I think there are parallels between a lot of the semiconductor supply chain shocks and the toilet paper shortages at retail.
TP and ICs both are relatively inexpensive and have the ability to be stored for a long time without functional degradation. If you think either will be in short supply, it’s not that costly nor inconvenient to buy 5-10x your normal supply “just in case”, especially if you can see that others are likely to do the same.
To the extent that’s been a driver, I can’t see this lasting longer than 5 years at the very outside. Once there’s confidence that parts will be available in distribution in 2028, people will stop over-buying and chew through their increased safety stock to return to some normal, which might mean more local stock held, but distribution flow will become closer to finished product flow again.
I’m just a small-time hobbyist and I bought several extra reels of passives (mostly MLCCs) and partial reels of a couple microcontrollers that I use in early 2020. There was no reason for me to do that before 2019.
What happen when the big business industry of stearic candles fall against electrical/gas light bulbs? One was hyper developed and fall badly, the other was not ready to grow at the same fast peace of demand. That's just as a single example.
The main point is: civilization changes, normally not so fast, but they do, and normally most are not ready and adverse any change, no matter if good or bad as a result we end up always in rushes. As a result scarcity, precarity, FUD happen. The key is an ancient Latin phrase "in medio stat virtus", or for the present topic being too efficient is not efficient because generally means not being flexible nor capable to evolve. On the other side of the spectrum we have seen in IT how hard are distributed systems respect of few centralized ones. In medio stat virtus again, we need a bit of efficiency and a bit of flexibility, and yes, it's not that easy, and yes it cost money.
You can click on the domain name in the post to see the few other submissions there have been; the most successful one was the Experimental Music Notation thread (https://news.ycombinator.com/item?id=17166055).
What’s stopping a bunch of small manufacturers from pooling together their capital to buy an upstream supplier?
After all if the price disparity is indeed so large and will be sustained then you should have wall street guys eager to lend to get in on the action and make easy returns?
If the returns are not there and the price of resources, labor, etc., have increased simply due to inflation and other factors then this is like saying Ferrari’s are scarce. Or $15000 new cars from Toyota are scarce. 50 years ago that would have bought a new Ferrari from the dealer, and now it can’t even buy a Corolla.
If so, I don’t see anything other than a trivial point.
> What’s stopping a bunch of small manufacturers from pooling together their capital to buy an upstream supplier?
Texas Instruments is worth ~160 billion dollars, ST Micro ~38 billion dollars. Even at Elon Musk level of wealth, that is unobtainable.
To make stuff worse, all the stuff in short supply (anything from power regulators to not-top-of-the-line SoCs) is made on "outdated" nodes. No one will go and set up a brand new 90nm fab (which is the node size of most current STM32 models), it's just not worth the expense without massive government subsidies.
Why jump all the way to the big fish? It’s more sensible to start with a distributor like Digikey since presumably they’re not buying directly by the pallet from TI or ST Micro if they are ‘small manufactuers’.
If they don’t think they can do Digikey’s job better than what’s already being done then they have their answer.
The folks upstream and at Wall st can presumably do math and think a few steps ahead, so if they saw an opportunity for outsized returns they would likely already have plans for it.
Just like why nobody is making a $15000 Corolla equivalent in 2022, because it’s not profitable enough.
Unfortunately there's another angle here which is that the little fish are competing for supply with higher priced, higher volume products like cars, where a 10x increase in one component can be amortized over a much higher total cost and passed on to the customer. Nobody is going to notice an extra couple hundred bucks on the price of a car, but they'll definitely notice when it's on something that itself only costs a couple hundred bucks.
Big consumers are able to outbid even companies like DigiKey because they can afford to spread the cost around. TI had one of its best quarters ever over the last 12 months, but just try to buy TI parts anywhere - they're being priced through the roof. A $0.30 TI regulator was going for $30.00 on the secondary markets. I feel for any small company trying to build hardware right now.
Indeed that seems like the market telling small manufacturers of lower priced goods, who are unable to increase the price of their product, that other manufacturer’s goods are providing more value.
For those who were expecting cheap electronics to be preordained forever this serves as an effective wake up call, and that nothing on this Earth is truly guaranteed even one second into the future.
Not 90nm, but a 28nm fab still looks quite viable, as AFAIK, its cost per transistor is still lower than the top of the line ones.
Anyway, the reason nobody is investing is because everybody expects the problem to go away by itself in a year or two, and if they invest now, they will start production just in time to get destroyed by the opposite phase of the whiplash effect.
> Anyway, the reason nobody is investing is because everybody expects the problem to go away by itself in a year or two, and if they invest now, they will start production just in time to get destroyed by the opposite phase of the whiplash effect.
Or rather because in two or three years the top CPU/GPU manufacturers will have moved to TSMC's "2nm" node size, which frees up capacity at larger nodes and that will propagate across the chain.
With ST, specifically, I'm not seeing a lack of Silicon but a lack of lead frames. The recent shutdowns in China caused several batches of lead frames to expire unused, creating a big gap moving through the supply chain. Chip-scale packages have better availability.
