed to for servo arms, and production would have peaks and valleys until the additional supply smoothed things out.
That depends. Either way, what I'm describing are my own observations in Factorio are those instances where shortages actually do end up causing bullwhips.
And shortages are not always easy to correct: A "mere bottleneck" is what we have in the chip shortage. Production capacity is the bottleneck, and correcting it is complicated.
That is what I also observed in Factorio: A small increase in demand that exceeds supply creates a bottleneck because production can't keep up. It propagates throughout the system causing other shortages where the components 4 or 5 layers down the chain need the scarce material. And also causes surpluses on multiple levels because components can't be used to make the components that also need the scarce material.
Fixing the shortage isn't always easy. If you need a lot of a specific material to build the production capacity to fill the shortage, that may reveal yet another bottleneck in producing another material, and so on.
IRL, it's possible something like this could occur with the increased uptick in fab production. ASML might be a bottleneck in providing drastically more equipment for more fabs. Or someone up the supply chain that feeds them resources could be bottlenecked. I guess we'll see how it plays out though.
A lot of Factorio newbies play with a "Spot bottleneck. Increase production. Repeat" mindset, though. The bullwhip effect absolutely shows up here, particularly if you're sharing intermediate products on a main bus. You start constructing purple circuits; that means your red circuit production is no longer adequate and you've starved blue science. You make more red circuit factories; now you're not constructing enough green circuits, and your whole factory has ground to a halt, but the red circuit buffer is filling up. You increase green circuit production; now you're starving the rest of your factory of iron and your steel production is lagging, dropping purple science production.
(And not limiting boxes)
It's just some parts can't keep up and everything after goes as fast as it can with the supplies it gets, running pretty steadily. It's what you want to happen when supplies are low, for the most part.
why do you believe this? we worked through that simulation as part of my mba program some years ago, and mba friends at other schools said they did the same. the bullwhip effect is such a core concept to operations management that it's hard to believe it wouldn't continue to be taught, though perhaps with an updated simulation demonstrating its effects.
Giant stocks of everything started to vaporise overnight, all at the around same time last November.
I have a few STM32 reels of different models some of which I can now sell at $10 per MCU. A 100-fold increase in price, that beats any bitcoin.
The starvation of seventies, and hungry eighties have taught Chinese a thing, or two about shortages.
People in the ex-USSR countries had the same tendency to run hoarding stuff on every economic doom announcement.
And that is that semiconductor manufacturing has been chronically and almost criminally underinvested. The coronavirus pandemic was just a triggering event that brought the whole house of cards down.
The thinking in the financial world for the past 10-20 years has been that software is the hot, growing license to print money industry and actually manufacturing chips is a dirty, competitive, low margin, low return on capital industry best left to the Asians. And of the actual chip manufacturers the ones to be most favored were the ones making digital chips and the ones making analog and power semiconductors were the dirtiest most disliked commodity like companies.
Even within the semiconductor world, the financial world heaped money on fabless semiconductor companies like stock market darling Nvidia and did not give much respect to the companies that owned the fabs and made the chips. And even less respect was paid to the ones that made power and analog chips.
Guess where the biggest shortage is now. In power and analog. Also, in packaging, another field that was considered unsexy, low margin and too competitive to bother with.
Everybody talked about how software is eating the world and heaped massive amounts of capital on software companies but they kind of neglected the important role of semiconductors. And even when they talked about semiconductors they talked about the "important" ones, the "brains" of the computer, the cpus and the graphics chips.
So while software companies (especially SAAS ones) would have license to lose large amounts of money and still have receive endless streams of capital, companies that manufacture chips would be judged severely by the market on profitability and free cash flow. This is especially true for analog and power semi manufacturers.
Free cash flow is a very dangerous metric for a growing industry. Capital spending gets subtracted from cash flow to arrive at free cash flow, which means that if a CEO is judged by free cash flow, he will try to minimize capital spending. That of course will hurt future growth.
