Tag: tsmc

Samsung Foundry drops further behind TSMC after disasterous Q4
TSMC is the largest contract chip foundry in the world. Looking at its roll call of customers, you can see why it is number one. After all, the Taiwan-based foundry manufactures chips based on designs submitted by companies such as Apple, Nvidia, AMD, MediaTek, Qualcomm, and Broadcom. Data compiled by TrendForce that The Chosun Daily published says that during the fourth quarter of 2024, TSMC’s market share rose 2.4 percentage points to 67.1%.

That means approximately two out of every three chips produced during the last three months of the year were built by TSMC. As for second-place Samsung Foundry, its market share dropped 1 full percentage point from 9.1% to 8.1% during the same quarter. Sammy’s foundry has had to deal with poor yields on its advanced chips, which this year resulted in the use of Qualcomm’s Snapdragon 8 Elite for Galaxy application processor (AP) on every Galaxy S25 series unit.

A repeat could be in store for next year’s Galaxy S26 line unless Samsung Foundry can hike its yield on 2nm production from the current 30% before mass production of the Exynos 2600 AP begins. If there is enough of an improvement, the Exynos 2600 SoC will power the Galaxy S26 and Galaxy S26+ in all markets except for the U.S., China, and Canada.

But we digress. The gap in market share between TSMC and Samsung Foundry increased sequentially from 55.6% in Q3 to 59% in Q4. TrendForce says that TSMC is benefiting from the AI buzz as demand for Nvidia’s GPU chips used for AI training, inference (predictions and decisions), and computing picks up. Those chips are made by TSMC.

Also helping TSMC is the demand for high-end smartphone chips including those used to power the iPhone, and Samsung’s flagship Galaxy S and foldable Galaxy Z models. Starting this year the new Tensor G5 AP, designed by Google from the ground up, will be built by TSMC using its third-generation 3nm node (N3P) and deployed in the Pixel 10 line. That is the same node that TSMC will use to build Qualcomm’s Snapdragon 8 Elite 2 for next year’s Galaxy S26 Ultra.

You might be wondering which foundry is in third place behind Samsung Foundry. That would be China’s largest foundry, SMIC. While the latter appears to be handling the production of Huawei’s APs, it also could be making some money building Huawei’s Ascend AI accelerators. Long-time PhoneArena readers probably are aware that thanks to U.S. sanctions, SMIC cannot purchase extreme ultraviolet lithography machines preventing SMIC from building chips using a process node more advanced than 7nm.

TSMC generated $26.85 billion in revenue during Q4 2024, a gain of 14.1% year-over-year. Samsung Foundry’s Q4 revenue declined 1.4% to $3.26 billion.
Lees ook:
- Stablecoins Boost Capital Flight From China
March 13, 2025
Apple chip builder TSMC not yet ready to move to next-gen lithography system
Back in December, Dutch tech giant ASML shipped its first High-NA Extreme Ultraviolet (EUV) lithography machine (EXE:5000) to Intel. The $400 million machine will take us to the next tier of chip production with a process node of 2nm and less. The original EUV machines, also made by ASML, were necessary to get chip foundries to manufacture components under 10nm. A lower process node means smaller transistors, meaning more can fit inside a chip. The higher a chip’s transistor count, the more powerful and/or energy efficient it is.

The reason that the EUV machine is so important is that it prints circuitry patterns on silicon wafers that are thinner than human hair. This is required when you are building a chip with billions of transistors inside. The 7nm A13 Bionic SoC, used to power 2019’s iPhone 11 series, contained 8.5 billion transistors. The 3nm A17 Pro used to run the iPhone 15 Pro and iPhone 15 Pro Max sports 19 billion transistors.

The 8 nm resolution of the EXE:5000 means that chipmakers can pack more transistors into a single chip. Smaller transistors are more energy efficient – that means the chips will be able to do more with less.-ASML

The new High-NA EUV machine has a 0.55 numerical aperture (NA) lens giving it a resolution of 8nm compared to the current machines’ resolution of 13nm (.33 NA). This means that the new machines can print transistors 1.7 times smaller resulting in transistor densities 2.9 times larger with a single exposure. The result? More powerful or energy-efficient chips. The new machines can also print 185 wafers per hour, rising to 220 by 2025. That compares to the 160 wafers per hour that can be printed with the current EUV machines.

The current Low-NA EUV machines can produce the same resolution but only after two exposures are made using double patterning. However, there are risks with double patterning including longer production times, and increased risk that a defect will occur. It can also lead to performance variability among manufactured chips.

ASML is quick to tell you what can go wrong with double patterning since it would prefer foundries spend the bigger bucks on the newer machines. But TSMC’s N3B process node, which supposedly relies on double patterning, was used to manufacture the A17 Pro application processor used on all of those iPhone 15 Pro and iPhone 15 Pro Max units made by Apple, and the M3 chip used to power high-end Macs. ASML says that its clients are just now doing their research into the High-NA EUV.

