TSMC Sees HPC Driving Business

发布时间:2018-01-22 00:00
作者:
来源:Alan Patterson
阅读量:1809

Taiwan Semiconductor Manufacturing Co. (TSMC) says that high-performance computing (HPC), including demand for chips used in cryptocurrency mining, has replaced smartphones as the key driver of its business.

While low- and mid-tier smartphones still show signs of growth, sales of leading-edge handsets are likely to be “flat”, according to the world’s largest foundry, which makes all of the application processors used in Apple’s latest iPhones as well as silicon used in handsets from a range of Chinese manufacturers such as Huawei, Oppo and Xiaomi.

HPC, which under TSMC’s definition includes applications for AI, GPUs and cryptocurrency mining, is picking up just as smartphone demand is losing momentum, the company said.  While AI ASICs such as those used for cryptocurrency mining may to a degree be eroding sales of GPUs, demand for both types of chips is quite “solid”, according to TSMC.

TSMC Sees HPC Driving Business

TSMC Chairman Morris Chang

TSMC believes it will continue to lead chip industry growth in 2018 with the company’s annual revenue gaining by as much as 15 percent from the $32 billion in sales recorded in 2017. The overall growth rate for the foundry segment this year will be about 10 percent while the semiconductor industry will grow by as much as 8 percent, according to TSMC Chairman Morris Chang, speaking at an event to announce the company’s fourth-quarter 2017 results.

IDMs Go Fab Light

Contributing to TSMC’s industry-leading growth is a renewed trend toward more outsourcing by IDMs, according to Chang.

“Almost without exception, every IDM has become fab-light,” he said. “Those will become fab lighter and lighter and lighter.”

IDMs now depend almost solely on foundries for support below 90 nm, according to TSMC Co-CEO Mark Liu. Much of the new IDM outsourcing to TSMC will be for automotive semiconductors, he said.

While the company agreed with UBS analyst Bill Lu that foundry supply is becoming tight, TSMC doesn’t plan to increase its capital expenditures for 2018 significantly compared with last year. TSMC said its capex target for this year is $10.9 billion, little changed from the $10.8 billion the company spent in 2017. During the next few years, the company plans to keep capex in a $10.5 billion to $11 billion range.

The company’s latest 10 nm process, which it ramped up in 2017 to make the A11 application processor on a sole source basis for Apple’s latest iPhone X, accounted for 25 percent of TSMC’s revenue during the fourth quarter of last year. The contribution of 10 nm to TSMC’s sales during the first quarter of this year will drop due to the seasonality of the smartphone business, according to TSMC. While some 10 nm customers migrate to 7 nm, the company said it expects sales of 10 nm to grow in 2018 to meet demand for application processors, cellular baseband and ASIC CPUs.

In the meantime, TSMC has continued to enhance its legacy 28 nm/22nm and 16 nm/12 nm nodes with low-power and high-performance tweaks. The company said that in the next few years, 22 nm will play a more important role than the 28 nm process that has been the company’s cash cow for more than five years.

Leading Geometries

TSMC expects to have a 100 percent share of the 7 nm business when it starts production of the new process technology in June this year.

That expectation reflects on Samsung and Qualcomm, according to one analyst.

“This implies Qualcomm is switching back to TSMC for its first 7 nm product ramp,” said independent analyst Andrew Lu in a report for Smartkarma. Samsung may have delayed its 7 nm ramp as a result of its decision to fully implement EUV at the 7 nm node, according to Lu. Qualcomm several years ago switched from TSMC to Samsung as its foundry for Snapdragon processors.

TSMC said it expects to do 10 customer tapeouts on 7 nm during the first quarter and a total of 50 tapeouts during 2018.

Most of TSMC’s initial 7 nm production will go to customers in the smartphone business, followed by those in HPC. The company forecast that 7 nm will account for about 10 percent of its sales during 2018.

Following the introduction of 7 nm, the company will in 2019 introduce its N7+ process, an enhancement of the 7 nm process to include extreme ultraviolet (EUV) lithography at several layers. TSMC expects that N7+ will increase dies per wafer by about 10 percent compared with N7 because of EUV.

TSMC said its 5 nm process is on schedule for risk production in the first quarter of 2019 and that 5 nm customer test chips are running in its fabs.

The company said it is using EUV equipment operating at 160 watts to support TSMC’s 7 nm and 5 nm rollout. EUV scanners with source power at 250 watts have been installed in the company’s fabs, and TSMC said it is confident that EUV will be ready for the introduction of N7+ in 2019 and N5 in 2020.

The company said that even as Moore’s Law loses validity, it is positive about making 3 nm chips in the future using EUV and its chip-on-wafer-on-substrate (CoWoS) packaging technology.

“We are still pathfinding,” said TSMC Chairman Chang. “There was a question on whether it was feasible.”

