The Key Technologies of EUV Lithography and Network Architecture Selection

The Key Technologies of EUV Lithography and Network Architecture Selection

Introduction:
In the world of technology and semiconductor manufacturing, two key areas have recently gained significant attention: EUV lithography and network architecture selection. Both these areas play a crucial role in advancing the development of next-generation semiconductor chips. This article aims to explore the common points between these topics and provide insights into their significance.

EUV Lithography:
Extreme Ultraviolet (EUV) lithography is a critical step in the manufacturing process of next-generation semiconductor chips. It utilizes EUV light generated by a high-temperature plasma produced by high-purity tin. Solid tin is melted inside a droplet generator, producing over 3 million droplets of 27µm size per minute in a vacuum chamber. A 25kW carbon dioxide (CO2) laser with an average power forms and ionizes the tin droplets in two consecutive pulses. Initially, thousands of watts of EUV light are generated, but due to absorption and scattering losses along the optical path, only a small fraction of the light reaches the photomask used in lithography. The output power and beam quality of 13.5nm light are inferred from measurements using an indirect scintillator camera. A multilayer focusing mirror system guides the light onto photosensitive polymers or photoresists, transferring the patterns onto silicon wafers. A reflective mirror is protected from tin debris by a constant flow of H2 gas. An automated wafer stage positions the wafer with a resolution of ≤0.25 nm after each exposure, undergoing 20,000 cycles per second. Overall, this process requires precise coordination among various engineering systems.

Network Architecture Selection:
When it comes to network architecture selection, reducing end-to-end communication latency between multiple machines and multiple graphics processing units (GPUs) is a crucial aspect. The key technology for reducing latency is Remote Direct Memory Access (RDMA), which allows one host to directly access the memory of another host, bypassing the operating system kernel. There are four ways to implement RDMA: InfiniBand, RoCEv1, RoCEv2, and iWARP. RoCEv1 technology is currently considered outdated, and iWARP has limited usage. The two main solutions widely used for RDMA technology are InfiniBand and RoCEv2. By bypassing the kernel protocol stack, both InfiniBand and RoCEv2 can improve latency performance by several orders of magnitude compared to traditional TCP/IP networks. In laboratory testing, the end-to-end latency at the application layer can be reduced from 50us (TCP/IP) to 5us (RoCE) or 2us (InfiniBand) when bypassing the kernel protocol stack in scenarios where the communication is within a single hop. The key components of an InfiniBand network include the Subnet Manager (SM), InfiniBand network adapters, InfiniBand switches, and InfiniBand cables. In 2021, NVIDIA introduced the Quantum-2 series switch with a speed of 400Gbps (64*400G). The switch features 32 800G OSFP ports, which need to be converted into 64 400G QSFP ports using cables. InfiniBand switches do not run any routing protocols. The forwarding table of the entire network is calculated and uniformly issued by a centralized Subnet Manager (SM). Besides the forwarding table, the SM is responsible for managing the configuration of InfiniBand subnets, partitions, and Quality of Service (QoS). InfiniBand networks require dedicated cables and optical modules for interconnecting switches and network adapters. InfiniBand's Adaptive Routing is based on per-packet dynamic routing, ensuring optimal network utilization in massively scaled networks. InfiniBand networks have been widely adopted in the industry for large GPU clusters, including Baidu Intelligent Cloud and Microsoft Azure. The major InfiniBand network solutions and equipment suppliers in the market include NVIDIA, Intel Corporation, Cisco Systems, and Hewlett Packard Enterprise. RoCE, on the other hand, offers more versatility and relatively lower pricing compared to the InfiniBand solution. It can be used not only in high-performance RDMA networks but also in traditional Ethernet networks. However, the configuration of parameters related to Headroom, PFC, and ECN on switches is relatively complex. In massively scaled scenarios like GPU clusters, the throughput performance of RoCE networks is slightly weaker compared to InfiniBand networks. Various switch vendors support RoCE, with prominent ones being Xinhua 3Com and Huawei. NVIDIA's ConnectX series of network adapters currently hold a significant market share in RoCE-compatible network cards.

Common Points and Insights:
Despite their seemingly different domains, EUV lithography and network architecture selection share some commonalities. Both fields require advanced technology and precise coordination among various components/systems. In EUV lithography, the process involves generating and manipulating high-purity tin plasma, directing EUV light onto photosensitive polymers, and protecting mirrors from debris. Similarly, in network architecture selection, RDMA technology bypasses the kernel protocol stack, enabling direct memory access and improving latency performance. Both areas require specialized equipment and components, such as InfiniBand switches and network adapters for high-performance RDMA networks. Additionally, both EUV lithography and network architecture selection play crucial roles in advancing technology and driving innovation in their respective fields.

Actionable Advice:

1. Stay updated with the latest advancements in EUV lithography: As EUV lithography is a key step in the semiconductor manufacturing process, staying informed about the latest advancements and technologies in this field can provide a competitive edge. Keep an eye on industry conferences, research papers, and news to stay up-to-date.

2. Evaluate network architecture options based on specific requirements: When selecting network architecture, carefully evaluate the specific requirements of your application. Consider factors such as end-to-end latency, scalability, cost, and compatibility. InfiniBand and RoCEv2 are popular options for high-performance RDMA networks, while RoCE offers versatility and lower pricing.

3. Collaborate with industry leaders: To leverage the benefits of advanced technologies like EUV lithography and RDMA, collaborate with industry leaders and experts. Engage in partnerships or join industry consortia to gain insights, share knowledge, and stay at the forefront of technological advancements.

Conclusion:
In conclusion, EUV lithography and network architecture selection are two significant areas in the technology and semiconductor manufacturing domains. Despite their apparent differences, these fields share common points, such as the need for advanced technology, precise coordination, and specialized components. By understanding the key technologies and considering actionable advice, businesses and professionals can navigate these areas effectively and contribute to advancing technology and innovation.