TSMC has halted shipments to Chinese chip designer Sophgo after a chip it manufactured was reportedly found in a Huawei AI processor, according to Reuters, which cited two sources familiar with the situation.

The Reuters report noted that Sophgo had ordered chips from TSMC identical to the one detected in Huawei’s Ascend 910B processor. Huawei, restricted from acquiring certain technology to protect U.S. national security, is under stringent export controls. However, Reuters reported that it remains unclear how the chip ended up in Huawei’s product.

On Sunday, Sophgo posted a statement on its website asserting it has never engaged in any direct or indirect business relationship with Huawei and conducts its operations in strict compliance with all applicable laws, including U.S. export control regulations. The company affirmed it has never breached any of these laws or regulations.

Sophgo, which is affiliated with cryptocurrency mining equipment maker Bitmain, also noted it had submitted a detailed investigation report to TSMC to confirm its non-involvement with Huawei.

TSMC declined to comment, according to Reuters, while Huawei has not yet responded to requests for comment. The U.S. Department of Commerce acknowledged awareness of potential export control violations but declined to comment on any active investigations.

(Photo credit: Sophgo)

According to a report from Bloomberg, TSMC’s first wafer fab built in Arizona, USA, has achieved a major breakthrough in early production yield, surpassing that of similar factories in Taiwan.

The report in Bloomberg indicates that, citing the words from TSMC’s U.S. division president Rick Cassidy, the yield of chips produced at TSMC’s facility in Phoenix is about 4 percentage points higher than that of comparable facilities in Taiwan.

The recent yield improvements are significant for TSMC, which has traditionally maintained its most advanced and efficient facilities in Taiwan. According to Bloomberg, TSMC’s Arizona plant faced early challenges due to a shortage of skilled workers for advanced equipment installation and issues related to safety and management. However, TSMC resolved these setbacks by reaching an agreement with construction labor unions late last year.

Cassidy further pointed out that TSMC may intend to further expand its presence in the United States, partly depending on whether the government will provide more subsidies, referring to the early conversations in Washington about a second CHIPS Act. He mentioned that there is space for at least six total fabs at the Phoenix complex, according to the report in Bloomberg.

Earlier this year, the U.S. government has officially announced that it will provide subsidy about USD 6.6 billion to TSMC, and TSMC will build its third fab in Arizona.

According to the press release from TSMC, Arizona’s first fab is on track to begin production using 4nm technology in the first half of 2025. The second fab will produce the world’s most advanced 2nm process technology with next-generation nanosheet transistors in addition to the previously announced 3nm technology, with production beginning in 2028.

The press release further states that the third fab will produce chips using 2nm or more advanced processes, with production beginning by the end of the decade.

The report in Bloomberg pointed out that, citing the words of Chief Executive Officer C.C. Wei, TSMC expects volume production of its first fab in Arizona to start in the beginning of 2025, and are confident to deliver the same level of manufacturing quality and reliability  in Arizona as from fabs in Taiwan.

On the other hand, according to the report in Bloomberg, Samsung’s investment in the U.S. is facing challenges. Meanwhile, Intel, despite being a major beneficiary of the CHIPS Act, is experiencing financial difficulties due to delays in global projects and may consider selling off assets.

According to a report from MoneyDJ, TSMC previously has announced that its first US fab in Arizona has begun producing engineering wafers using the 4nm process in April but did not provide additional details about the yield. Investors are concerned about the yield and expect that the company’s gross profit to maintain stable.

Addressing this issue, according to the report from MoneyDJ, TSMC mentioned that the gross profit rate will remain 53% or even higher, and the net profit margin has maintained above 36% in the last four years.

(Photo credit: TSMC)

The discovery of TSMC-manufactured chips in Huawei devices has sparked significant concern among U.S. lawmakers. John Moolenaar, Chairman of the U.S. Congressional Committee on the Chinese Communist Party (CCP) and a Republican Congressman, criticized the reports, calling it a “catastrophic failure” of U.S. export control policies. He urged both the U.S. Commerce Department and TSMC to promptly explain the affected parties and the scale of the incident.

In his statement, Moolenaar said, “Reports that cutting-edge TSMC-manufactured chips have contributed to Huawei’s AI development represent a catastrophic failure of U.S. export control policy. AI accelerators, like the one these chips powered, are at the forefront of our technological competition with the CCP, and I fear the damage here will have significant consequences for national security. Congress needs immediate answers from both BIS and TSMC regarding the scope and volume of this disaster. The U.S. government must take immediate steps to ensure this does not happen again.”

The controversy revolves around a Huawei chip produced by TSMC. Last week, a report by The Information revealed that the U.S. Commerce Department is investigating whether TSMC has been involved in manufacturing AI chips designed by Huawei, which have become popular with Chinese customers as alternatives to NVIDIA chips, now restricted due to U.S. export regulations.

