Chapter 451: Where Is The Limit
This is the last day of 2025, Pengcheng’s evening rush hour has passed, and the workers waiting in droves for the company shuttle, dragging their feet until eleven o’clock like worker bees, have all returned to their residences.
At this time, the light in Lin Wenjie’s office is still on; he is the vice president in charge of supply chain at a chip design company in Pengcheng, China.
The words just now came from his mouth.
On the other side of the video call is Chen Zhiming, the client manager at TSMC in Hsinchu, Taipei.
After hearing this, Chen Zhiming looked somewhat dazed; he had heard such words countless times before: we need to de-risk, we need to adjust orders.
Although the words are similar, there are many subtle differences between the past and the present.
The first is the language.
In the past, hearing such words would be in English Language, Japanese Language, Korean, or even Minnan dialect—anything but Mandarin Chinese.
These companies spread across the globe in the free world, under the White House’s baton, speaking to him in different languages with the same demands: we need to adjust the supply chain, we need to consider de-risking, we are dispersing orders from upstream and downstream of the supply chain to other countries.
Chip production orders still go to TSMC, but further downstream encapsulation and testing used to go to Mainland China’s Changdian, Jingfang Technology, Tongfu Microelectronics—these Chinese Companies have cost advantages.
But now we are considering de-risking, so we need to give orders to encapsulation companies in other countries to support their growth.
As for clients from Mainland China? These clients speaking Mandarin Chinese get whatever they want from him; forget de-risking—whatever TSMC gives them, they can swallow it.
But now, Lin Wenjie, with a professional smile on his face, had said earlier: “Our order reduction for the next quarter will be forty percent.”
Followed by de-risking.
Who is the risk?
TSMC?
How dare Chinese person? Such a sentence flashed through Chen Zhiming’s mind.
Inside TSMC, there are also two factions: locals and outsiders. The former want to strictly follow White House regulations, avoiding any risky operations; for orders from the mainland, we reject them if there’s even the slightest risk.
The latter are the opposite: as long as we can skirt the edges, we do it, under the slogan of “it’s foolish not to make money when it’s there,” providing certain support to China’s semiconductor industrial chain.
Chen Zhiming, assigned to handle Greater China business, is naturally the latter, but even he never imagined that China could really catch up, using the same slogans as foreign clients to de-risk TSMC.
“Wenjie, your company’s order reduction this time is no longer a simple inventory adjustment; it’s undermining our long-term cooperative relationship.” Chen Zhiming reminded.
At the same time, he felt faintly anxious inside.
Because it’s too fast.
So fast that he even suspects China has already overcome the 5nm process.
Chip manufacturing doesn’t appear out of thin air; it requires a process from testing to production, then gradually mastering the production process and ramping up.
The general process is: initially, conduct technology verification and trial production to ensure the process is feasible and solve core defects discovered.
Any tiny deviations in equipment parameters, chemicals ratio, temperature control, etc., can cause chip failure.
These must be resolved one by one.
Then comes yield rate improvement; once the process flow stabilizes, in this phase, the chip’s yield rate climbs from 5% to 50%—that’s phase two.
Phase three is production capacity ramp-up and cost optimization; full-speed production begins only in phase three, when yield rate has reached over 70% and stabilized.
In this phase, the wafer fab introduces automation systems to reduce human error and focuses on cost control.
Management will identify bottlenecks in production line equipment. For example, if a certain inspection equipment’s processing speed can’t keep up with the lithography machine, then procure new equipment or optimize the process.
Phase three’s yield rate climbs from 70% to 85%.
Finally, phase four, the mature phase: the process is mature, the wafer fab produces without R&D, and technology is invested in the next-generation process.
Whether Chen Zhiming or other semiconductor practitioners in East Asia, everyone thought China’s 7nm process wafer fabs were still in phase two, barely able to produce.
Neither yield rate nor stability could be guaranteed.
Now it seems things are not like that.
Lin Wenjie on the video leaned back, staring at the screen and said faintly: “No, it’s not undermining; it’s going with the flow.
Just like TSMC’s past cutoff to Huawei; we are just going with the flow.”
After hearing this, Chen Zhiming knew he had to do something to get more intelligence through words: “Wenjie, we understand your concern for supply chain resilience.
