Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/the-five-biggest-mcu-suppliers-accounted-for-82-of-2021-sales.16228/page-2
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021171
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

The Five Biggest MCU Suppliers Accounted for 82% of 2021 Sales

jms_embedded

Active member
Cool, thanks for the reference; I'll check it out. I've been keeping my eye out for any public statements I can find

Serious question on what is driving the uptick in demand though on these specific nodes

I have some clues here, and a possible answer but not a definite answer.

The MCU market has constraints and optimizations for process size.

The constraint is the availability of embedded flash memory -- my understanding is that it goes down to 28nm (see linked article) but even that is relatively new, and 40nm or larger applies to the majority of MCUs. NXP has a 16nm automotive MCU but it relies on external NOR flash.

Optimizations are for cost-effectiveness. I don't have access to the kind of data that would point to which technology nodes are most cost-effective for various mixes of MCU complexity (if we say "cost modeling" three times, perhaps Scotten Jones will appear and shed some light on the situation) but I suspect that on the low end, if you have a relatively low-end electronic feature X in a car that requires an MCU which needs to cost, say, Y cents or less in 10M quantity, then depending on what Y is, you'll get different optimal technology nodes that are larger than 40nm. I'm just going to throw out some made up numbers: maybe at Y = 50 cents, the optimal node size in 2022 is 110nm, and maybe at Y = $1.00 the optimal node size is 55nm. The 16nm NXP parts look like they're in the $40 range in low quantities so maybe $15-20 at very large quantities.

One problem with cars (unlike phones) is that the electronics needs to be distributed; you can't have one big application processor to do everything, so the level of integration has its limits, and with these little cheap MCUs you're stuck on whatever process is lowest-cost / lowest-risk at the time of IC development.

As for the "why didn't they see this coming" question -- that is MUCH harder to answer, and my only guess is that it is a game of chicken, kind of an anti-Moore's Law where the auto manufacturers and their Tier 1-2 suppliers said "well we can just rely on the MCU companies for JIT, we don't have to pre-order" and the MCU companies said "well, we're just going to handle it quarter by quarter, and the foundries are going to have some more capacity by the time we need it" and the foundries said "well, we're just going to focus our capex on the more advanced nodes and the MCU companies will shift away from mature nodes when they need to" and now we have our train wreck. It would have happened without COVID but with the supply/demand shocks of COVID it just happened faster.

That last paragraph is a supposition on my part; I can't back it up.
 
Last edited:

jms_embedded

Active member
Check here for complete coverage of the TSMC Technical Symposium:
I saw that, and it's interesting, but not sure how it's relevant to MCUs, except for this one sentence:

TSMC will continue to invest in mature node and specialty technologies with a 1.5x capacity expansion from 2021 to 2025 which includes Fabs F14P8 (Tianan), F16 P1B (Nanjing), F22 P2 (Kaohsiung) and fab F23 P1 (Kumamoto Japan).

Did they give any breakdown on what this "1.5x capacity expansion" means? Kumamoto is new construction, but it's expected to be 12nm - 28nm, which doesn't really qualify as "mature nodes" in anyone's mind except maybe TSMC's. This is a joint venture with Sony and Denso, and presumably Sony's interest allocates a good chunk of the 22-28nm capacity for image sensors, with Denso's interest for ADAS. (I wonder if you look at 40nm and above, how much capacity expansion is TSMC investing in?)

TSMC lifts Japan chip plant investment, with Denso following Sony​

Additional funding will help pay for 12-to-16-nm semiconductor line

Nikkei staff writers
February 15, 2022 18:20 JST Updated on February 15, 2022 21:48 JST

TOKYO -- Taiwan Semiconductor Manufacturing Co. will increase its investment in a project that will see the construction of the company's first plant in Japan, as auto parts maker Denso joins in the project along with Sony Group, the three companies announced Tuesday. The announcement confirms Nikkei's earlier report.

The chip plant will be built in Kumamoto, in southern Japan. It will be undertaken by a joint venture called Japan Advanced Semiconductor Manufacturing or JASM, which will be majority-owned by TSMC.

The three companies said Denso will provide $350 million in equity for JASM, taking a stake of over 10%.

The expected increase in funding will help TSMC to build a production line for 12-to-16-nm chips in Kumamoto, in addition to the 22-to-28-nm chip production line envisaged in the original plan.

"JASM is not only an opportunity for TSMC to support growing market demand for specialty technologies, it enables us to leverage Japan's top-notch semiconductor talent and contribute to the growth of the global semiconductor ecosystem," C.C. Wei, TSMC's chief executive, said in a joint statement.
The project was originally estimated to cost 800 billion yen ($7 billion), but the three companies say the project will be now worth $8.6 billion. Construction is scheduled to begin this year, with production to start by the end of 2024.

The project was unveiled in November and is expected to produce semiconductors, including chips used in image sensors, a key product for Sony. Sony intends to provide 57 billion yen, or about $500 million, in capital for the project over the next two years, acquiring a stake in JASM of less than 20%.

The Kumamoto plant will turn out logic chips, which serve as main processors in computers. Chips in the 10-nm class are necessary to handle the complex data generated by autonomous vehicles and advanced driver-assistance systems designed to prevent collisions. With its contribution, Denso aims to secure a stable source of the advanced chips needed for autonomous vehicles.

"Semiconductors are becoming increasingly important in the automotive industry as mobility technologies evolve, including automated driving and electrification," Koji Arima, president and CEO of Denso, said in the statement. "Through this partnership, we contribute to the stable supply of semiconductors over the medium to long term, and thus to the automotive industry."
 

Paul2

Active member
Twitter press coverage of the European session of TSMC Tech Symposium quotes as 65/55 is the most over-subscribed node currently:

However, their capacity build-out only seems to be for 28/22nm and newer (based on their comment of 50% increase for 28nm capacity, at least). So customers will have to consider migrating at some point. Serious question on what is driving the uptick in demand though on these specific nodes, because one could assume that many of these "10yr+" auto grade supply parts have been in production already for some time, and demand forecast shouldn't have gone 2-3x just for this year. Was there a shift in ECUs or similar controller specific to EV management that would correlate (maybe a specific customer's battery charging monitor IC)? I would have though those were on older nodes still. It can't all be Apple's VR demand, or CoWoS.

No wonder at all. 65nm is a wonderful node:

1. Design portability
2. The last dry single expo node
3. Productivity, and yield maxed out long ago
4. Low cost design — South Asian outsourcers can still work with it, no gazillion dollars EDA packages needed
5. Mixed signal, flash, high voltage, RF - all available

An average microcontroller die is already so tiny at that node that you will need extra silicon for pads anyways.
 
Top