本帖最后由 kukiyi 于 2021-06-08 18:43 编辑
您需要 登录 才可以下载或查看，没有帐号？成为超大会员
28nm芯片形成了中高端集成电路制造的分界线。正是这些以及中低端芯片将满足未来对芯片的大部分需求，因为人工智能 (AI) 特性和功能已嵌入到快速增长的自主连接设备中，这些设备的范围从汽车到智能交通灯，从配套机器人到生物医学设备。
28nm 和低端芯片无法满足的功能，将在很大程度上使用 14nm 和 20nm 技术来满足。与28 nm相比，14 nm在性能方面更接近 7 nm技术，而设计和制造成本比 7 nm低得多。
例如，去年，在速度和功耗方面，英特尔的 14 nm Skylake 台式机处理器与 AMD 的 7 nm锐龙处理器没有明显区别，尽管围绕 7 nm进行了营销炒作，而且制造成本更低。
只有一小部分下游 5G 应用需要比14 nm处理器和支持芯片更强大的东西。将需要前沿的芯片组设计、基于微控制器的系统、传感器融合、先进的封装和第三代化合物半导体材料，而不是 7nm 的前沿制造，更不用说更薄的线宽晶体管芯片了。
与此同时，5G 支持的“万物互联”正在迅速形成，到 2030 年将实现空间网络。据 GlobalData 估计，到 2024 年将有超过 110 亿台企业物联网设备。中国在 5G 部署方面至少领先美国和欧盟两年，并且已经发射了 6G 试验卫星。
中国最大的芯片代工厂中芯国际 (SMIC) 是中国走向未来的前沿和中心，自去年以来，它一直在加大关键 28 nm芯片的生产。今年实现 28nm 规模对于中国国内半导体生态系统发展的长期过程，并且减少将生产外包给中国大陆以外地区的代工厂，尤其是台积电的需求具有重要意义。
明年，同样重要的是，14nm 芯片也将大幅增加。在没有匹配的本土替代品的情况下，中芯国际使用由应用材料、LAM 和东京电子，以及ASML DUV（深紫外）光刻机，并将在可预见的未来继续这样做。
然而，一台国产的 28 nm深紫外光刻机计划在今年年底前从上海微电子设备厂（SMEE）生产出来，并在上海专有的生产线上为物联网设备制造 48 nm和 28 nm芯片。
在 3 月份的上海 SEMICON上，该公司展示了一款工作在 90nm 的扫描仪。管理层最近报告说，提高 48nm 和 28nm 的良率仍然是一个挑战，但 SMEE 技术现在拥有基本的本土 UV 能力，无需美国 IP 来制造芯片。
现在评估 SMEE 可能会改变游戏规则的程度，以及到 2025 年 SMEE 和其他中国DUV 机器可能在低至 5nm 的情况下运行的规模还为时过早。但 SMEE 在一个极其困难的技术领域取得的进展已经出人意料。
ASML 仍然屈服于美国对 EUV 机器的压力，这样的光刻机可以使中芯国际能够像台积电一样向 7nm 和更先进制程方向发展。
不过，欧盟内部有迹象表明，ASML 与欧盟委员会以及英飞凌和意法半导体等领先的欧盟公司一样，正在对美国干涉其事务进行控制。毕竟，到 2030 年，预计中国将至少占全球芯片制造产能的 40%。
类似的动态正在美国专业生产设备和 EDA 设计工具供应商之间聚集力量，因为在某些情况下，中国占其收入的 50% 以上。
明年，中芯国际将增加 14nm 芯片，并在 2023 年增加 7nm 芯片。它已经进入另一种芯片的小批量生产，这些芯片介于 14nm 和 7nm 之间，采用N+1 工艺技术。
在深圳、北京和上海等政府的财政支持下，该公司正在进行价值 120 亿美元的产能扩张项目。重点将放在提高 28 nm产量上，同时也适当关注 14 nm及其 7 nm版本——后者可能会在 2024年实现量产。
通过满足中国不断增长的设计自己芯片的无晶圆厂公司不断增长的需求，中芯国际和其他六家中国代工厂很可能能够制造中国到 2025 年所需的大部分芯片——其中包括 Hi Silicon、阿里巴巴、百度、腾讯、地平线机器人、寒武纪、小米、OPPO 和字节跳动。
华盛顿面临着一个问题，即其政治目标与其国内芯片行业持续不断地进入中国市场的巨大重要性发生冲突，无论是高通、应用材料还是新思科技。高达 1000 亿美元的年收入和多达 150,000 个工作岗位受到威胁，同时也是研发资金的重要来源。
BCG 和 SIA 预测，到 2030 年，全球半导体行业的规模将是目前的两倍多，收入将达到 1.4 万亿美元，中国将占到其中的 60%，并占全球产能增长份额的40%。
China’s rapid progress in pivotal 28nm and 14nm chips
China focuses on 28nm and 14nm chips – core enablers of the internet of everything – 28nm chips form the dividing line between mid and high-end integrated circuit manufacturing. It is these and midrange and low-end chips that will cater for most of the future demand for chips as Artificial Intelligence (AI) features and functions are embedded in the rapidly growing population of more or less autonomous connected devices. These devices range from cars to smart traffic lights and companion robots to biomedical devices.
What capabilities that 28nm and lower end chips can’t satisfy, will be very largely be met using 14nm and 20nm technologies. 14nm stands a lot closer to 7nm technology in terms of performance than to 28nm, while involving much lower design and fabrication costs than 7nm.