(备注:文章来源于网络,信息仅供参考,不代表本网站观点,如有侵权请联系删除!)

在线留言询价

相关阅读
TSMC Plans Mini-Loop at Kaohsiung to Speed CoWoS Tool Validation; Reportedly Eyes AUO Site for CoPoS
  As AI demand drives growth in advanced packaging, TSMC is strengthening equipment and materials validation across its supply chain. According to Commercial Times, construction of Kaohsiung’s Baipu Industrial Park began on September 21, with TSMC expected to anchor the cluster. The company plans to set up a modular “Mini-Loop” line to validate new equipment and process integration before the equipment is introduced into CoWoS (Chip-on-Wafer-on-Substrate) production, aiming to improve validation efficiency by 25% to 50%.  As the report highlights, the move could signal a more direct role for TSMC in coordinating its supply chain. TSMC Vice President of Advanced Packaging Technology and Service Jun He estimates that CoWoS capacity will grow at a compound annual rate of more than 80% through 2027, with growth continuing through 2029. Unlike major front-end equipment makers, many back-end equipment suppliers are smaller and have more limited R&D resources, making it increasingly difficult for them to conduct equipment and process validation independently as AI chips evolve rapidly.  Meanwhile, ASE reportedly plans to build two facilities at Baipu after breaking ground on a new Renwu facility in April. If the investment moves forward, it would bring TSMC’s equipment and materials validation platform closer to ASE’s packaging and testing capabilities, as the report points out.  TrendForce notes that GPU suppliers and hyperscale CSPs are continuing to develop increasingly powerful and versatile AI chips, making larger package sizes a key development priority for 2.5D packaging. Although TSMC’s CoWoS-L faces competition from Intel’s EMIB-T, its advantages in technology maturity and yield are expected to keep it the mainstream packaging solution for AI chips through 2028.  TSMC Explores Glass Substrates as CoPoS Packaging Scales Up  TSMC is also reportedly exploring technologies for next-generation CoPoS as AI chips grow in size. According to another report from Commercial Times, the company is working to shift packaging substrates from round wafers to square panels while developing glass-core substrates. TSMC’s 3DIC R&D center in Japan is already developing glass-substrate technologies involving through-glass vias (TGVs), redistribution layers (RDLs), and GPU and HBM integration.  According to a July report from Economic Daily News, TSMC is currently building a CoPoS pilot line, with the technology and yields expected to take about another year to mature before production is introduced with customers.  Notably, TSMC has recently made multiple visits to AUO’s Central Taiwan Science Park facility, reportedly involving both a potential facility transaction and its CoPoS plans. The Commercial Times report notes that AUO’s experience in large-format panel processing, glass fabrication, and factory infrastructure overlaps with some of the capabilities required for panel-level packaging.  CoPoS Push Puts TGV and Inspection in Focus  For glass substrates, TGV formation is expected to become a key area of supply-chain competition. As Commercial Times highlights, the market currently favors ultrafast lasers combined with wet etching, using picosecond or femtosecond lasers to modify the glass before chemical etching forms the vias. Inspection is also expected to be a critical step in CoPoS mass production. The report notes that microcracks formed during laser drilling, copper plating, or thermal processing could spread to adjacent TGVs, affecting electrical performance and packaging yields.
2026-09-22 14:14 阅读量:298
TSMC: SoIC, CoWoS to Drive 50× Compute by 2029; Silicon Photonics to Top 50% of Transceiver Market by 2027
  TSMC is stepping up its technology roadmap to meet surging AI computing demand. According to TechNews, April Li, Global Head of AI and HPC Business Development at TSMC, said at SEMICON Taiwan 2026 that AI computing demand is growing fivefold annually, putting increasing pressure on compute capacity. In response, TSMC is integrating advanced process, 3DFabric, and silicon photonics into its HPC platform. Notably, the company expects SoIC and CoWoS integration to boost overall system computing performance to 50 times its 2024 level by 2029.  As the report indicates, Li said heterogeneous integration is the only way to push beyond current computing limits. TSMC plans N2P-on-N3P 3D integration in 2026, followed by A14-on-A14 in 2029, with interconnect pitch shrinking to 4.5 microns.  TSMC Advances COUPE and Silicon Photonics Design  TSMC is also continuing to advance its COUPE technology. According to TechNews, its 2–4-micron bonding pitch limits transmission loss at 112 Gbps to just 0.06 dB, compared with 1.38 dB for micro-bumps, helping reduce power consumption and latency to 10–20 nanoseconds.  