I have been investing in analog and power semiconductor companies for many years now, and the CEOs of all these companies knew that a jump in demand was coming. But they all wanted to keep their jobs so they all talked about how they will grow their revenues by minimizing capital outflows etc. If they were cheered by the market for having cash outflows of the type many SAAS darlings have, we would be swimming in semiconductors right now.
For example, if some one wants to do some further reading, check out old investor presentations of Texas Instruments. By old I mean a couple of years ago, before coronavirus. Probably the largest power and analog semiconductor manufacturer in the world. Their entire sales pitch to investors was about free cash flow and capital return to shareholders. That is great for many types of shareholders, but this is not something a high growth company should be doing. Needless to say now they are swamped by demand and do not have production capacity anywhere near to what demand is.
There is another peculiar quality of semiconductors. They are easy to store and they do not go bad after taking some minimal precautions in storage. This means that any rumor of a shortage and price increases causes everybody that uses semis to go out and buy out everything the suppliers will let them buy.
So there is a very unstable situation. A- there is a shortage and B - even the possibility of a shortage causes hoarding behavior that makes the shortage worse.
That already unstable situation existed before coronavirus. And then coronavirus came. And it triggered a heightened demand for semiconductors. And that triggered a snowball effect of hoarding and higher prices, more hoarding, etc.
The solution is that prices should go up (already happening). Stock prices of semiconductor manufactures should go up. And here I mean actual manufacturers, not fabless chip designers. This should happen especially for analog and power semiconductors where the biggest demand and the biggest underinvestment lies. This has not happened as much as it needs to. Higher stock prices should cause more money to go to the fab business and that should eventually result in the necessary growth in semiconductor manufacturing.
All of this will take time. And meanwhile there will be a lot of hoarding. There may also be occasional panics where the hoarders dump a lot of hoarded inventory thinking prices are going down, cause prices to go down and later discover that prices are shooting back up again after their inventory gets used up.
But in the end hopefully we should end up in a world where (1) there is much more investment in semiconductors and (2) we eventually get all the benefits of the wealthy, high production, environmentally sustainable, all electric future we have been dreaming of.
TSMC all-time high is 632 and is trading now at 620. I don't know about tanked or how you lost money (unless you put a x10 leverage when the stock was at ATH)
So much from before seems old and foreign. Definitely one of those events that marks a definitive before & after... and during, since we still don't quite know how or when things will settles out to a new normal.
> The solution is that prices should go up
Prices at resellers are going up of course, but there's a reason why TI, On Semi, Nexperia, etc might prefer to stock out at a low price than to have product available at a high price: they still want to be selected by cost-sensitive design engineers who are designing products that will be on the market well after the shortage is over.
It's a little dumb that basically all advanced chip manufacturing in the world happens within a thousand mile radius of Beijing and we're over here an ocean over wondering where our supply went. More energy needs to be focused on made-in-USA and less on the thing we have no control over
Are we shipping more chips than even, only demand is allocating unevenly?
Are covid + multiple factory incidents at fault?
Even the cost the stay at the cutting edge facts seem a little suspect for at least some chip markets. e.g. embedded chips such as for automotive use are generally are multiple generations behind fab technology - and really don't need to stay on a cutting edge fab to keep profitably producing for many industrial embedded components.
Like, if a washing machine or a car's fuel injector needs some chips, we should be able to reuse chips from older washing machines and older fuel injectors which we generally throw out because of a mechanical failure with all their electronics intact, but currently it seems that all these chips get just discarded, no one cares about them.
So if I design my washing machine around a particular chip, I could order a million of them for 5 cent each, or I would have to:
* somehow identify other models which have the chip
* find 1 million of them about to be discarded
* rip them open, desolder, collect these parts
* discard the rest of the machine
* do this for about 10 cents per washing machine
combined with the fact that washing machines usually last a decade and will be dying out over a time period spread out over 10-20 years after manufacturing time. How on earth would it be possible to pull off and by that time, these chips will be very obsolete / impossible to order new ones.