But you must admit, in yield rate, technology ecosystem, delivery stability, and cost, no mainland domestic wafer fab can compare to us now.
If you shift these forty percent of orders, do you dare guarantee your new-generation AI chips’ yield rate won’t plummet?”
He needed verification.
To verify his judgment: how many 7nm wafer fabs does China really have, and are they in phase two or phase three?
It couldn’t possibly be phase four, right? A dangerous thought entered Chen Zhiming’s mind.
Actually, he already had a guess in his heart: phase four.
Simple, because in China’s semiconductor map, Lin Wenjie’s company is not small in scale with strong backing capital, but it’s not important—to all of China, it’s not important.
There are too many similar companies; China is raising gu; as long as one emerges, that’s fine—who emerges? China doesn’t care.
To semiconductor practitioners in East Asia outside Mainland China, China’s semiconductor is most importantly Huawei, SMIC, and Moore Threads under Lin Ran’s name.
If even Lin Wenjie’s company can get forty percent of TSMC’s orders capacity, it means China’s 7nm production capacity is sufficiently abundant.
Abundant enough that semiconductor companies in Mainland China’s Yangtze River Delta, Pearl River Delta, Jingjinji can all share the benefits.
Lin Wenjie sighed, his tone softening a bit, with some apology to an old friend: “Of course I dare not guarantee.
I even know that this batch of shifted orders will initially bring at least 15% yield rate loss, and costs won’t have short-term advantages over yours.
But Zhiming, you must look at the problem from my perspective.”
He spread his hands, emphasizing: “For our H Company, the biggest risk is no longer chip yield rate, but cutoff.
If one day the external chains tighten, all our production lines will be choked, just like what Moore Threads once suffered: S90 could have been produced, but suddenly one day, an update on the White House official website turned their hugely invested S90 design to ashes.
This is not what we want, not what we can accept.
Especially today, with domestic 7nm wafer fabs already in production.
Now, China’s domestic wafer fabs tell us they can guarantee never to cutoff; even if yield rate is a bit low, even if price is a bit high, they can guarantee 100% security.
Do you think my superiors, board of directors, our shareholders will make what choice?”
Chen Zhiming fell silent; Lin Wenjie said a lot, he heard it all, but didn’t care, as if it didn’t exist at all.
His mind only echoed one point: initially will bring at least 15% yield rate loss.
This sentence was like a spell, instantly striking down his suspended heart.
It’s phase three.
Undoubtedly phase three.
China’s 7nm process has completed production capacity ramp-up.
Worse, China’s progress is too fast; Taiwan Stock will be harvested even more efficiently.
Any news of a technical breakthrough from the mainland will be exploited to the limit by the new MacArthur big brother in the north Japan.
The opponent seems born for this, seizing the best timing, using minimal funds for the sharpest harvesting.
On X, there’s already more than one post suspecting that China and America are teaming up to harvest them.
Even so-called local Taiwan independence factions are starting to doubt America.
You know, this was untouchable in the past.
Seeing Chen Zhiming’s face stiffen to an ugly degree, Lin Wenjie chuckled lightly: “Of course, it’s not unnegotiable; if TSMC can convince the White House to do 3nm foundry for us, I think we still have plenty of cooperation opportunities.
Even including shifting 7nm orders, we can talk.”
Chen Zhiming’s fallen heart immediately suspended again: “Wenjie, why 3 nanometer instead of 5 nanometer?”
He didn’t even have time to nitpick Lin Wenjie’s wording—what does “TSMC convince White House” mean? Can defiance succeed?
Lin Wenjie explained: “Because the gap between 7nm and 5nm is not big enough to be a decisive chip, a reason for us not to support mainland companies and continue giving orders to TSMC.
You need to convince the White House; we similarly need to convince Yanjing.
Writing reports is not easy.”
Good, good, not that you’ve broken through to 5nm, Chen Zhiming thought; this is the only not-bad news that’s good news today.
Chen Zhiming then smiled bitterly: “TSMC convince White House? That’s the funniest joke I’ve heard recently.”
The two looked at each other in silence.
Clearly both speaking Mandarin Chinese, yet controlled by someone ten thousand miles away speaking English Language.
