News: 1643980751

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Dear chip designers, if you're struggling to get components made, try 28nm. Supply is set to overtake demand

(2022/02/04)


Analysis Amid the semiconductor crunch, there's an interesting cliff forming at 28nm.

While demand for other process nodes exceeds supply, the tech world's need for that mid-level node may drop below available manufacturing capacity, if not already.

Early indications of a potential oversupply at 28nm emerged during an earnings call with UMC, a top contract microchip manufacturer headquartered in Taiwan.

[1]

"On the supply side, based on the announced capacity expansion plan, we do see the oversupply situation at 28nm to happen beyond 2023, not before 2023," Jason Wong, president of UMC, [2]told analysts on the call.

[3]

[4]

Fabs, such as those owned by UMC, TSMC, and Intel, generally can't churn out chips fast enough to satisfy demand. Microchip makers are adjusting manufacturing capacity accordingly and bringing fresh factories online, raising the possibility of chips flooding the market, with supply eventually matching or outstripping demand.

Analyst firms believe chip shortages will end sometime in 2023, and IDC has [5]predicted an oversupply scenario emerging.

[6]

This oversupply issue may in particular hit the 28nm node, which is used by UMC to make microcontrollers as well as integrated circuits for wireless communications, sensors, and mixed-signal applications. UMC also makes chips on a 22nm node.

Financial analysts on a TSMC earnings call last month probed the biz about 28nm oversupply, which plagued the Taiwanese giant in previous years. There is a concern that TSMC and its peers could end up with a glut in capacity after competing to establish multiple rival plants.

TSMC built the world's first 28nm fab in 2011, and last year invested in a new 22/28nm factory in Japan, and is expanding capacity in Nanjing, China. Rival chip manufacturers are also increasing their 28nm capacity; SMIC last year started building a 28nm fab that will begin operation in 2022.

[7]

On his call with analysts, TSMC CEO C. C. Wei admitted that 28nm was problematic before the pandemic: his fab lines for that node had a utilization rate just a little bit above 80 per cent in 2018 and 2019. He said he hopes a repeat of this situation can be avoided as chip supplies normalize.

NAND flash controller house Silicon Motion Technology last week said its supplier TSMC was short on capacity at many nodes, though not for 28nm. And if Silicon Motion can't get enough dies on non-28nm, it'll rethink how it can use 28nm to get product out the door, it said.

"We are in severe shortage in certain technology nodes, including 55nm and all advanced technology nodes, 16nm, and 12nm. However, I think we are comfortable in 28nm wafer supply," said Wallace Kou, CEO of Silicon Motion Technology, on an earnings call.

"If we cannot get any additional wafer supply from TSMC, it all depends on how we can fully utilize 28nm with a better product mix."

[8]There's a new Chipzilla in town: Samsung topples Intel to become largest chip maker

[9]You're fabbing it wrong: Chip shortages due to lack of investment in the right factories, says IDC

[10]Now that's wafer thin: Some manufacturers had less than five days of chip supplies, says Uncle Sam

[11]Intel plunking down $20bn for at least two chip factories in Ohio

The demand for microchips is expected to [12]generate $680.6bn in revenue for semiconductor companies in 2022, trending to a trillion dollars in the future with the electrification of cars and companies stuffing more electronics into products.

The age-old pattern

Chip companies, particularly memory makers, have dealt with volatility for decades, and it is considered part of the semiconductor demand and supply dynamics.

The boom and bust cycle for chip makers are exacerbated by macro events, such as economic crises, natural calamities, and over-or-under investments. Historically, the semiconductor market has crashed after events such as the 2007-2008 financial crisis, which halted IT purchasing, drying up demand for chips.

But the COVID-19 pandemic worked in the favor of chip makers. The demand for chips skyrocketed as people worked and learned at home, leading to PC upgrades and cloud players buying in equipment to power remote collaboration and communication. This collided with semiconductor content increasing in electric vehicles, and already high demand for smartphones, laptops, custom silicon, and servers.

Chip companies have cashed in with record revenues, and committed billions to expand manufacturing via facility upgrades or new factories. Most investments are in state-of-the-art nodes for compute-intensive parts such as graphics accelerators and smartphone system-on-chips, and older nodes for cost-optimized chips, from microcontrollers to integrated analog circuits.

'Precarious'

The 28nm node may be snubbed or skipped by fab customers in favor of other nodes that offer more performance or are more cost-effective, which could create an oversupply scenario, said Akshara Bassi, an analyst at Counterpoint Research. "So 28nm could be stuck in that precarious transition cycle, where some products are moving to more-advanced-than-28nm faster than expected," Bassi said.