Following the Information report, a teardown of Huawei’s Ascend 910B AI chip revealed that it was manufactured by TSMC using its 7-nanometer process. In response, Reuters reported on the 22nd that TSMC had notified Washington about a possible attempt by Huawei to bypass U.S. export controls.

TSMC reportedly informed the U.S. Commerce Department after receiving an order for a chip similar to Huawei’s Ascend 910B, a processor designed for training large language models, according to Financial Times. How this chip ended up in Huawei’s possession remains unclear.

Before U.S. sanctions were imposed, TSMC had produced an earlier version of the 910B chip, Financial Times notes.

In addition, Liberty Times reported, citing industry insiders, that a Chinese design company suspected of acting as a proxy for Huawei submitted an order to TSMC earlier this year for a 7-nanometer project, paying several hundred million dollars in full for wafer production.

(Photo credit: TSMC)

According to a report from Wccftech, Google is rumored to switch to TSMC’s second-generation 3nm process, “N3E”, for its Tensor G5. The report also claims that for the Tensor G6, Google will utilize TSMC’s “N3P” process instead of the 2nm process speculated earlier. Its entire Pixel 10 series set to be launched next year will use TSMC’s 3nm process.

The report suggests that the Tensor G4 processor may be the last Google chipset manufactured by Samsung using its 4nm process, since the Tensor G4 reportedly offers only a slight upgrade compared to the Tensor G3 in the Pixel 8 smartphone, as it continues to use Samsung’s older FO-PLP packaging technology instead of the newer FO-WLP packaging, which is more capable in preventing overheating.

Notably, the report states that Qualcomm and MediaTek have also adopted the 3nm “N3E” process for their Snapdragon 8 Elite and Dimensity 9400 chipsets for the first time this year, having bridged the technological gap with Apple.

Therefore, the report notes that Google will still be a year behind in the competition when it announces the Tensor G5 and Tensor G6, since Qualcomm, MediaTek, and Apple are already using TSMC’s 3nm process and are expected to shift to N3P.

According to the report, the Tensor G5 was reported to have reached the tape-out stage earlier this year and is expected to use TSMC’s InFO-POP packaging, which allows the chipset’s packaging to be smaller and also more power efficient.

Regarding Tensor G6, it will likely launch two years from now. Although it was previously rumored that Tensor G6 will use TSMC’s 2nm process, the report indicates that it will instead utilize TSMC’s enhanced version of 3nm, N3P node.

According to a report from AnandTech, as a more enhanced process node compared to N3E, N3P offers better performance with higher transistor density and reduced power consumption.

The report indicates that it makes sense that Google is not opting to adopt the 2nm process immediately, especially since even Apple is expected to wait until the launch of the iPhone 18, two years from now, to introduce its first A-series chipsets that will exceed the 3nm barrier. Additionally, it is expected that the 2nm A-series chipsets will not be included in all iPhone 18 models due to their high cost.

Read more

• [News] TSMC Rumored to Secure Tensor G6 Orders from Google with Its 2nm Process
• [News] Google’s Tensor G5 Reportedly Manufactured with TSMC’s 3nm and InFO-POP Packaging

(Photo credit: Google)

According to a report from the Reuters, NVIDIA CEO Jensen Huang stated that the design flaw in the Blackwell GPU was entirely NVIDIA’s fault, dispelling rumors that TSMC was to blame. Huang emphasized that TSMC help fix the problem and resume the manufacturing “at an incredible pace.”

When initial reports of the design flaw emerged, some media outlets suggested that TSMC was at fault, speculating that this could strain the decades-long partnership between NVIDIA and TSMC. According to the Reuters, Huang referred to the reported tensions with TSMC as “fake news.”

A report from Tom’s Hardware indicated that the now-fixed Blackwell GPUs designed for AI and supercomputers are set to enter mass production in late October and are expected to begin shipping early next year.

Aside from addressing the design flaw in Blackwell AI chips, Huang also commented on the state of AI in Europe. According to another report from the Reuters, Huang stated that the EU is currently falling behind the U.S. and China in terms of AI investment.

The report from the Reuters mentioned that the EU has implemented the world’s first comprehensive regulations governing AI, which took effect in August. However, there are relatively few AI companies in the EU. Huang emphasized that the EU should accelerate its advancements in AI, noting that lots of countries are beginning to recognize data as a national resource.

Huang was in Denmark to launch a new supercomputer jointly developed by NVIDIA, the Novo Nordisk Foundation, and Denmark’s Export and Investment Fund. The supercomputer, named Gefion, is equipped with 1,528 GPUs and will be used for drug discovery, disease diagnosis, treatment, and addressing complex life science challenges, as reported by the Reuters.

Read more

• [News] NVIDIA Reportedly in Talk with Samsung for Potential GPU Orders amid Rumored Tensions with TSMC
• [News] NVIDIA’s Blackwell Overcomes Delays, as GB200 Reportedly Sets for December Mass Production

(Photo credit: NVIDIA)