The best time to plant a tree is 20 years ago. The best time to build a semiconductor foundry is 5 years ago
- Reference: 1623238091
- News link: https://www.theregister.co.uk/2021/06/09/let_the_chips_fail_where_they_may/
- Source link:
For a resource so fundamental to the connected economy, it seems quite amiss that suddenly semiconductors have fallen into such tight supply. The proximate causes, we're told, include: orders cancelled in the pandemic panic; a drought in Taiwan; a blizzard in Texas; a fire at a fabrication facility; and the [3]unexpected uptick in orders of PCs – again, another outcome of the pandemic.
All of these have some impact on supply chains, to be sure, but none of these – even when combined – add up to more than a bit of temporary tightening around the edges. As this shortage will not be temporary, it's looking increasingly as though all of those "causes" merely hastened an inevitability.
[4]
Someone's forgotten to do their sums.
[5]
[6]
Planning has always troubled the semiconductor industry. From its very earliest days, "wildcat" manufacturers of transistors sucked profits away from the bottom line of the majors with a here-today-gone-tomorrow approach to business. The same thing happened – more than once – with DRAM. The whole history of the industry is littered with firms that either failed to plan – or planned to fail.
[7]Blessed are the cryptographers, labelling them criminal enablers is just foolish
[8]Harassers and bullies succeed in tech because silence is encouraged
[9]A Code War has replaced The Cold War. And right now we're losing it
[10]How do we combat mass global misinformation? How about making the internet a little harder to use
The majors managed their planning well – or well enough to avoid disaster. That's bred them to be somewhat risk averse, conservative in their planning, and parsimonious with their investments. While that approach makes sense in mature markets, one lesson we can glean from the history of computing is that it's all change, all the time. Yesterday's must-have PCs are tomorrow's already-cast-aside smartphones. Forms change, processes change, needs change: if we haven't learned that over the past 50 years, we haven't learned anything at all.
Analysts and shareholders and boards drove a conservative streak deep into businesses that should have spent their time and resources continuously flirting with the edge of disaster. It's that conservatism that's led us to this crisis – well-intentioned, but entirely wrong for the business these companies are actually in: disruptive innovation.
The proofs of this are visible within the remedy: in May we saw well over half a trillion dollars of capital committed to semiconductor fabrication, [11]with $450bn from the South Koreans , around $130bn in [12]President Biden's massive infrastructure spend , and [13]another $100bn from TSMC – the current Master of the Chipmaking Universe. That kind of sudden and overwhelming commitment of capital to anything is completely unprecedented in the history of business. It brings into high relief the central failure here: a long-term aversion to invest in capacity.
Someone's forgotten to do their sums.
For at least a generation, companies have chosen to pass their profits back to their shareholders, rather than investing them productively. Such a strategy, pursued over the long term, inevitably hollows out a business. In the fast-moving semiconductor sector, that vacuum forms at considerable speed and terrifying acceleration. So now, governments in South Korea, the United States, and Taiwan are doing their best to stuff that hole with every dollar they can find.
The best time to plant a tree, it's said, is 20 years ago. The best time to build a next-generation semiconductor foundry is five years ago. Last week, [14]TSMC broke ground on its current-generation fabrication facility in Arizona. It's a start – but won't produce any chips until at least 2024. In the semiconductor industry, even all the money in the world can't instantly transform into productivity. These things take time – and decisions that should have been made years ago.
[15]
We can hope that the semiconductor majors have learned their lesson, and now understand that their futures rely on being willing to take a long walk on the wild side. ®
Get our [16]Tech Resources
[1] https://www.theregister.com/2021/06/01/automakers_pc_chip_shortage_explained/
[2] https://www.theregister.com/2021/04/16/tsmc_chip_forecast/
[3] https://www.theregister.com/2021/05/28/desktop_renaissance_no_its_just/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YMDlni9UI7Rl@LPXbb3tPQAAAA0&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YMDlni9UI7Rl@LPXbb3tPQAAAA0&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YMDlni9UI7Rl@LPXbb3tPQAAAA0&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2021/05/12/blessed_are_the_cryptographers/
[8] https://www.theregister.com/2021/04/20/harassers_and_bullies_succeed_in/
[9] https://www.theregister.com/2021/03/10/code_war/
[10] https://www.theregister.com/2021/02/08/make_the_internet_harder_to_use/
[11] https://www.theregister.com/2021/05/14/south_korea_semiconductor_stimulus/
[12] https://www.theregister.com/2021/03/31/president_biden_broadband_chips_infrastructure/
[13] https://www.theregister.com/2021/04/02/tsmc_fab_investment/
[14] https://www.msn.com/en-gb/money/technology/tsmc-says-has-begun-construction-at-its-arizona-chip-factory-site/ar-AAKC2L7
[15] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YMDlni9UI7Rl@LPXbb3tPQAAAA0&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[16] https://whitepapers.theregister.com/
Is this fair?