Financial institution China Renaissance (via

Eventually, TSMC, Samsung Foundry, and others will have to join Intel and start making the investments needed to move to the High-NA EUV. That time might end up being soon with TSMC and Samsung looking to start 2nm production in 2025 moving to 1.4nm production in 2027.
Lees ook:
February 1, 2024
TSMC shows off 2nm chip prototype to Apple

TSMC is the largest foundry in the world and Apple is its biggest customer. Apple was able to reserve most of TSMC’s initial 3nm production for the A17 Pro chipset which is being used to power the iPhone 15 Pro and iPhone 15 Pro Max. To keep things simple, just remember that as the process node number gets lower, the transistors used by a chip get smaller in size which means that more of them can fit into the small space inside an SoC. The higher a chip’s transistor count, the more powerful and/or energy-efficient it is.

Looking at the iPhone as an example, the iPhone 11 series was powered by the 7nm A13 Bionic which carried 8.5 billion transistors. The 3nm A17 Pro has 19 billion transistors. And since it takes a few years to build the fabs and order the machinery needed, we’ve known for some time that TSMC is working toward 2nm chip production. Apple has already had a chance to check out 2nm prototypes built on TSMC’s N2 node. The report noted that another huge TSMC customer, NVIDIA, also was able to view the 2nm prototype chip.

TSMC previously said that it would start 2nm volume production in 2025 and in a statement to The Financial Times, the company said that it is “progressing well and on track for volume production in 2025, and will be the most advanced semiconductor technology in the industry in both density and energy efficiency when it is introduced.” Assuming that there are no delays, the iPhone 17 Pro and iPhone 17 Pro Max could be the first Apple handsets to be powered by a 2nm application processor (AP), possibly the A19 Pro.

At 2nm, TSMC will debut its new Gate-all-around (GAA) transistors which cover the channel on all four sides reducing current leaks and delivering gains in power efficiency. Samsung Foundry already uses GAA with its 3nm production.

Speaking of Samsung Foundry, the latest word is that Qualcomm will be switching from TSMC to Samsung to produce the 3nm Snapdragon 8 Gen 5 AP in 2025. Qualcomm left Samsung Foundry in 2022 after the latter reported low yields building the Snapdragon 8 Gen 1. TSMC took over to produce the Snapdragon 8+ Gen 1 and has had the business ever since, although that seems likely to change in 2025. Samsung Foundry’s yield is about 60% for the most basic of 3nm production. The yield is expected to drop when building APs for smartphone.

At a 60% yield, 40 out of every 100 dies cut from a wafer fails to meet quality control. And since the foundry’s customers are usually responsible for the cost, yield is a major factor when determining which foundry a chip designer does business with.

Samsung says that it is ready for 2nm production. “We are well-equipped to set up for SF2 mass production by 2025,” Samsung said. “Since we were the first to take the leap and transition to GAA architecture, we are hoping the progress from SF3 to SF2 will be relatively seamless.” While Apple and NVIDIA were able to get a look at their future, TSMC is putting together the final list of its 3nm and 2nm clients according to DigiTimes.

Intel is also looking to grab some contract business away from TSMC and Samsung Foundry. Intel’s next-gen 18A node (1.8nm) could take process leadership away from the pair and it is offering chip design firms free test production. Intel could shake up the market. Longtime TSMC customer AMD said in July that it would “consider other manufacturing capabilities.”

Leslie Wu, chief executive of consulting firm RHCC, said that companies looking to manufacture their chip designs using a 2nm node might start having their chip production spread across multiple foundries. With the concern about China invading Taiwan always something to worry about, Wu said, “It’s too risky to rely on TSMC solely.” Yet Apple has been relying solely on TSMC for years.
Lees ook:
December 14, 2023

Intel gets confidence boost from anonymous customer pre-paying for 18A production

Last week, Intel CEO Pat Gelsinger reiterated his goal for Intel to take foundry leadership from TSMC by 2025. The executive said, “We’re 2.5 years into the transformation. Now, it’s gone the way I would have expected at the time in terms of rebuilding the company. You have to be much less skeptical about our ability to pull this off.” Intel not only had to watch as TSMC took its foundry leadership away, but it also saw TSMC’s largest customer, Apple, replace its Intel chips with M-series chips built by TSMC.

During last week’s Deutsche Bank conference, Gelsinger stated that the 18A node will return process leadership to Intel. That node would be considered 1.8nm while TSMC and Samsung would be shipping at 2nm during the same time. The CEO also let it be known that Intel has received a large prepayment from a customer for Intel’s 18A capacity. This gives Intel the confidence to believe it is on the right track and will lead the company to speed up the build-out of 18A fabs.

In the long term, Intel sees the contract foundry business as its biggest opportunity. One interesting comment made by Gelsinger is that Intel knows TSMC’s wafer costs, ASPs, and targets. With this knowledge, he says that Intel wants lower costs than TSMC to win business from the global foundry leader. He says that Intel is trying to get its internal cost structure in line with TSMC. To this end, Intel will study its metrics such as headcount per wafer start, and use more AI/Machine Learning to improve efficiency.