A big assumption that I'm making is that you would be extract the exact same components from many different devices, so you can build a washing machine with a component supply that's part new components and part 'reused' from all kinds of older washing machines, fridges or cars or smart toasters or whatever. This IMHO is true for generic electronic components (which are quite cheap but still add up in volume) but might not hold for chips - however, my initial argument probably was that if "chips-for-not-so-powerful-appliances" currently are structured to not be interchangeable, then perhaps they should be.
I would have no issues going out and buying the latest macbook right now but basic other products are missing.
https://www.cnet.com/tech/services-and-software/samsungs-war...
Even if I look at my parents' TVs - all smart but they don't need or use that, they just watch cable.
I'd like to see more power efficient dumb tvs. My favorite TV is a 65" led tv that sips power. At the same time, I've read that manufacturers providing data on their customers helps subsidize lower TV prices, and in that case adding apps likely doesn't incur that much more cost.
"The customers must love this since they're buying 3x as many TVs! Keep em coming!"
Yes, and that's the problem.
This also offers more product flexibility from the same basic chassis. You can offer a "dumb" module for security-centric "cannot have unapproved devices on the network" markets, or a module that has specialized software or connectivity (i. e. a built-in KVM-over-IP client, or something bidirectional for hotels with internal cable and pay-per-view technologies)
Aside from the consumer benefit of "You can just snap in a new $100 module in 3 years and it will be snappy and up to date instead of scrapping a $600 set", this could be a good way to manage the ATSC 3 rollout. You can sell a 4K set today without an ATSC3 tuner, then the customer can add it as an "updated smart module" for $100 later once the format is live in their market.
In my case, I have an HD tv from 2006, it ain't smart, but it does tune to what ever port I want by default. I've got roku on there now, but in the past I've had cable boxes, consoles, even some rabbit ears.
Probably a healthy ability to reuse, repurpose, repair and recycle can be quite effective against the inability to just buy a new thing (regardless of whether the new thing is hi or lo tech)
For a car, my guess is it's impossible to comply with modern emissions and safety regulations without ICs, not to mention the redesigning and retooling that would be required. I would suspect that these regulations have also basically evolved symbiotically with the industry to be a most for them, and they would have no interest in pushing to allow less complex cars.
For TVs and for cars, there are big revenue streams associated with things chips do, like serving ads and tracking what you do with your car, and well as making it impossible to maintain without a dealer. Companies would be very reluctant to shut this down.
None of this helps consumers, but I think that's where we are. Time to look at getting an old motorcycle maybe
A special class of theorem provers could be developed, proving that a program runs below a certain level of spacetime complexity.
This would entail a great increase in energy efficiency.
I also endorse holy information warfare against inefficient proof of work cryptocurrencies and a transition to efficient proof of stake.
That would come at some kind of compile time or efficiency cost since you couldn’t anymore optimise for those but I’m sure that’s something one could opt for.
I don’t think binary size matters though. Storage is very cheap these days.
At least the prices are almost normal again, now that Chia's over.
You could also have self modifying code such that the size of the binary automatically changes as needed.
If you ship a lisp interpreter instead of rust, you can have the interpreter recode itself and any lisp files to a smaller size. You’d just implement a compression algorithm that preserves the functionality of the code compressed.
I think you could do that with rust too with a self compiling binary. It’ll require some real technical skill but if you hire a real hacker you can pull it off.
My cousin his solution for the competitive programming contest had something like that.
I do have a project a bit like that, but I'm not using Rust for it. I was trying to make my own language (like Rust, but more powerful and also smaller), but I'm probably just going to use a modified (safer) C.
If I were to write it in Rust, I'd have to compile it in the first place… and if I could do that easily, I would simply use Rust.
If you're building a docker image for example Rust is going to be smaller.
Sure, I could probably halve that by forking every dependency so they aren't duplicating versions, but that's a lot of work. (It's a shame Rust doesn't let you do conditional compilation based on dependency versions, or this would be a lot easier. As it is, we have to resort to the Semver trick: https://github.com/dtolnay/semver-trick/ — not that many people do that, so it's functionally useless.)