As Mirror Daily indicates, TSMC’s self-developed COUPE platform uses TSMC-SoIC bonding technology to integrate electronic integrated circuits (EICs) with photonic integrated circuits (PICs), while supporting both grating couplers (GCs) and edge couplers (ECs). To further scale bandwidth, TSMC is pursuing two development paths. One will increase per-channel speeds from 200 Gbps to more than 400 Gbps while expanding channel counts from 16 to over 128, lifting total bandwidth from 3.2 Tbps to more than 12.8 Tbps. The other will expand wavelength-division multiplexing (WDM) from a single wavelength to 4, 8, 16, and beyond.  The company is also addressing testing and design challenges as CPO moves toward broader deployment. According to Mirror Daily, TSMC has enabled wafer-level optical testing for edge-coupled devices, allowing defective dies to be screened out before costly CPO assembly. TechNews adds that TSMC has introduced an optical PDK that enables electronic-photonic co-simulation to address key silicon photonics design challenges, while Mirror Media notes that the PDK supports Cadence and Synopsys tools to help accelerate product development.  Silicon Photonics Gains Ground as CPO Nears Mass Production  Beyond TSMC’s technology roadmap, silicon photonics is moving toward broader adoption and large-scale CPO deployment. As noted by Central News Agency, TSMC Vice President of Advanced Packaging Technology and Service K.C. Hsu said optical transceiver architectures are undergoing a structural shift, with silicon photonics expected to account for more than 50% of the optical transceiver market by 2027. CPO represents the next step, with some suppliers expected to begin mass production in the second half of 2026.  Hsu described optics as the “nervous system” of data centers as demand for high-speed connectivity grows. As noted by Central News Agency, he said optical transceiver sales rose 25% in 2025 and are expected to grow another 50% in 2026, while the high-end 100G-and-above market doubled in 2024 and expanded another 60% in 2025.  Hsu noted that Taiwan’s silicon foundry ecosystem is already capable of producing optical engines at scale and with high precision, while real-world data and market validation are helping ease longstanding concerns over CPO reliability. The bigger challenge to large-scale deployment now lies elsewhere in the supply chain, including lasers, optical fibers, fiber connectors, and product testing, he said, as reported by Central News Agency.
2026-09-01 15:44 阅读量:389
TSMC Arizona Profit Soars 663% YoY in 1H26; U.S. Earnings Pull Back 8.2% QoQ in 2Q as Depreciation Bites
  As TSMC’s Arizona fabs ramp up to meet strong demand from NVIDIA and other Big Tech customers, the U.S. unit has overtaken TSMC Nanjing as the foundry giant’s most profitable overseas subsidiary. Citing TSMC’s 1H26 interim report, Commercial Times and the Economic Daily News report that TSMC Arizona posted a record NT$36.066 billion profit in 1H26, up 662.8% YoY.  As TSMC’s overseas footprint moves into a profit-harvesting phase, its four major manufacturing subsidiaries generated a combined NT$58.529 billion in 1H26, up 215.4% YoY. TSMC Arizona alone contributed more than 60% of the total, underscoring its growing weight in TSMC’s overseas earnings mix, as highlighted by the Economic Daily News.  Arizona Fab Profit Pulls Back QoQ as Depreciation Pressure Mounts  However, the Arizona profit surge showed signs of moderating in Q2. Its profit slipped 8.2% QoQ to NT$17.259 billion, even as it remained up a hefty 307.8% YoY, according to the Economic Daily News. Commercial Times further notes that TSMC recognized about NT$14.603 billion in investment income from its U.S. fab in the second quarter, down 13.6% from NT$16.909 billion in the first quarter.  Analysts cited by Commercial Times pointed out that as construction of subsequent fabs accelerates, higher depreciation and amortization expenses could put pressure on profitability.  TSMC has already warned that the overseas fab ramp-up will weigh on margins. CFO Wendell Huang said at the July earnings call that overseas expansion is expected to dilute gross margin by 2–3 percentage points in the early stages, widening to 3–4 points as the fabs mature.  As highlighted by Commercial Times, TSMC’s first Arizona fab entered 4nm mass production in 4Q24, while the second, focused on 3nm, is slated to ramp in 2H27. With the latest $100 billion investment, total planned U.S. investment has reached $265 billion, with the Arizona site ultimately expected to house 10 fabs and two advanced packaging facilities, according to MoneyDJ. How this aggressive expansion will impact profitability remains a key watchpoint.  Nanjing, JASM Profits Ease as Overseas Earnings Face New Tests  TSMC’s Nanjing fab remained a key profit driver, posting NT$14.978 billion in 1H26 and ranking second among its four major overseas operations. However, Q2 profit edged down to NT$7.389 billion from NT$7.589 billion in Q1, according to Economic Daily.  Meanwhile, JASM’s earnings also cooled slightly in Q2. As noted by Commercial Times, the Japan subsidiary, 73%-owned by TSMC, turned profitable this year after posting a loss in 2025, earning NT$951 million in Q1 and NT$727 million in Q2.  The bigger question now is whether JASM can keep that momentum after the July 7.1-magnitude Kumamoto earthquake. Commercial Times notes. The fab has resumed normal production and passed structural safety checks, but equipment requalification, calibration and wafer losses could reportedly weigh on Q3 results.