Intel is still relying on TSMC to build parts of its next-generation ‘Meteor Lake’ chips. While Intel will use its Intel 4 node (7nm) to build the chip’s CPU tile, the GPU tile will use TSMC’s 5nm node. The chip’s SoC tile, an ultra-low-power tile that supports media, imaging, display, and the connection to memory, will be built on TSMC’s 6nm node as will the chip’s I/O Extender tile.

As Tech journalist Leo Waldock notes in a post on X, “All that could be correct yet it still seems weird to me that Intel is buying tiles from TSMC for its own products while touting foundry services to 3rd parties. Eat your own dog food, no?”

Will Intel make TSMC and Samsung Foundry squirm? Will Apple stay loyal to TSMC? Will geopolitical rumblings make TSMC customers worried enough to bring their business to another foundry? These are questions that we will have to wait to answer.

September 4, 2023

Apple paying TSMC special rate for 3nm A17 Bionic

TSMC’s yield rate on its 3nm production is said to be at 55%. At that rate, a bit less than half of the silicon wafers used to produce Apple’s A17 Bionic and M3 chips are frisbees or extra-large drink coasters. Remember, the iPhone 15 Pro and iPhone 15 Pro Max will be the only smartphones powered by a 3nm chipset this year due to the high price of the wafers.

But Apple reportedly has worked out a deal with TSMC and will pay only for known good die rather than the $17,000 per wafer price. But Brett Simpson, senior analyst at Arete Research, provided EE Times with a report in which he said that once yields get to 70%, TSMC will put its most lucrative client back on standard wafer prices.

Simpson wrote, “We think TSMC will move to normal wafer-based pricing on N3 with Apple during the first half of 2024, at around $16-17K average selling prices. At present, we believe N3 yields at TSMC for A17 and M3 processors are at around 55% [a healthy level at this stage in N3 development], and TSMC looks on schedule to boost yields by around 5+ points each quarter.”

The report from Arete Research says that the A17 Bionic requires 82 mask layers and with a die size in the range of 100-110 square mm, each wafer can yield 620 chips with a wafer cycle time of four months. That’s the time it takes a wafer lot (usually 25 wafers) to move from start to finish in a fab. The report added that the M3 chip is likely to be around 135-150 mm square die size and yield up to 450 chips per wafer.

The first generation 3nm chips from the foundry are using the N3B process node for the A17 Bionic. In 2024, Apple might switch to the N3E node for the A17 Bionic which will have lower production costs and higher yields. The only downside, according to one tipster, is that it supposedly delivers less of a performance increase than the N3B node delivers.

This is a rumor, but even so, it isn’t clear whether Apple would stick with the N3B node to manufacture the A17 Bionic and M3, or switch to the less-expensive but slightly less impressive N3E node. If Apple decides to go with the N3E node, you might see prospective iPhone 15 Pro and iPhone 15 Pro Max buyers trying to figure out which variant of the A17 Bionic SoC is inside the phone they are about to buy.

TSMC CEO C.C. Wei said during a conference call with analysts, “Our 3-nm technology is the first in the semiconductor industry to high-volume production with good yield. As our customers’ demand for N3 (3nm) exceeds our ability to supply, we expect N3 to be fully utilized in 2023, supported by both HPC and smartphone applications. Sizable N3 revenue contribution is expected to start in the third quarter, and N3 will contribute a mid-single–digit percentage of our total wafer revenue in 2023.”

TSMC will start 2nm production in 2025

Mehdi Hosseini, senior equity research analyst with Susquehanna International Group, says that in the battle between TSMC and Samsung Foundry, TSMC remains on top. “TSMC, in our view, remains the preferred foundry choice for leading-edge nodes as Samsung Foundry has yet to demonstrate a stable leading-edge process technology, all while IFS [Intel Foundry Services] is years away from offering a competitive solution,” he wrote.

In the foundry business, you don’t get a second to look back at your accomplishments. TSMC says that N2 production will start in 2025. TSMC’s Wei states, “At N2, we are observing a high level of customer interest and engagement. Our 2-nm technology will be the most advanced semiconductor technology in the industry in both density and energy efficiency when it is introduced and will further extend our technology leadership well into the future.”

The recent chip inventory correction has been worse than TSMC expected and the company said that it might report a drop in annual revenue (for 2023) which would be the first drop in a decade. The average inventory holdings for TSMC’s fabless clients (chip designers that don’t own a factory and turn to TSMC to make their chips) is 92 days. During Q4 of 2022, Nvidia had over 200 days of inventory with Marvell around 180 days and Qualcomm with about 160 days of inventory.

July 14, 2023

TSMC’s current 3nm chip production not enough to satisfy Apple’s needs

The world’s leading foundry, TSMC, is having problems meeting the demand for 3nm chips from Apple. The latter is TSMC’s largest customer, accounting for 25% of its revenue. Apple reportedly locked up all of TSMC’s 3nm production for this year and plans on debuting the 3nm A17 Bionic chipsets with the iPhone 15 Pro and iPhone 15 Ultra.

