Taiwan Semiconductor Manufacturing Co. on Thursday confirmed it has suspended processing new orders from key customer Huawei Technologies to comply with U.S. export regulations,
https://asia.nikkei.com/Spotlight/H...s-to-halt-chip-supplies-to-Huawei-in-2-months"We are complying fully with the new [U.S.] regulations. We did not take any new orders [from Huawei] since May 15," TSMC Chairman Mark Liu told an investors conference, confirming an earlier report by the Nikkei Asian Review. "Although the regulation just finished its public comment period, the BIS [Bureau of Industry and Security] did not make a final ruling change. Under this circumstance, we do not plan to ship wafers [to Huawei] after Sept. 14."
Under the tightened export control rule, non-U.S. chip companies must apply for licenses to use American technology and tools to supply to Huawei, the biggest Chinese tech company. TSMC and other chipmakers were not allowed to process new orders from Huawei or its chip design arm HiSilicon after May 15 without a license and must ship any orders already in the pipeline before Sept. 14.
Liu did not say whether TSMC plans to apply for licenses to ship to Huawei. A TSMC spokesperson said the chairman's comment is based on current rules, and it is not clear whether there will be any further regulation changes from the U.S. government, given that the comment period just ended a few days ago.
https://www.techpowerup.com/270903/...peng-cpu-reportedly-beats-intel-core-i9-9900kThe specific chip used in this upcoming computer is the Kunpeng 920 3211K CPU which features 24 cores clocked at 2.6 GHz paired with 8 GB DDR4 memory, 512 GB Samsung SSD, and an AMD Radeon RX 520 GPU.
In terms of expansion, this desktop does not support other types of graphics cards
The chip itself was rectangular, but it was cut from a single wafer, and contained 400,000 cores, 1.2 trillion transistors, 46225 mm2 of silicon, and was built on TSMC’s 16 nm process.
Yield
While AMD and Intel are moving to multi-chip packaging (or what Intel calls glued together chips), to make bigger silicon complexes while maintaining yield. Cerebras is going the other way with a giant chip. As a result, defects are expected on every wafer and therefore every chip. The Cerebras Wafer Scale Engine design expects these defects and has extra cores and interconnect wires to handle these defects.
Using this approach, Cerebras can be “defect tolerant.” While NVIDIA is focused on getting smaller perfect or near-perfect dies, Cerebras designed the Wafer Scale Engine to have multiple defects.
Since the Cerebras Wafer Scale Engine is so big, it cannot be cooled via air and power cannot be delivered using a traditional planar delivery method.
https://www.tsmc.com/english/newsEvents/blog_article_20200801.htmTSMC marked an amazing milestone in the past month of July – we manufactured our one-billionth good die on our 7-nanometer (7nm) technology; in layman’s terms, that would be one billion functional, defect-free 7nm chips.
A remarkable achievement for a technology that entered volume production in April 2018. Since then, we have manufactured 7nm chips for well over 100 products from dozens of customers.
https://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do?action=detail&language=E&newsid=PGWQISPGTHIn US dollars, third quarter revenue was $12.14 billion, which increased 29.2% year-over-year and increased 16.9% from the previous quarter.
Gross margin for the quarter was 53.4%, operating margin was 42.1%, and net profit margin was 38.5%.
...
7-nanometer and 16-nanometer accounted for 35% and 18% respectively. Advanced technologies, defined as 16-nanometer and more advanced technologies, accounted for 61% of total wafer revenue.
Table 3 shows capacity node for 7nm and lower nodes for TSMC and Samsung. Capacity for TSMC was about 5-6 times greater at the 7nm node in 2019 and 2020 and 3.5 times greater forecasted for 2021. At the 5nm node, TSMC’s capacity is about four times greater.
TSMC, which serves more than 460 customers, counts its leading customers among the top global fabless companies, including Apple (AAPL), Broadcom (AVG), HiSilicon, AMD (AMD), MediaTek (OTCPK:MDTKF), NVIDIA (NVDA), Qualcomm (QCOM), and Intel (INTC).
Table 4 shows these companies and their percentage of TSMC’s revenues for 2019-2021.
https://seekingalpha.com/article/44...rtake-taiwan-semi-despite-massive-capex-spendIn 2020, Samsung used 60% of its foundry output for internal use, primarily the Exynos chip for its smartphones. The remainder of capacity came from non-captive customers, Qualcomm (20%) and the remaining 20% split among Nvidia, IBM (IBM), and Intel. As Samsung increases its sub 7nm output in 2021, captive consumption will decrease, possibly to 50%.
https://www.nextplatform.com/2021/03/30/neocortex-supercomputer-to-put-cerebras-cs-1-to-the-test/Over the next year, we should get a good sense of how the Cerebras CS-1 system performs for dual HPC and AI workloads between installations at Argonne and Lawrence Livermore labs, EPCC, and the Pittsburgh Supercomputing Center (PSC).
O yeld é de 100%, segundo eles. Aliás, parece que a 1ª versão tinha redundância a mais.Isso deve ter um yield de 1 chip per waffer, a correr bem.
Isto não é um "concept". Isto é usado na prática. Eles têm vários deployments feitos e anunciados.Isto é basicamente usado como proof of concept, certo? Não é propriamente utilizado ou é?
Premium Performance: The Dimensity 1200 has an octa-core CPU designed with an ultra-core Arm Cortex-A78 clocked up to 3GHz for extreme performance, three Arm Cortex-A78 super cores and four Arm Cortex-A55 efficiency cores. With a nine-core GPU and six-core MediaTek APU 3.0, the Dimensity 1200 delivers a new level of premium performance. The Dimensity 1100 is designed with an octa-core CPU which includes four Arm Cortex-A78 cores operating at up to 2.6GHz and four Arm Cortex-A55 efficiency cores, along with a nine-core Arm Mali-G77 GPU. Both chipsets are manufactured on TSMC’s advanced 6nm process technology.