2026-08-18 11:25 阅读量:395
TSMC and Researchers Overcome 2D Semiconductor Bottleneck with Epitaxial Interface Engineering
  As conventional silicon semiconductor technology approaches its physical limits, 2D semiconductors are increasingly viewed as a key enabler for extending Moore’s Law. According to TechNews, citing Taiwan’s National Science and Technology Council (NSTC), a research team led by Dr. Iuliana Radu of TSMC, in collaboration with National Yang Ming Chiao Tung University (NYCU), has developed a high-performance monolayer molybdenum disulfide (MoS₂) top-gate transistor by addressing the long-standing interface challenges facing 2D semiconductors. The study was published in Nature Electronics.  The report explains that as silicon transistors approach their physical scaling limits, shrinking channel dimensions lead to higher leakage current and weaker electrostatic control. This has driven researchers to explore alternatives, with monolayer MoS₂ emerging as one of the most promising 2D semiconductor candidates. Despite its atomically thin structure, it retains semiconductor properties, making it well suited for smaller, more power-efficient transistors. However, 2D semiconductors require an ultrathin gate dielectric to control current flow, and depositing this layer often introduces dielectric-related scattering that degrades electron transport and device performance.  To overcome this challenge, the researchers focused not on discovering a new material, but on redesigning the interface through epitaxial interface engineering. Using ultra-high vacuum technology, the team first deposited an ultrathin aluminum layer onto monolayer MoS₂, then oxidized it to form an approximately 0.42nm-thick aluminum oxide (Al₂O₃) layer as a high-quality template for subsequent material growth. According to the report, this atomically thin interface improves the quality of the interface between the 2D semiconductor and the gate dielectric while reducing electron scattering, enabling both an ultrathin structure and high transconductance.  The resulting high-performance monolayer MoS₂ top-gate transistor delivers ultralow leakage current and excellent operational stability even at extremely small dimensions, the report says. By reconciling two long-standing tradeoffs—an ultrathin dielectric layer and high carrier mobility—the work demonstrates the practical potential of 2D semiconductor devices. According to the report, the technology could enable faster, lower-power electronic devices while integrating seamlessly with existing semiconductor manufacturing processes.  Leading Chipmakers Pursue 2D Transistor Technologies  TSMC has been advancing 2D-material transistor development. In June, TSMC, imec, and ASML unveiled a scalable 300mm integration route for 2D-material-based n-type field-effect transistors (nFETs) and p-type field-effect transistors (pFETs). According to an imec press release, the collaboration demonstrated, for the first time, 50nm contacted poly pitch (CPP) transistors using MoS₂, WS₂, and WSe₂ channel materials, establishing a manufacturing-compatible integration approach for next-generation 2D transistors.  Intel has also been advancing 2D transistor research. According to Tom’s Hardware, Intel Foundry and imec demonstrated a 300mm-ready integration of key process modules for 2D field-effect transistors (2DFETs) in late 2025. The collaboration featured a fab-compatible contact and gate-stack integration scheme, marking another step toward future manufacturing of 2D-material-based transistors. While commercialization remains a long-term goal, the work helps de-risk the development and eventual production of chips based on 2D materials.
2026-08-10 15:15 阅读量:441
  • 一周热料
  • 紧缺物料秒杀
型号 品牌 询价
CDZVT2R20B ROHM Semiconductor
TL431ACLPR Texas Instruments
RB751G-40T2R ROHM Semiconductor
BD71847AMWV-E2 ROHM Semiconductor
MC33074DR2G onsemi
型号 品牌 抢购
BU33JA2MNVX-CTL ROHM Semiconductor
BP3621 ROHM Semiconductor
IPZ40N04S5L4R8ATMA1 Infineon Technologies
ESR03EZPJ151 ROHM Semiconductor
STM32F429IGT6 STMicroelectronics
TPS63050YFFR Texas Instruments
热门标签
ROHM
Aavid
Averlogic
开发板
SUSUMU
NXP
PCB
传感器
半导体
相关百科
关于我们
AMEYA360微信服务号 AMEYA360微信服务号
AMEYA360商城(www.ameya360.com)上线于2011年,现 有超过3500家优质供应商,收录600万种产品型号数据,100 多万种元器件库存可供选购,产品覆盖MCU+存储器+电源芯 片+IGBT+MOS管+运放+射频蓝牙+传感器+电阻电容电感+ 连接器等多个领域,平台主营业务涵盖电子元器件现货销售、 BOM配单及提供产品配套资料等,为广大客户提供一站式购 销服务。

请输入下方图片中的验证码:

验证码