Certain products, such as memory controllers and I/O dies, could also be in a transition phase from lagging nodes to a "little more advanced matured nodes for better cost savings and to meet environmental, social and governance commitments from OEMs," Bassi added.

Microcontroller chips and other embedded parts are advancing to the 14nm node, STMicroelectronics CEO Jean-Marc Chery said in an [13]earnings call last week. He signaled that 28nm is standing at an uncomfortable cut-off point for foundries investing billions in factories.

For chip designers, advancing from 28nm to a node like 22nm is a reasonable jump that, for one thing, opens up the possibility of obtaining more dies from fabs and thus increasing profitability, said Brian Matas, vice president of market research at IC Insights, a semiconductor number-crunching firm.

By shrinking a design to a 22nm node, one ought to get more dies on a wafers and ship more finished components, for example.

"You're able to reduce the prices and pass that on to some of your customers, and at the same time, you're still making more money off that smaller process node," Matas said.

UMC and TSMC acknowledged that a correction will be coming as chip momentum slows down, and that the negative impact on 28nm will be a temporary minor blip, while applications emerge to make the process relevant again.

"We do observe that our long-term structural demand at 28nm was to be well supported by multiple specialty technologies such as CMOS image sensor for multi-camera trend and better nonvolatile memory application and other specialty technologies," TSMC's Wei said. "The enrichment in the silicon content in many end devices that develop in recent years helped to support this demand."

UMC's Wong said oversupply on 28nm in 2023 will be mild for his firm, and that ultimately the node will emerge in the future as a sweet spot for many applications. "Then expect the demand will continue to migrate to 28nm and that 28nm demand will continue to grow," he argued. ®

Get our [14]Tech Resources



[1] 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=2Yf1bt8fN7K@rZ@zTgDMvNQAAAIU&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[2] https://seekingalpha.com/article/4481465-united-microelectronics-corporation-umc-management-on-q4-2021-results-earnings-call

[3] 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=44Yf1bt8fN7K@rZ@zTgDMvNQAAAIU&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[4] 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=33Yf1bt8fN7K@rZ@zTgDMvNQAAAIU&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[5] https://www.theregister.com/2021/09/20/chip_shortage_to_end_2023_idc

[6] 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=44Yf1bt8fN7K@rZ@zTgDMvNQAAAIU&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[7] 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=33Yf1bt8fN7K@rZ@zTgDMvNQAAAIU&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[8] https://www.theregister.com/2022/01/31/samsung_intel_swap/

[9] https://www.theregister.com/2022/01/28/idc_chip_shortages/

[10] https://www.theregister.com/2022/01/25/doc_chip_shortages/

[11] https://www.theregister.com/2022/01/21/intel_us20bn_ohio/

[12] https://www.theregister.com/2022/01/10/chip_growth_this_year/

[13] https://seekingalpha.com/article/4482022-stmicroelectronics-nv-stm-ceo-jean-marc-chery-on-q4-2021-results-earnings-call-transcript

[14] https://whitepapers.theregister.com/



Sort of inbetween

imanidiot

I think the problem is that 28nm right now is sort of nowheresville when it comes to fab nodes. A lot of the currently in use designs for microcontrollers and the like use older nodes like 30-ish and up from what I can see as that is what previously was available cheaply, while the current state of the art for (phone) CPUs has moved on to lower nodes (probably 14nm and down). As fab capacity in the 28nm node becomes available (and cheap) we'll probably see a lot more chips designed on that node to fill up the available capacity.

Re: Sort of inbetween

Electronics'R'Us

The 28nm node is perfectly fine for many quite advanced devices (the first series 7 devices from Xilinx are on that process and they are very capable with multi gigabit ports).

If the process is underutilised but could produce what the manufacturers / market needs, then it would make sense to move over to it.

Re: Sort of inbetween

imanidiot

Thing is that it's not very easy to move a part taped out for a certain node over to a different node size. That would require a near complete redesign. So manufacturers tend to stick to the node size part was designed for once it's been produced at that node, they tend to only move to smaller nodes for new products during the initial design

Good news for Raspberry Pi

Old Used Programmer

The Pi4 SoC (BCM2711) is a 28nm part.

"Seven years and six months!" Humpty Dumpty repeated thoughtfully.
"An uncomfortable sort of age. Now if you'd asked MY advice, I'd have
said 'Leave off at seven' -- but it's too late now."
"I never ask advice about growing," Alice said indignantly.
"Too proud?" the other enquired.
Alice felt even more indignant at this suggestion. "I mean,"
she said, "that one can't help growing older."
"ONE can't, perhaps," said Humpty Dumpty; "but TWO can. With
proper assistance, you might have left off at seven."
-- Lewis Carroll, "Through the Looking-Glass"