Take GanttProject as an example. It's 20.6MiB of files, plus the JVM. I challenge any of you to make a Rust version (with accessibility support in the GUI) that can open (something resembling) its XML files and draw some (vague graphical Proof of Concept) representation on the screen (with editable text fields), in less than 114+21=135 MiB of binary. And then tell me how, because I've been trying to do that kind of thing for over a year.
¹: I can get it down to around 8MiB with release mode, lto etc., but that significantly increases the build time and only about halves the weight of the intermediate build files.
How dare you claim a larger binary...
Let me guess; Go?
Are you building in "release" mode? The debug builds can be 10x bigger.
cargo build --release
But yes, I have tried my own custom debug profile that turns on the optimisations to try to get the size down. The final binary is smaller, but `cargo build` still regularly leaves me with just kilobytes of space remaining, and then fails outright until I `cargo clean` and try again (which I think is build script related).
I would personally prefer a language model designed to emit x86 assembly, akin to Github copilot, but I understand that to be a minority viewpoint.
Do not rewrite software in Rust, because Rust is not an efficient language. It's a memory, and RAM hog, and it's unstable, with major breakers every release. Switching to Rust is not a thing for a profit seeking enterprise.
In practice, it's an n-fold downgrade from good C on performance.
I know there exists a formally verified secure C compiler. Ideally you would have a similar compiler which guarantees low energy uses.
In my experience Rust already does a great job there but I’m happy to behold whatever evidence exists for C being more energy efficient.
Source? I was under the impression that Rust took an aggressive approach to back compat.
And I'm also not a proponent of rewriting in Rust for it's own sake, but the other commenter was suggesting it in place of Python, which would probably be a huge net gain in efficiency.
Once the OEM sold the board, they made all the money so they don't care about long term support.
I'm not sure when this became a thing, but it's very real to me. It's a snowball effect; these folks get better and better at buying everything up, and then get all the capital to continue and expand the process. I have stock notifications everywhere and everything is gone as soon as it is posted.
Regarding the article - it is definitely not all about fancy fabs that can't be built - these are analog and power parts! And microcontroller fabs are fancier, but usually many many nodes behind state of the art. Maybe it is a capital thing, but every fab I've ever encountered is more than busy. Sure, you lose money trying to do the hardest thing and failing, but the problem is so widespread across so many non-edge products that it must be explained by other factors.
It's consolidation of manufacturing. Many products and product lines have exactly 1 manufacturer today. No competition. No reason not to just run existing fabs, forego investing in more capacity and make bank while prices skyrocket.
How much of this is caused by component fragmentation?
As an example, Digikey lists a staggering 55k separate microcontroller variants (active products):
https://www.digikey.com/en/products/filter/embedded-microcon...
I think we've always had a pretty large variety of microcontrollers, just something like an 8051 was available in hundreds of options from various companies. Microchip has always had lots of different ones with slightly different options. Sometimes (often?) those are the same die in a different package with some different build time options enabled.
Say you use 50 microcontrollers in a car. Probably 15+ separate part numbers? One or two of those can't be acquired, so boards need to be redesigned and the production line is held up.
Now, imagine a different world with wide cross-licensing between manufacturers and instead of doing hundreds of separate variants slighly cost optimized for various purposes for each major microcontroller design, let's just have a small number of kitchen sink variants.
It's primarily a tradeoff between a) increased chip size, b) number of external chip connections, c) cost, d) probability that having a small number of footprints/designs will help.
So, in short: cost vs winnings from having a lean library of component.
The diversity should mean that when Audi picked some specific microcontroller for their ECU there's not too many others that are using the same part at volume. So this should isolate them to some degree from fluctuation in demand. If there's only one vendor/part, like we're seeing with GPUs, then nobody can get any, period. Even if Audi can't get their CPU for some reason then GM who uses a different CPU might be able to get it. Well, unless it's the same factor and the fabrication is the bottleneck...
Here's examples of differences:
- Power consumption (some applications care less, some care more).
- Number of pins. Sometimes you just need one input and one output, shoving a high pin count device in there doesn't make sense.
- Pin functions. Higher current capability or other specialized functions like A/Ds D/As etc.