For example, last year, in terms of speed and power draw, Intel’s 14nm Skylake desktop processors were not noticeably different from AMD’s 7nm Ryzen processors, despite the marketing hype around 7nm, while costing less to make.
Only a tiny fraction of downstream 5G applications will need anything more powerful than 14nm processors and support chips. The call will be for leading edge chipset designs, microcontroller-based systems, sensor fusion, advanced packaging and down the line third generation materials rather than the leading-edge fabrication of 7nm, let alone even thinner line width transistorised chips.
Meanwhile, the 5G enabled ‘internet of everything’ is forming rapidly to realise the spatial web by 2030. There will be over 11 billion enterprise IoT devices by 2024, according to GlobalData estimates. China is at least two years ahead of the US and the EU in terms of 5G deployment and has already put up an experimental 6G satellite.
Production of 28nm chips is being ramped up
Semiconductor Manufacturing International Corp (SMIC), China’s largest chip foundry, is front and center of China’s drive into the future, and it has been ramping up the production of critical 28nm chips since last year. Achieving scale at 28nm this year will be highly significant in the longer-term process of developing a more complete Chinese domestic semiconductor ecosystem, and in the drive to eliminate the need to outsource production to foreign foundries, most notably TSMC.
Next year, equally significant, there will be a major ramp up of 14nm chips as well. In the absence of a matching homegrown alternative, SMIC uses ASML DUV (deep ultra-violet) lithography machines supported by specialist US equipment from Applied Materials, LAM and Tokyo Electron, and will continue to do so for the foreseeable future.
Free from dependence on US
However, a home grown 28nm DUV lithography machine is scheduled to spin up from Shanghai Microelectronic Equipment by the end of this year, and to be fabricating 48nm and 28nm chips for IoT devices on a proprietary Shanghai production line.
At SEMICON 21 in Shanghai in March, the company showed a scanner operating at 90nm. The management recently reported that improving yields at 48nm and 28nm remains a challenge, but SMEE technology now has the basic home-grown UV capability free of US IP to fabricate chips
It is too early to assess just how much of a game changer SMEE might prove to be and at what scale SMEE and other Chinese DUV machines may be operating by 2025 at down to 5nm. But SMEE has already defied expectations with its progress in a fiendishly difficult area of technology.
N+1 process technology
Meanwhile, in late March, SMIC clinched an on/off/on again contract with ASML to buy over $1 billion worth of DUV machines this year. SMIC, like other Chinese foundries, had been turning to the global second-hand market in UV machines and other production equipment such as etchers, vapor deposition and wafer inspection components.
ASML is still yielding to US pressure over EUV machines that would enable SMIC to move to 7nm and thinner like TSMC.
There are though signs within the EU to suggest that ASML, like the EU commission and leading EU companies such as Infineon and STMicro, is bridling against the US meddling in its affairs. After all, China is forecast to account for at least 40% of the global expansion in chip making capacity by 2030: a not to be missed market prospect for ASML, SMEE notwithstanding.
A similar dynamic is gathering force among specialist US suppliers of production equipment and EDA design tools since China accounts for over 50% of their revenues in some cases.
Next year, SMIC will be ramping up 14nm chips and in 2023 7nm chips. It is already into small batch production of the latter chips, which stand somewhere between 14nm and 7nm, with its N+1 process technology.
It has $12bn worth of capacity expansion projects underway with financial support from provincial governments in Shenzhen, Beijing and Shanghai. The primary focus will be on boosting 28nm output with due attention also being paid to 14nm and its 7nm version– the latter likely to see volume production by 2024.
It is a high probability that SMIC, plus half a dozen other Chinese foundries, will be able between them to make the vast bulk of the chips China needs by 2025 by meeting the rising demands from China’s growing base of fabless companies designing their own chips–these include Hi Silicon, Alibaba, Baidu, Tencent, Horizon Robotics, Cambricon, Xiaomi, Oppo and ByteDance.
The chips are down in the US sanctions poker game
The Chinese leadership assumes that America will continue to weaponize its core semiconductor IP to try and hold China back. In extremis, China may need to counter the threat of tougher US sanctions over the mid-term by closing vital Chinese markets to US companies, notably Apple, and/or by cutting off supplies of rare earths, industrial grade battery materials or APIs.
As Dr Handel Jones, the highly respected Founder/CEO of International Business Strategies, puts it: ‘’the Chinese are master strategists.’’
Washington faces the jagged problem that its political objectives collide with the huge importance to its domestic chip industry of continued and growing access to the Chinese market whether it’s Qualcomm, Applied Materials or Synopsys. Annual revenues of up to $100bn and up to 150,000 jobs are at stake as well as a vital source for R&D funding.
BCG and the SIA forecast that the global semiconductor industry will be more than twice its current size by 2030, with revenues of $1.4 trillion, and that China will account for 60% of this and build out 40% of the global growth in production capacity by then
As the decade unfolds and moves towards a post Moore world shaped by big data, AI and by new chip architectures, packaging and materials, China could emerge as the leader at the post Moore inflection point.
It has done so at previous technology inflection points. For example, it has achieved leadership in 5G, in high-speed trains, in quantum communications and in big data driven AI.
It certainly has the motivation, capital and human resource, magnetism to attract foreign talent, and sheer entrepreneurial energy and ingenuity to do so. It would be unwise to bet against it.