The smaller the process node, the smaller the chip’s feature set, including transistors. Smaller transistors mean that more can fit inside a chip, which is important because the higher its transistor count, the more powerful and energy-efficient it is. For example, the Apple iPhone 11 line was released in 2019 and was powered by the 7nm A13 Bionic which contained 8.5 billion transistors in each chip. Last year’s iPhone 14 Pro models were powered by the 4nm A16 Bionic SoC with a transistor count of 16 billion.

The aforementioned A17 Bionic that will power Apple’s premium iPhone 15 models this year will be made on the 3nm node and could include more than 20 billion transistors. The iPhone 15 Pro and iPhone 15 Ultra could be the only two phones from a major brand to use a 3nm chip under the hood this year. One reason is cost. With the price for each silicon wafer used for 3nm chip production tagged at approximately $20,000, moving to 3nm this year is an expensive proposition, especially with yields still improving.

As TSMC and Samsung jockey for 3nm leadership, TSMC CEO C.C. Wei recently spoke to analysts during a conference call and said, “Our 3-nm technology is the first in the semiconductor industry to high-volume production with good yield. As our customers’ demand for N3 (TSMC’s 3nm production) exceeds our ability to supply, we expect N3 to be fully utilized in 2023, supported by both HPC (High-Performance Computing) and smartphone applications.”

The executive added, “Sizable N3 revenue contribution is expected to start in the third quarter, and N3 will contribute a mid-single–digit percentage of our total wafer revenue in 2023.” The sizable N3 revenue contribution Wei sees in Q3 has to do with the release of the 2023 premium iPhone models.

While TSMC and Samsung are the top two chip foundries in the world, Intel has joined the fight and promises to regain process node leadership in 2025. Right now, that crown belongs to TSMC. Mehdi Hosseini, senior equity research analyst with Susquehanna International Group says, “TSMC, in our view, remains the preferred foundry choice for leading-edge nodes as Samsung Foundry has yet to demonstrate a stable leading-edge process technology, all while IFS is years away from offering a competitive solution.”

Besides the A17 Bionic, TSMC will also produce Apple’s M3 chip using the 3nm node. Brett Simpson, senior analyst at Arete Research, said in a report provided to EE Times, “We think TSMC will move to normal wafer-based pricing on N3 with Apple during the first half of 2024, at around $16-17K average selling prices. At present, we believe N3 yields at TSMC for A17 and M3 processors are at around 55% [a healthy level at this stage in N3 development], and TSMC looks on schedule to boost yields by around 5+ points each quarter.”

As is the case in the chip industry, there is no time to rest because you always have to look ahead. With 3nm heading for the iPhone this year, 2nm production will start in 2025. TSMC CEO Weil says, “At N2, we are observing a high level of customer interest and engagement. Our 2-nm technology will be the most advanced semiconductor technology in the industry in both density and energy efficiency when it is introduced and will further extend our technology leadership well into the future.”

Even with Apple’s 3nm business, 2023 is not shaping up as a good year for TSMC and revenue might fall this year for the first time in a decade. Year-over-year sales could decline by a mid-single–digit percentage point. Even with business slowing, the foundry expects capital expenditures to remain in the range of $32 billion to $36 billion.

The Apple A17 Bionic has a die size in the range of 100-110 mm square allowing it to produce 620 chips per wafer. With a die size of 135-150 mm square, TSMC’s yield for the Apple M3 is 450 chips per wafer.

April 28, 2023

Apple’s 5G modem will be built by TSMC using its 3nm process node

Apple has scooped up all of TSMC’s 3nm production capacity for this year. This will be the first shipment of 3nm chips made by the world’s largest foundry. To explain in simple terms what this is all about, a lower process node means smaller transistors are used. This allows more transistors to fit inside a chip and this transistor count is important. The more transistors in a chip, the more powerful and energy efficient it is.

For example, the A13 Bionic used in the iPhone 11 line back in 2019 was produced using the 7nm node and contained 8.5 billion transistors. The 5nm A14 Bionic used in the 2020 iPhone 12 series was equipped with 11.8 billion transistors. The following year, the A15 Bionic, made using the second-gen 5nm process node, was used on the iPhone 13 line and carried 15 billion transistors. And last year’s iPhone 14 Pro models feature the A16 Bionic. The latter was manufactured using a 4nm process node and the transistor count is close to 16 billion.

Since Apple has wrapped up all of TSMC’s 3nm production for 2023, it makes sense to expect Apple’s in-house 5G modem to be produced using the 3nm node. According to the Commercial Times, supply chain sources state that risk production of the modem will start in the second half of this year. This is when the foundry builds chips while still working out any issues. Yields are low. That is followed by volume production. Wafer output is expected to increase slowly in the first half of next year.

Based on this timeline, it appears that Apple will continue using Qualcomm’s Snapdragon 5G mobile chips this year. The iPhone 14 line uses the Snapdragon X65 5G modem chip while the iPhone 15 series will most likely be equipped with the Snapdragon X70 5G modem chip. This would dovetail with reports from reliable analyst Ming-Chi Kuo, Qualcomm CEO Cristiano Amon, and reliable tipster Ross Young, all of whom expect Apple to debut its in-house 5G modem chip with the iPhone 16 series in 2024.