- On-board peripherals. Some microcontrollers have on-board motor controllers, networking, specialized counter hardware, real time clocks, DSP blocks etc.
- Integrated memory and types of memory.
- Oscillator options/frequencies.
Add to that all the different CPU architectures and vendors.
It's more like the microcontroller replaces a bunch of chips or custom chips in certain applications. It's less like you can build one microcontroller to rule them all. Those are still generally mass produced devices so there's really not a lot of economy of scale, i.e. the one micro to do everything (if that was even possible, which it is not) would just be more expensive and less optimized.
Seems like it would be a great addition to any existing marketplace.
We've all been bitten by someone substituting in a 'equivalent' that 'ought to be the same'.
Mixing in random chips into random boards would be a nightmare to track down when something goes wrong
Hoarding is a component of the shortage, and may be the both trigger and cause for extending the problem from months to years. Semi companies are buying new equipment at record levels, but even the equipment companies are thrashing to find chips to control their equipment.
Analysts may want to keep an eye on balance sheet inventories at electronics manufacturers like Jabil, Flextronix, Fabrinet and their ilk to get an idea of the extent of hoarding in the market.
The part brokers have locked in their quantities months ago, also increasing demand. What are you going to do.
I wonder what would be the long lasting effect for these companies where entire products will get redesigned around what's available - many times these product designs live for 5-10 years. How many companies had to get TI parts out of their products? How's that going to affect TI revenue in few years time?
Many of them have quantity thousands in stock. Especially the MSP430 line.
Granted, maybe the chips you need are not available, but you're exaggerating a bit.
TI temperature sensors, 300 in stock out of 600 normally stocking.
It’s the often used parts that go first.
But let’s put TI aside for a second: search STM32, there are 45 parts out of 1571 normally stocking (!).
Hey, this "supply and demand" curve picture on Wikipedia is tiny, anyone got a freely licensed higher resolution one they can upload?
https://en.wikipedia.org/wiki/File:Fig5_Supply_and_demand_cu...
It’s a wacky world out there right now.
If China goes hard on chip manufacturing, they could economically crush Taiwan.
It's the increased consumption of these goods because of insane lockdowns all around the world.People did not spend money on holidays but decided to upgrade their homes and electronics.
I am honestly tired of this. Yes we had a pandemic but we didn't have a war that killed and destroyed the very thing that produces the things that we are short of right now.
(Might be a good business opportunity if so)
That might be a similar part by another company, or it might be an identical part in a different package (often ICs are available in two or three package shapes, and you might not be able to predict which ones will be in stock).
I sense a potential misunderstanding here: Not every chip runs code. Microcontrollers and processors do, but there are many different kind of chips. Some convert voltages, some massage signals, some drive motors, etc.
Why is this important? Porting code will only help you when you are swapping microcontrollers or processors. Which is just a tinny sliver of all the ICs in use. Porting code won't get you anywhere if what you are missing from your BOM is a boost-buck IC, or an opamp.
I suspect this makes it somewhat hard to restart the economy as there isn't any surplus of a lot of materials. And since downstream consumers are blocked, they fail to generate the demand needed to justify bringing up production.
https://pages.stern.nyu.edu/~adamodar/New_Home_Page/datafile...
How beautiful would it be if chips, and many other things, were built here in the US? In addition to the myriad of obvious benefits, another would be not having to wait over a month or more between a chip “rolling off the line” and finally making the month or more long trip from overseas.
Sure, most news articles regarding this issue will mention offshore production in passing - its impossible not to. But very few if any are actually making that a focal point of the article.
Its called demand > supply.
But if you're GM, you don't need the latest Xeon. You need a 100 MHz single-core microcontroller, that normally sells for $2 to a hobbyist ordering a single chip, but probably something like $0.50 if you're GM and you order a million.
GM would probably happily pay $20 for that microcontroller if it's holding up a $40,000 truck.
If those microcontrollers cost 0.1% of what a Xeon costs, shouldn't that mean, to a first order of magnitude, the factory that produces a 100 MHz microcontroller should cost 0.1% of what the factory costs to produce a Xeon?
