Semiconductor average lead time breaks half-year barrier We all know the global chip shortage has been bad, though here's a new data point: semiconductor lead times grew to an average of 26.6 weeks in March. For those who have, in this era of perpetual pandemic, understandably forgotten how calendars work, this means it now takes chipmakers more than half a year, on average, to …
What I am finding is that relative old tech, that has almost commodity status, is not too difficult to source, so this could be rated short lead time. However, the bigger chips, that do specialist jobs, and are only produced by one manufacturer, tend to be the ones with the really long lead times. There may be some well-established and popular single source products that are available to a fairly long (months) lead time, but that can be worked around. Then you have the stuff that probably won't be available for production until after I have retired, if current lead times are to be believed.
The point is, you can construct a statistical distribution with the bulk of components being readily available but rather old tech, a central family of medium tech on longish but manageable lead times, and a tail of unicorn kit. The mean of this distribution might be six months, but that conceals the fact that in practice, you can't build a product unless you can purchase the entire kit of parts, and if the design uses a few parts on two year lead time, that means your whole production is on two year lead time, even if you can buy the jelly bean components ex-stock.
"Susquehanna said there were a variety of events that impacted the supply chain in the first quarter, including Russia's invasion of Ukraine, an earthquake in Japan, and two pandemic-driven lockdowns in China. The firm suspects the effects of these may linger for the rest of the year."
I always get an itch when I read things like these.
If the shortages are seen as caused by lockdowns, wars and what not, there will linger a belief that if the lockdown ends, the shortages will end too. But fab expansion is going to take 2-3 years, and will not be homogeneous across all nodes, afaik most of the investments are below 20nm. And demand will continue to increase as people want cars to drive themselves, products to build themlselves etc.
Computerization of cars, everyone and their grandmother wanting a supercomputer, edge computing with its sister cloudification (i.e. datacenters), private 5G, IoT, the electrification of everything under the sun have all been aggressively pushed in discourse for years, and it looks like with the lock downs people saw an opportunity to implement it and took it, causing an unprecedented surge in demand, with all market segments mentioned above seeing double-digits growth until 2025 and/or 2030. And let's not forget that, as some have pointed out, there is a strong component of inventory hoarding as well.
https://corporate.vanguard.com/content/corporatesite/us/en/corp/articles/the-chip-shortage.html
It might be just me looking for confirmation bias, but there s some support for this reading elsewhere too:
https://corporate.vanguard.com/content/corporatesite/us/en/corp/articles/the-chip-shortage.html
“All of the connected everythings--smart cities, smart roadway, smart campuses, smart airports, smart, autonomous everything--I think this [shortage] was going to happen anyway, it just happened faster,” said Fenn.
Sure, then there s remote work and laptops, China's draconian (weaponized? or not... who knows, might be the perfect crime) lock downs... It's all a big soup. When talking about these things presenting a plurality of opionions is even more needed than usual. One can argue about relative weights, but using lockdowns and now war as a focal point to explain everything away is just smoke in the eyes.
I wondered about that. Last time I was involved in semiconductors we were getting custom stuff made for cheap on obsolete 3" wafer lines.
My understanding is that simple micro-controllers made on 60-100nm designs use what were cutting edge fabs a decade ago. Nobody is going to build a new 100nm fab because the margins on these parts is 1% of bugger-all so it would never pay.
Now in another 10 years there will be lots of 'obsolete' 5-7nm fabs around, but you aren't going to be making power switches on 5nm gates. And even if you just scaled up the design, the costs of a 5nm EUV process, even with the plant paid off, is unsustainable.
So are we going to need to start building lower end fabs? Or are we going to rely on a new round of 3rd world countries, after India and China catch up to TSMC? In 10 years are we going to be moaning about a car shortage because of a fab shutdown in Somalia?
I suppose I should set up a lucrative sideline in punditry because all I'd have to do to explain the current lead times is list off a few headlines that sound tolerably relevant.
I'd actually like much more relevant information. Semiconductors aren't made in one place, for example, so where is the bottleneck? Is it a shortage of raw material? Is it backlogs in process steps? Is it the product of embargoes or tariffs? Is it speculative cornering in the market for a handful of key semiconductor types? Understanding this might give us a clue as to when things might ease; at the moment parts seem to be either 'in stock' or 'not in stock and we have no idea when they'll be back in stock'. Not a lot of help to us.
Nowadays, a lot of my work as an electronic designer is helping out the purchasing department with sourcing alternative components, where the part I originally specified is on a stupid lead time. When designing new products under these circumstances, the whole process is bent of of shape, because there is no point designing a product if you can't build a prototype for a year or two. This gums up the development and production engineering process. As a result of this, I am more or less forced into an extremely conservative design strategy: no new chip types, because you might not be able to buy them. It would be an understatement to say that I find all of this very frustrating, because in the past, I have designed in new chip types, found they work nicely in testing, and then they become standard parts in our inventory, that I can rely on for future designs. This keeps our products fairly up-to-date with developments in semiconductors, resulting in improvements in performance and profitability.
I think what the article is missing is how many small and medium scale electronic manufacturers purchase their semiconductors, which is primarily through distribution, not directly from manufacturers. This is where I get my two year lead time from. I think that what has happened is that larger buyers have bought the stock in advance, so most of the output from manufacturers is spoken for, and the distributors are left with what is left, which is not enough. One of our suppliers, who is both a manufacture and a distributor, encouraged this advance scheduling of orders. I dare say it suits them to be able to plan their business for years ahead. I have seen it before. The supplier offers a discount, and some guarantee of availability, in return for a customer committing to advance orders. But two years lead time? Come on, someone is taking the piss.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