Apple was hoping to use its in-house 5G modem chips with the iPhone 15 series but patent issues and other problems have forced Apple to delay the launch of the component. In the summer of 2019, Apple spent $1 billion to buy most of Intel’s smartphone modem business and has been developing the component ever since.

While Apple executives reportedly had a high opinion of the quality of Qualcomm’s modem chips, it wasn’t happy with how Qualcomm sells these chips to smartphone manufacturers. First, Qualcomm demands that customers pay for a license (no license-no chips is Qualcomm’s mantra) and Qualcomm also charges for the chips themselves.

The Nothing Speaker’s design has leaked, showcasing a unique new product

March 9, 2023

Apple has sharply reduced the size of the order it has with TSMC

Taiwan Semiconductor Manufacturing Co. Ltd., better known as TSMC, is the world’s largest foundry. Only TSMC and Samsung currently mass-produce chips using their respective 3nm process nodes (which use different types of transistors, as we will explain later). Apple is TSMC’s largest customer and is believed to account for a quarter of the company’s revenue (that’s 25% for you numerical types).

With the overall weakness of the global economy and the hangover from last year’s chip shortage, some major brands have been canceling orders placed earlier with TSMC. And believe it or not, that goes for Apple as well. You know that we often cite Twitter tipsters and check out social media posts from Weibo writers for the sake of alliteration. Weibo is a Chinese social media site and one subscriber, whose user name translates to something like @CellPhoneChipExpert posted some big Apple news.

The tipster, who allegedly gets news directly from the semiconductor industry, says that Apple has reduced its orders with TSMC by as many as 120,000 wafers. The canceled orders were for chips that would be made using TSMC’s N7, N5, N4, and even some N3 nodes. Rumors call for the A17 Bionic, which is expected to be found under the hoods of the iPhone 15 Pro and iPhone 15 Ultra, to be produced using TSMC’s N3 (3nm) process node. Apple’s M2 Ultra and M3 chips might also use the same node.

Digitimes reports that TSMC has been able to beat out Samsung Foundry when it comes to signing up customers. Samsung’s 3nm node uses Gate-All-Around (GAA) transistors that allow the gate to come into contact with the channel on all sides resulting in less current leakage and higher drive current compared to the FinFET transistors used by TSMC for its 3nm node. GAA features vertically placed horizontal nanosheets while FinFET uses horizontally placed vertical “fins.” TSMC is expected to go to GAA for its 2nm production in 2025-2026.

Despite sticking with FinFET for 3nm production, TSMC still has a pretty full-order book with big names like Qualcomm, MediaTek, and Nvidia reserving production capacity for 2023 and 2024. Samsung, though, has been handicapped in its efforts to secure more business due to low yields. That means that a higher percentage of the chips it cuts from wafers fail to pass quality control.

The report says that in order to give Apple most of the 3nm production capacity later this year, Intel agreed to change its roadmap and delay receipt of its 3nm orders. TSMC’s enhanced 3nm process node (N3E) could be launched this year even though there might not be a big rush for orders considering the cost.

Digitimes points out that wafer prices have skyrocketed with 90nm wafers selling for about $2,000 in 2004. By 2016, 10nm wafers cost $6,000 and that price hit $16,000 for 5nm which debuted on consumer electronics in 2020. Wafer prices for 3nm chips are priced in the $20,000 area.

Intel has said that it will take process leadership from TSMC and Samsung by 2025 thanks to two developments. It will use RibbonFET transistors which is another term for the Gate-All-Around (GAA) transistors already employed by Samsung. And it will use a feature known as PowerVia or backside power delivery. This allows transistors to obtain power from one side of a chip while using the other side to connect to data communication links.

Today’s chip designs have transistors trying to handle both functions from the same side which makes the process more complicated and limits the use of miniaturization. Intel will also be the first to use high numeric aperture Extreme Ultraviolet Lithography (EUV). This machine will etch higher-resolution circuitry patterns on wafers which could reduce the time it takes to add additional features to a pattern.

February 20, 2023

TSMC suspends production of powerful GPU chip for Chinese tech firm

Taiwan Semiconductor Manufacturing Company Ltd. is known throughout the planet as TSMC. The largest foundry in the world produces chips based on the designs presented to it by companies like Apple, Qualcomm, Nvidia, MediaTek, and more. In fact, Apple is TSMC’s largest customer and accounts for approximately 25% of the company’s revenue.

TSMC currently produces powerful and energy-efficient chips such as the Apple A16 Bionic found inside the iPhone 14 Pro series, and the Qualcomm Snapdragon 8+ Gen 1 found in newer high-end Android phones including the Samsung Galaxy Z Fold 4, Galaxy Z Flip 4, and the Motorola Edge 30 Ultra. TSMC has suspended production for Chinese start-up Biren Technology.