The next day, TSMC executives from around the globe boarded flights back to Hsinchu headquarters.
The meeting was held late at night; the crisis response room next to the president’s office was brightly lit.
On the huge screen, several key charts were displayed side by side:
China wafer foundry production capacity ramp-up curve: 7nm yield rate curve showed abnormally steep growth in the past six months, crossing the 50% profit line, approaching the 75% stability line.
TSMC next two years revenue forecast: curve brutally adjusted downward, especially at 7nm and 10nm nodes, expected revenue loss of billions of US Dollars.
Both were urgent speculations rushed out by TSMC’s strategy analysis department.
On the geopolitics risk map, the lines between Taipei, Washington, and BJ flashed red light.
In the room sat several of the company’s top core figures: President Wei Zhekai, several co-CEOs, and senior vice presidents in charge of R&D and business.
“The intelligence brought back by Zhiming gives us a clearer judgment of the situation.” Wei Zhekai broke the silence: “Lin Wenjie’s words have confirmed our worst guess: initially will bring at least 15% yield rate loss.
This means China’s domestic 7nm process has completed phase three’s production capacity ramp-up.”
“Not running data in the laboratory, not small-batch trial production,” he tapped the table with his finger, emphasizing, “but has entered the commercialization initial stage of large-scale mass production, honing and improving yield rate with real money from major clients.
At this moment, right now, SMIC’s identity has shifted from chaser to competitor.
SMIC, backed by all of Mainland China, can now threaten us.”
Wei Zhekai felt deep fatigue inside, fatigue from the White House’s strangulation.
In the past, China’s chasing didn’t pressure TSMC much; they hadn’t even mastered 28nm, and after getting 28nm from Japan, it wasn’t a big issue, because TSMC’s profit mainly came from advanced processes.
That is, 7nm and below, accounting for seventy percent of revenue; 28nm process only a negligible seven percent.
But 7nm is different; it’s a threshold, with 17% of revenue.
What worried Wei Zhekai more was the constant meddling from the board of directors and Aethelred Capital representative cooperating with the White House, urging them to expand investment in America.
They not only demanded TSMC increase investment in America, demanding TSMC bring engineers and their families to Arizona Phoenix, with America green cards fast-tracked.
They also had more excessive demands: requiring TSMC to jointly R&D with old rival Samsung and new rival Rapidus.
Called joint R&D, but actually technology transfer; these two companies have no technical bottlenecks—they are free world companies; what they need transferred is process, TSMC’s mature advanced process.
The latter, Japan’s Rapidus established in 2022, even caused a scandal by instructing TSMC employees to sneak photos of internal process flow.
Finally, under Wall Street and Silicon Valley coordination, constant short selling and going long on Taiwan Stock against TSMC.
These combined form America’s systematic strangulation of TSMC.
This is a scenario Wei Zhekai never imagined after taking TSMC from Morris Chang.
Can Americans be so shameless? Use such tactics against a company? Can Chinese give Americans such pressure? Forcing them to exhaust ways to secure their moat in semiconductors?
These now hit one after another, giving Wei Zhekai factual answers.
The R&D vice president, a technical iron man who worked at TSMC for thirty years, now appeared unusually angry.
“This is a betrayal of engineering ethics!” he said loudly: “We spent a full five years, mobilizing the world’s top-tier talent and over twenty billion US Dollars in capital to complete the 7nm ramp-up; China actually did it in one year, leaping from 28nm foundation to this stage! I seriously suspect issues inside us; I think we need a thorough investigation of internal employees’ loyalty.”
His tone lowered: “Even all second- and third-generation outsiders, we need to investigate.”
A co-CEO countered: “Oh? According to our intelligence, China didn’t use ASML’s lithography machines, nor photoresist from Sumitomo Chemical or Shin-Etsu Chemical, nor Zeiss’s lens system.
First time I know our 7nm process suppliers have all achieved self-development.
Manager Chen has made such great merit for the company; I must speak for you at the board of directors!”
“That’s not what I mean; I mean our processes and equipment differ, but technology paths are the same; suppliers differ, but the engineering principles in the process are the same—just like when Liang Mengsong defected to Samsung, quickly helping them master 14nm, causing us a ten billion US Dollars loss.