The article mentions the billions that will be spent by South Korean and Taiwainese companies and yet they are the only companies that have already been spending on both process and fabs.
I think todays problems are from two sources. One is Intel and the missteps over process. It is behind in manufacturing because it went down the wrong path. And it would be strange to spend billions building fabs based on a process that does not work, so they haven't. They just spent billions on a process that did not work.
Secondly though TSMC and Samsung (and Intel) have spent to be able to build the latest and greatest, most companies have not. And the latest CPU/GPUs require the latest and greatest and those CPU/GPU are in more demand than ever. So blame all of the other companies that dumped their fabs to avoid the capital costs, leaving fewer and fewer manufacturing companies.
True for some parts
The high performance processors (CPU / GPU) are definitely affected due to a lack of investment, but there are far more devices made and sold than those.
Those parts aren't sexy but a processor, memory and a GPU alone do not a computer make (well, not a very useful one anyway).
There are the parts in the power supply (and with modern parts we have gone to local point of load converters due to the very high core currents required; as core voltages go down, the current goes up for the same power) which is difficult to distribute, to say the least.
I remember designing a 1.8V 30A point of load converter almost 20 years ago and I had to use what is known as [1]remote sense (you sense the voltage at the device you are powering) because even using very heavy copper and multiple layers, I still had voltage drops over a few inches of about a quarter of a volt.
That was not the highest current regulator I designed; 100A at 1.2V still haunts my dreams.
Then there are workhorse parts such as configuration memory, peripheral controllers and the like.
In the wider market there are parts that are simply not suitable for a 7nm process (it is believed that the smallest feature size for NOR flash is 65nm to get decent yields for example although there is nothing stopping a 7nm line from producing them but it seems like a waste of a top end fab line). Parts such as ADCs, DACs, ethernet PHYs and the like sell many more parts than processors and GPUs combined and the fabs that make them are churning this stuff out day in and day out.
So the shortage is for the latest and greatest parts on very tiny nodes (FPGAs included although I am not sure I want to pay upwards of $2K for a single part).
There doesn't currently seem to be a shortage of all those essential but often forgotten parts.
[1] https://www.ti.com/lit/an/slyt467/slyt467.pdf?ts=1623171819911&ref_url=https%253A%252F%252Fwww.google.com%252F
Re: True for some parts
Aren't most of the fabs making 65-130nm parts redundant CPU fabs from a couple of generations ago?
Are these going to keep running indefinitely because you aren't going to be able to use redundant 5nm fabs in 5 years to make power electronics? Or is there somebody making brand new 65nm lines today ?
Re: True for some parts
These fabs are smaller and have no need to go to a new process node until they update the equipment.
Going to a smaller process node does not preclude making 65nm (or larger) parts; that is simply the smallest feature size supported.
They are mostly owned by the vendors (TI, Analog devices, Spansion, Hitachi etc) who are still pumping out 68HC devices in some cases.
TI is still making [1]741 opamps (originally released by Fairchild in 1963!).
No doubt it is on a newer process node, but the specifications compared to a modern device are pretty rough but if you have a circuit that was proven with them (there are lots of those still around) then that is what you will need to buy.
[1] https://www.ti.com/lit/ds/symlink/ua741.pdf
Shortages enable to make prices higher. They aren't always unintended.
So
Planning sucks and the only viable way forward is market freedom. I am sure someone will be shocked but I dont know who.
The central failure
We're going to see another central failure in a short time : building all of your new fabs in the driest State of the Union.
Arizona. Who the blazes thought it would be a good idea to build there ? Well, apart from the high-level suits who undoubtedly extracted a monstrously insignificant tax agreement and didn't give a flying one about the additional drain this is going to put on what's left of the Colorado river.
Additionally, cooling costs are going to be through the roof.
Way to go for ecology all around.
Big customer said "we don't need any chips"
The auto industry figured they'd not sell any cars during Covid, so they said "we don't need any chips"
So Intel/AMD/etc said "awright mate, we'll make other stuff"
Now the auto industry complains of shortages.
Hmmm... Sounds like an own goal to me.
...and the best people to pay for it
are the taxpayers. $450Bn from South Korea's government, $130Bn from the US government. Of the three listed in the article, only the TSMC investment is from a company.
Optimistic?
Not me.
Everyone in the industry with more than two communicating neurons could see this coming, but the people determining the finance are most certainly not in the industry.