The reason for halting production for this company is that TSMC is following U.S. regulations that prevent it from making chips for the Chinese-based firm. Part of the reason for this is that Biren’s products outperform Nvidia’s A100 Graphics Processing Unit (GPU) silicon based on what the English language South China Morning Post calls “information in the public domain.” The U.S. is trying to keep cutting-edge chips away from China.

This past September, the U.S. ordered that Nvidia stop shipping the A100 chip to China to “…address the risk that products may be used in, or diverted to, a ‘military end use’ or ‘military end user’ in China.” The A100, according to Nvidia, is used to “power the world’s highest performing elastic data centers for AI, data analytics, and high-performance computing (HPC) applications.”

The U.S. Commerce Department last month expressed a goal to “keep advanced technologies out of the wrong hands.” China called it a “tech blockade.” At the same time, commerce ministry spokesperson Shu Jieting said that “the U.S. continues to abuse export control measures to restrict exports of semiconductor-related items to China, which China firmly opposes.”

Biren was trying to raise funds earlier this year at a valuation of $2.7 billion. The company designs its BR100 and BR104 processors to be competitive with GPUs designed by Nvidia and AMD that work with AI and Machine Learning models and algorithms.

TSMC itself isn’t sure that Biren’s chips are covered by U.S. regulations but has decided to halt their production anyway. Biren, of course, says that its AI chips are not covered by U.S. export restrictions. A TSMC spokesperson made a limited statement noting that the foundry complies with all relevant rules. The U.S. Commerce Department’s Bureau of Industry and Security (BIS) announced new semiconductor restrictions on October 7th.

A spokesperson for the U.S. Commerce Department said, “While BIS cannot comment on company-specific actions, we expect all companies to comply with export controls. Since the rule’s release on October 7, BIS has been undertaking a vigorous outreach effort to educate those impacted by it to aid compliance efforts.”

One of the Biren GPU chips that TSMC was going to produce for the company was the BR100 GPU which was manufactured using TSMC’s 7nm process node and features 77 billion transistors in each chipset. This particular component was said to be 2.8 times faster than Nvidia’s A100.

Export rule changes have been used before to restrict the distribution of silicon to China. A restriction announced by the U.S. Commerce Department in 2020 prevents foundries using American technology to manufacture advanced chips from shipping these chips to Huawei. The latter is considered a national security risk by both major political parties and the restriction has forced Huawei to abandon its own Kirin 5G Application Processor (AP) chips. Its current flagship Mate 50 and Mate 50 Pro handsets are powered by the Snapdragon 8+ Gen 1. While this is Qualcomm’s current top-of-the-line mobile AP chipset, the chip sold to Huawei is tweaked to prevent it from working with 5G networks.

October 24, 2022

Samsung rumored to begin mass production of 3nm chips next week

Right now there are only two independent foundries that are able to manufacture cutting-edge chips. The companies are TSMC and Samsung Foundry. Independent foundries take chip designs created by other firms and build the actual chips off of that design. Both TSMC and Samsung are working on building chips using their 3nm process nodes. The smaller the process node used, the larger the number of transistors found inside a chip.

This is important because the more transistors employed in a chip, the more powerful and energy-efficient that chip can be. Every two years or so, the process node gets smaller and more transistors fit inside an integrated circuit. This is the famous “Moore’s Law” that you’ve heard about, named after Intel and Fairchild semiconductor co-founder Gordon Moore. Remember, this is not an actual law and as smaller and smaller components are being built, this “observation” no longer can be fully counted on to double the transistor count every other year.

Still, over time, you can see how “Moore’s Law” has forecast the incredible increase in processing capabilities over the years. Let’s take the iPhone X, which was released in November 2017 powered by the Apple A10 Bionic chipset. The latter carried 4.3 billion transistors in each chip. Now let’s move ahead to the 2021 iPhone 13 series which sports the A15 Bionic chipset. The A15 Bionic contains 15 billion transistors, up 27.1% from the 11.8 billion transistors employed by the A14 Bionic chip.

So now Samsung and TSMC are battling for supremacy in the business of building chips for third-parties. TSMC is number one in most metrics and its customer list includes top tech firms such as Apple (its number one customer), MediaTek, Nvidia, Qualcomm, and others. But it appears, according to ExtremeTech, that Samsung will soon beat out TSMC by starting mass production of chips made using its 3nm process node next week with TSMC starting 3nm mass production later this year.

Additionally, Samsung is going to use a new transistor structure on its 3nm chips called GAA or gate-all-around. With this structure, the flow of current is controlled by gates that contact the transistor on all four sides. TSMC will continue to use the FinFET structure which has been in place since the 22nm process node debuted. TSMC will finally scrap FinFET for GAA when it starts shipping 2nm chips in 2026.

Samsung’s GAA design is called Multi-Bridge Channel Field Effect Transistor (MBCFET) also known as nanowires. This is one of only two different gate-all-around designs currently available with the second one known as GAAFET or nanowire.

The report cites a major Korean news agency that says Samsung is expected to make a major announcement about its 3nm chip production shortly. It also notes that the move to gate-all-around from FinFet will reduce the area of a chip by 45% to help deliver a performance bump of 30% while reducing energy consumption by 50%. However, there is a big problem. Samsung was reportedly getting yields of only 10% to 20% at 3nm meaning that the vast majority of its 3nm chip dies cut from a wafer could not pass quality control.