Seems like anyone relying on a few critical special parts should just leave some space on the PCB around the part so that you can just replace the pads for one part with those for another and then route and jumper the pins to match the functionality of the old layout. It would make the boards a little bigger, but it would make it much easier to replace an MCU with another part. Zero ohm jumper resistors make adding jumpers for routing easy for pick-and-place machines. For small scape manufacturing the added costs would probably be minimal. If you're going to have to redesign the product anyway, might as well make it more survivable while you're at it.
Computing is baked into the design of almost everything because it was so good at driving down component counts and increasing reliability. It's ironic that the very devices that drove this change are now themselves the failure point.
I worry that this collapse will find its way into machine tools, the basis of all industry. If inserts (cutting tools used for almost everything) were to suffer a similar shortage, a lot of well tuned industrial processes would have to go through all the pain of startup again, instead of just being able to crank out more of the same parts every 6 months when a few more were needed.
This is a very analogous situation we now find ourselves in, in the electronics world. You can't just use a new chip without going through all the teething pains, the months of frustration as you keep finding new corner cases. There are very good reasons for conservative views in the machining and electronics worlds. The accountants who outsourced everything tore down Chesterton's fence... and it turned out to be holding up the world.
16 comments
[ 2.2 ms ] story [ 40.4 ms ] threadTP and ICs both are relatively inexpensive and have the ability to be stored for a long time without functional degradation. If you think either will be in short supply, it’s not that costly nor inconvenient to buy 5-10x your normal supply “just in case”, especially if you can see that others are likely to do the same.
To the extent that’s been a driver, I can’t see this lasting longer than 5 years at the very outside. Once there’s confidence that parts will be available in distribution in 2028, people will stop over-buying and chew through their increased safety stock to return to some normal, which might mean more local stock held, but distribution flow will become closer to finished product flow again.
I’m just a small-time hobbyist and I bought several extra reels of passives (mostly MLCCs) and partial reels of a couple microcontrollers that I use in early 2020. There was no reason for me to do that before 2019.
The main point is: civilization changes, normally not so fast, but they do, and normally most are not ready and adverse any change, no matter if good or bad as a result we end up always in rushes. As a result scarcity, precarity, FUD happen. The key is an ancient Latin phrase "in medio stat virtus", or for the present topic being too efficient is not efficient because generally means not being flexible nor capable to evolve. On the other side of the spectrum we have seen in IT how hard are distributed systems respect of few centralized ones. In medio stat virtus again, we need a bit of efficiency and a bit of flexibility, and yes, it's not that easy, and yes it cost money.
a great place if you're interested in tech and electronic music
After all if the price disparity is indeed so large and will be sustained then you should have wall street guys eager to lend to get in on the action and make easy returns?
If the returns are not there and the price of resources, labor, etc., have increased simply due to inflation and other factors then this is like saying Ferrari’s are scarce. Or $15000 new cars from Toyota are scarce. 50 years ago that would have bought a new Ferrari from the dealer, and now it can’t even buy a Corolla.
If so, I don’t see anything other than a trivial point.
Texas Instruments is worth ~160 billion dollars, ST Micro ~38 billion dollars. Even at Elon Musk level of wealth, that is unobtainable.
To make stuff worse, all the stuff in short supply (anything from power regulators to not-top-of-the-line SoCs) is made on "outdated" nodes. No one will go and set up a brand new 90nm fab (which is the node size of most current STM32 models), it's just not worth the expense without massive government subsidies.
If they don’t think they can do Digikey’s job better than what’s already being done then they have their answer.
The folks upstream and at Wall st can presumably do math and think a few steps ahead, so if they saw an opportunity for outsized returns they would likely already have plans for it.
Just like why nobody is making a $15000 Corolla equivalent in 2022, because it’s not profitable enough.
Big consumers are able to outbid even companies like DigiKey because they can afford to spread the cost around. TI had one of its best quarters ever over the last 12 months, but just try to buy TI parts anywhere - they're being priced through the roof. A $0.30 TI regulator was going for $30.00 on the secondary markets. I feel for any small company trying to build hardware right now.
For those who were expecting cheap electronics to be preordained forever this serves as an effective wake up call, and that nothing on this Earth is truly guaranteed even one second into the future.
Besides:
> No one will go and set up a brand new 90nm fab
Not 90nm, but a 28nm fab still looks quite viable, as AFAIK, its cost per transistor is still lower than the top of the line ones.
Anyway, the reason nobody is investing is because everybody expects the problem to go away by itself in a year or two, and if they invest now, they will start production just in time to get destroyed by the opposite phase of the whiplash effect.
Or rather because in two or three years the top CPU/GPU manufacturers will have moved to TSMC's "2nm" node size, which frees up capacity at larger nodes and that will propagate across the chain.
I worry that this collapse will find its way into machine tools, the basis of all industry. If inserts (cutting tools used for almost everything) were to suffer a similar shortage, a lot of well tuned industrial processes would have to go through all the pain of startup again, instead of just being able to crank out more of the same parts every 6 months when a few more were needed.
This is a very analogous situation we now find ourselves in, in the electronics world. You can't just use a new chip without going through all the teething pains, the months of frustration as you keep finding new corner cases. There are very good reasons for conservative views in the machining and electronics worlds. The accountants who outsourced everything tore down Chesterton's fence... and it turned out to be holding up the world.