How slow / expensive would it be if you wanted to set up a factory to pump out microcontrollers for GM trucks instead of Xeons for Amazon servers?
1. Disruption of supply chains
2. Money printing for Covid stimulus giving additional spending power to consumers increasing demand above the average.
3. Money printing for Covid stimulus reducing demand for low paying jobs.
4. Scalpers.
5. Speculation on the part of manufacturers that prefer to increase prices instead of investing long term to expand production.
6. Other.
Nintendo: Perfect, give me all of them.
There are also problems surrounding qualifications of parts and the supply chain restrictions that suppliers are held to by manufactures. So, trying to replace even just one part on a design that is locked down can be quite painful for a company. We are having many types of parts shortages at this time though, which exacerbates that experience.
Even so, a much “slower part” usually won’t have the same feature set, and may not have the same memory peripherals pinned out, requiring further qualification. Hardware changes of that level are usually done 2-3 years in advance for all of the paperwork and qualification steps. You’d have to build today-cars with standardized qualified parts from years ago. An outrageous example might be trying to stick a cassette-tape head-unit in your latest 2022 car.
(See Newark, for example)
https://www.newark.com/search/prl/results?st=usrp&sort=P_PRI...
https://en.wikipedia.org/wiki/Semiconductor_device_fabricati...
this extends to CMOS VLSI fabrication.
silicon wafers are required as general fodder, as copper clad pc board is to discrete analog production.
https://en.wikipedia.org/wiki/Wafer_(electronics)#Production
the wafers require pulled ingots of silicon, these must be extremly pure and of extremely regular crystal latticing
"factories are more advanced and cost over $20 billion each."
"Once you spend all that money building giant facilities, they become obsolete in five years or less. To avoid losing money, chipmakers must generate $3 billion in profit from each plan"
This is also known as inflation.
1959 Corvettes and Cadillacs wouldn't be bad either.
Even a 1965 Beetle when they still had a flat windshield. They would fly off the shelf.
Heavy overlap in raw materials and supply chain regardless.
Other chips are also made from silicon.
Manufacturers will use said silicon to make whatever chips make them the most money per manufacturing run.
Which, right now, is GPUs.
There are massive shortages at all process node sizes AIUI, not just the cutting edge.
Not very smart to boast about this. So AMD makes more performant chips with less complicated designs?
I implore the national security state and the intelligence community to take action and intensify this trend.
Thus the advance of artificial intelligence will be slowed and we will have more time to practice AI regulation and control.
Only then may strong AI be permitted to eventuate. Otherwise it may destroy us. He who develops AI would be punished severely were we to live in a just world.
We need a butlerian jihad against AI, led by the national security state, equipped by the brave men and women of Anduril incorporated.
The damage to the rest of the economy is unfortunate but a cost I am willing to accept.
Now what do we actually need that hardware for, aside from training language models or mining bitcoin?
Why don’t we also develop more efficient software? Why don’t we pay people to do so? Many of you in this thread could work on this, and be paid handsomely for it.
I expect it to instrumentally secure power to better pursue it’s goals, and in the process thereof establish hegemony over us and other AIs.
From there it could easily try to kill us. I don’t want any of that.
My preferred outcome seems to be what Elon Musk wants, human cyborgs amplified with brain computer interfaces and control over machines; never a machine with control over many humans or posthumans.
Yudkowskian AGI might fail in the long term since it would run into Gödelian problems it would never be able to solve but infinitely loop on.
That humans are immune to those and can behold contradictions testifies to a Penrosian supra-turing-machinic aspect to the human mind.
Man has a transcendent aspect even if we may not be meta-cognitively aware of it at all times.
AGIs could possibly be destroyed through the use of such problems, akin to magical spells wielded by powerful magi.
A secondary threat model of mine is based on a world of non-general, non-agentic strong AI with many discrete “tool AIs” with superhuman ability in one unique area.
The secondary model is mostly harmless except for military AI. The genius of Andúril Incorporated lies in their creation of drones designed to destroy other drones.
This future is still very dangerous, and I hope that the security state will take action to prevent it’s arrival.