Back in February, a report indicated that Samsung’s yield on 4nm production was only 35% resulting in Samsung losing some business from chip designer Qualcomm. The latter supposedly moved some of those orders to TSMC. However, if Samsung Foundry is about to announce the start of high volume manufacturing (HVM) at the 3nm process node, one could come to the conclusion that Samsung Foundry has improved its 3nm yield.

Speaking of 3nm, TSMC’s major customers such as AMD, Apple, Broadcom, Intel, MediaTek, Nvidia, and Qualcomm, have started to queue up for 3nm capacity. And next on the horizon, of course, is the 2nm process node that both TSMC and Samsung are working on. Another name is expected to join TSMC and Samsung on the list of cutting-edge foundries. Intel Chief Executive Pat Gelsinger has announced that the American company will regain process leadership from Samsung and TSMC by 2025.

June 27, 2022

TSMC founder Chang says that the U.S. has the world’s best chip designers

The two most advanced chip foundries in the world at the moment are Taiwan’s TSMC and South Korea’s Samsung. With both providing phone manufacturers with chips produced using their 3nm process node next year, TSMC recently said that it will start shipping 2nm chips to customers in 2026.

In theory, the lower this process node number goes, the smaller the transistors used in these chips become. That allows chip designers and manufacturers to put more transistors in these integrated circuits allowing them to be faster and/or more power-efficient.

The U.S., concerned about geopolitical events putting a crimp in the supply chain for chips (especially the possibility of China making a move on Taiwan) would love to become self-sufficient in the design and production of chips. Design is actually not an issue as U.S. firms like Apple, Qualcomm, Intel, and others have no problem designing chips (more on this later). But building a foundry takes oodles of money and time. It also requires proximity to a reliable supply chain.

TSMC is working with the U.S. to build a fab (a plant that manufacturers chips) in Arizona. The facility should start producing chips by 2024 although production will reportedly be limited at first to 5nm chips which will have been replaced by the 3nm process node by the time the fab starts running in the states.

But TSMC founder Morris Chang considers the U.S. attempt to be relevant in chip manufacturing to be “a wasteful, expensive exercise in futility.” According to The Register, Morris spoke as a guest of the Brookings Institution think tank and stated that the U.S. does not have the talent pool necessary to create a thriving business in the states manufacturing chips.

Chang cited Taiwan’s large population that helped TSMC become the world’s top independent foundry. While the U.S. moved away from producing manufacturing professionals, Taiwan was loaded with talent. As we mentioned earlier in this story, where the U.S. does have the talent is in chip design, something that Morris isn’t shy about saying.

The 90-year-old Chang has high praise for the chip design talent in the U.S. calling it “the best in the world.” He adds that “Taiwan has very little design talent, and TSMC has absolutely none.” As an independent foundry, TSMC wouldn’t be expected to produce its own designs since its job is to produce chips designed by other companies like Apple, Qualcomm, MediaTek, and more.

Talking about the costs of running a fab in the U.S., Chang says that the small plant in Oregon that TSMC has run for 25 years is making money but not enough to justify expanding the facility. “We were extremely naive,” said Chang, “in expecting comparable costs, but manufacturing chips in the US is 50 percent more expensive than in Taiwan.”

U.S. experts happen to agree with Chang’s take on the situation with one think tank expecting several thousand unfilled positions in any new U.S. based fabs. With Chang’s comments in mind, why would TSMC lay out $12 billion to build the Arizona factory? “We did it at the urging of the US government, and TSMC felt we should do it,” Chang said.

Morris also pointed out that while chip production state-side will certainly see an increase, “there will be a high per-unit cost increase, and it will be hard for the US to compete internationally.” Chang also worries about a possible war between China and Taiwan. If there is a war, U.S. chip production might prove to be profitable with TSMC possibly not able to continue normal production.

He adds that if there is a war between China and Taiwan, “the U.S. will have a lot more than chip manufacturing to worry about.” Back in December, we passed along a stat that said 90% of the world’s most advanced chips are shipped from Taiwan. Worried about how vulnerable that makes the country, and in turn big chip consuming nations like the U.S., foundries need to work in advance on contingency plans in case Taiwan-based fabs are damaged or destroyed in a war.

April 22, 2022

TSMC continues to dominate the global foundry business with Samsung a distant second

TSMC’s market share during the fourth quarter of last year was 52.1% based on its revenue of $15.7 billion during the three-month period. That was down from the 53.1% share of gross that the foundry had during the previous quarter, but was still well ahead of the next name on the list.

Behind TSMC was its biggest rival in the industry (at least for now), Samsung Foundry. The latter had 18.3% of industry revenue at the end of the last quarter of 2021. Even with strong 15.3% growth in its top-line on a quarter-over-quarter basis, Samsung’s gross for the period from October through December was only 35% of TSMC’s.