The problem behind is the process.” The R&D vice president retorted.
The co-CEO raised his tone: “Oh? Then we allow Japan and Korea’s engineers on our production lines, to watch our production, attend our production discussion meetings, view parts of our production records.
Is this leaking secrets? Need investigation? I see the local engineers involved were very enthusiastic, even learning ninety-degree bows to Japan’s engineers—more enthusiastic than to me.”
Seeing these two about to continue arguing endlessly over locals vs. outsiders, loyalty vs. betrayal, Wei Zhekai slammed the table hard: “Enough, this is going off track.
What we need to discuss now is how to respond, not find enemies internally.”
The business senior vice president quickly shifted the topic: “H Company only shifted forty percent this time, but it’s the wind direction.
Next, other mainland domestic design companies will follow.
Soon we’ll face dual squeeze: high-end chased by Samsung and Rapidus, mid-end swallowed by China domestic fabs; worst, this loss is irreversible.”
Wei Zhekai looked at everyone, tone calming: “Now, we must focus on the two chips Lin Wenjie proposed.”
He pointed to the text record on the screen:
“If TSMC can convince White House to do 3nm foundry for us…”
“Because gap between 7nm and 5nm not big, can’t be decisive chip…”
After a full hour of discussion.
Finally they reached consensus: on one hand, increase investment to accelerate advanced process R&D; on the other, agree to expand America investment, move one most advanced 2nm process production line to America, but in exchange for White House permission to sell 5nm to mainland.
“Engineer Liang, come sit.” Lin Ran pointed to the seat in front, saying to Liang Mengsong.
Sitting opposite was a true craftsman, a man with the soul of an engineer; if he were Japanese, calling him chip immortal would be no exaggeration.
Since joining SMIC as co-CEO in 2017, under his leadership, SMIC’s process R&D speed achieved astonishing leap; even without Lin Ran, SMIC achieved 7nm chip manufacturing capability without EUV and other cutting-edge equipment, small-batch for specific products.
Most importantly, the system; he built a complete semiconductor manufacturing and R&D system for SMIC.
Of course, technology is one aspect; what Lin Ran admired more was his character—donating 900,000 to establish Mengning Scholarship at HDU just two years after joining SMIC.
Very different style from other 4v semiconductor elites coming to mainland.
“General Manager Lin, our CG-1 lithography machine is still in technology verification; theoretically, it has no issues, just poor stability.
This metasurface lens system is indeed very advanced, high precision, but it has its problems too.
That is, we can’t easily adjust it; this lens system is too precise—any tiny temperature, vibration, or particle changes cause optical distortion.
Once distorted, we must stop production and redo weeks of alignment and calibration.
Cost of one calibration even exceeds producing 7nm wafers for a month.
Compared to traditional semiconductor production equipment, this technology path’s biggest problem is high debugging cost, both in time and money.” Liang Mengsong explained.
Just as TSMC seeks White House permission to sell 5nm chips to China, China has already reached the 4nm threshold.
This lithography machine named CG-1, though not EUV lithography machine since China hasn’t mastered the light source yet, is already the limit of traditional DUV lithography machine path.
“Engineer Liang, you’ve made an absolutely stable engineering mistake.” Lin Ran reminded softly.
Liang Mengsong was stunned, looking puzzled at Lin Ran.
In past collaborations, Liang Mengsong knew Lin Ran might not be as proficient in process as him, but his wild ideas, especially mathematical solutions to engineering, were absolutely Blue Star’s best.
Their internal team jokingly called: if Lin Ran can’t solve a math problem, no one on Earth can.
“We used metasurface lens’s ultimate precision to achieve 4nm theoretical resolution.
But you’re trying traditional static passive stability methods, whether isolating vibration or constant temperature and humidity to tame it.
We spend endless time debugging, pursuing an ideally eternally unchanging perfect focus.”
Lin Ran raised his hand, pointing to the air, as if CG-1 lithography machine was right there.
“But CG-1’s lens group can never be as stable as ASML’s pure quartz lenses.
Its materials, its structure, are destined to be more sensitive to environment.
Whether now or long-term, we must abandon the fantasy of absolute stability and turn to dynamic compensation.”