And things are about to get worse for Samsung. Low yields on the production of some Application Processors (AP) by Samsung, about half the yield rate achieved by TSMC, has led Qualcomm to move over production of some of its most powerful Snapdragon chipsets to TSMC from Samsung. Additionally, Samsung is investigating the disappearance of funds that the foundry unit had supposedly set aside to improve chip production yields.

Trend Force notes that while Samsung Foundry is showing strong revenue growth, the bottom line has not been able to stay in gear as “the slower ramp-up of advanced process capacity continues to erode overall profitability.”

TSMC’s results for the quarter were bolstered by strong growth in revenue connected to the production of 5nm chipsets for the iPhone 13 series. The only TSMC process node that showed a drop in the top line for the fourth quarter was the 7nm/6nm unit which was impacted by a weak smartphone market in China. Samsung’s strong 15.3% hike in sequential revenue came about from the completion of its 5nm/4nm advanced process node capacity.

After the well-known pair of TSMC and Samsung, Taiwan’s United Microelectronics Corporation (UMC) was next as its Q4 revenue comprised 7% of the industry’s revenue for the three months. For the quarter, it garnered $2.12 billion in gross, up 5.8% from its third-quarter revenue. In fourth place, GlobalFoundries kept the 6.1% share of global foundry revenue that it earned during Q3. At number five was China’s largest foundry, SMIC, with a market share of 5.2% and Q4 revenue of $1.58 billion.

What holds back SMIC is its inability to compete at the current cutting-edge process nodes of 5nm/4nm with both TSMC and Samsung knocking at the door of 3nm. Like the U.S., China is desperately looking to become self-sufficient in semiconductors, and its failure to do so worries experts who fear that China could use TSMC’s success as a reason to take over Taiwan and capture control of the world’s largest contract foundry.

The industry is far from balanced in terms of revenue with the top five foundries controlling 90% of global foundry market share. That is understandable though when you consider that TSMC and Samsung are the leaders in advanced process nodes.

The remaining foundries making up 6-10 on the list and their Q4 market shares include HuaHong Group (2.9%), PSMC (2%), VIS (1.5%), Tower (1.4%), and Nexchip (1.2%). The latter might have reported the lowest market share of global foundry revenue for Q4, but it did show the largest growth rate in revenue quarter-over-quarter at 44.2%.

For the fourth quarter, global foundry revenue came in at $29.55 billion, up 8.3% from the third quarter.

One of the reasons for TSMC’s success is its long working relationship with Apple. For every $100 in revenue collected by the foundry, $26 comes from Apple. Just last week Apple announced that by combining a pair of its M1 Max SoCs, it created the M1 Ultra which contains a whopping 114 billion transistors inside.

March 18, 2022

Samsung execs accused of doctoring 5nm chip yield results to hide stolen funds

On the heels of a report claiming that Qualcomm is so upset about the low 35% yield achieved by Samsung Foundry’s 4nm process node that it dumped Sammy for TSMC, another alarming report has surfaced. Samsung is investigating the possibility that fraud occurred inside the company’s foundry business.

Unnamed Samsung Foundry executives are being accused in the Korean media of fabricating the yield rate data achieved by Samsung Foundry for its 4nm and 5nm process node. This data, expressed as a percentage, shows how many chips in a wafer meet quality testing standards. Yesterday’s report indicated that the yield for Samsung Foundry’s 4nm process node was a poor 35% compared to the 70% yield achieved by rival TSMC.

The poor yield was mentioned in a report as the reason why Qualcomm supposedly replaced Samsung Foundry with TSMC for the production of next year’s Snapdragon 8 Gen 2 Application Processor. The new report claims that the 4nm and 5nm yields were faked by Samsung Foundry executives to make it seem that everything was going well with the division.

Samsung officials are trying to track down funds that were supposed to be used to improve the yield at Samsung Foundry. Reports of the poor yields and missing funds at Samsung Foundry come at a poor time for Samsung since it and TSMC are the only two foundries in the world able of churning out chips at a process node of under 5nm. Both are working hard to become the first foundry able to ship 3nm components.

TSMC already has plenty of business as its customer list includes heavyweights such as Apple, MediaTek, Nvidia, and more. TSMC is believed to be running into yield problems of its own at the 3nm process node although volume production is only supposed to be reached later this year.

The lower the process node, the higher the number of transistors that can fit in a chip. That is key to the performance of the component since the more transistors used, the more powerful and energy-efficient a chip is.

A Samsung official familiar with the situation said in a quote translated by Machine Learning, “Since the delivered quantities is struggled to meet the recent foundry order volume, we have doubts about the yield of the non-memory process, which has was known to be achieved.”

The official added, “The management consulting investigates the claims on the yield of a semiconductor foundry by former and current DS division executives. The consulting will determine whether the claims are false.”

If this story sounds strange and bizarre, that’s because it is. Frankly, we can’t think of a story that is similar to this one. And that means that until more official statements are issued by Samsung or even the police in Korea, we should take this report with a grain of salt.

TSMC To Expand New Japan Chip Factory, Denso Takes Stake

Intel to meet this month with TSMC to avoid a fight with Apple