Internet traffic has been growing at an exponential rate, and the amount of data generated by Internet traffic is very large. Data centers are critical to supporting data communication, storage and processing through cloud computing. Optoelectronic devices play a vital role in modern data centers, which poses a major challenge for data center mass production (HVM).
Trend:
As shown in Figure 1, a recent Cisco study shows that global IP and global data center traffic are growing at a compound annual growth rate of around 25%. The proliferation of emerging applications such as the Internet, video streaming, 3D sensing, smart cars and VR/AR has generated huge bandwidth and new data center needs. By 2020, projections show that 50% of data centers will have a very large scale, and about 40 super-large scale data centers will be established each year for the next three years.
Figure 1 Market promotes the development of optoelectronic industry
The development of the optoelectronic industry plays a crucial role in bandwidth growth. In the past, telecom demand has driven bandwidth growth, and now more is the data center business to promote extraordinary growth in bandwidth. Figure 2 shows the global sales revenue of optical modules for telecommunications and data communications, respectively. The key turning point occurred in 2017-2019, and the sales revenue of optical modules in the digital communication field exceeded the revenue of optical modules in the telecommunications field. In addition, the study predicts that the future growth rate of data communications will also increase significantly.
Figure 2 The global sales revenue of data center-driven digital pass-through modules exceeds the global sales revenue of telecom optical modules.
As data center demand becomes the new growth engine, the optoelectronics industry is experiencing the challenges of a new data center business model that is distinct from traditional telecom models. The following diagram summarizes the main differences between telecom-centric and data center-centric business models. The most critical factors affecting optoelectronic manufacturing are high volume, low predictive visibility, rapid innovation, fast product response and low cost.
Table 1 Comparison of two terminal customer business models in the optoelectronic industry
Due to the extensive deployment of telecommunications, the optoelectronic industry has been greatly affected by telecommunications. Due to the harsh operating environment (eg, -40 ° C to 85 ° C, RH up to 85%) and the high cost of system deployment and maintenance, each telecom upgrade cycle is long and requires careful planning. The annual forecast visibility from operators to equipment vendors and upstream device vendors is quite good. In this long deployment and use cycle, it is easier to accept sophisticated and expensive optical components. Device vendors have more time to plan and respond to customer needs, and unlike the short-term we see in the data center business, we need fast product updates and high-volume rapid response.
Data center services run most of their networks in a controlled environment, such as buildings with temperature and humidity control (eg 0°C to 55°C, RH 40% to 60%). Most of the capital expenditure (Capex) is used for land, buildings and wiring. Most operating expenses (Opex) are used for energy consumption and environmental control. While optoelectronic device spending is relatively small in terms of capital expenditures and operating expenses, they play an important role in cloud service providers’ revenue-generating revenues from customers. In a controlled environment like an office, the upgrade cost is lower. The lower update cost and the low cost of optoelectronic devices in total capital expenditures and operating expenses combine to enable data center providers to reduce the upgrade cycle for existing data centers to three to five years. For each upgrade, they want to maximize the bandwidth and state of the art optoelectronic technology to maximize revenue in the same building.
It is difficult for data center providers to accurately predict the start-up time of a particular data center project, as new data center construction projects require local government support and approval for site selection and land purchases, which greatly reduces the visibility of optoelectronic device suppliers to the market. prediction. On the other hand, once the project is approved, data center providers naturally hope to build up as quickly as possible to quickly increase revenue sources. After two to three years of system construction, the new data center will be ready to install state-of-the-art optoelectronic equipment so that it can create maximum revenue potential on the first day of operation.
Finally, the rapid update of products for data center applications presents a dual challenge for high-volume and high-hybrid integrated optoelectronics manufacturing. As shown in Figure 3, in the past few years, optical modules have been upgraded from 10G to 40G. Now, 100G is surpassing 40G as the leading technology for new batch deployments. At the same time, the 200G / 400G solution is in the sample design and small batch production phase. All of these different technologies and products will coexist over a period of time on the same optoelectronic device manufacturing platform.
Figure 3: Rapid innovation in optoelectronic manufacturing drives high-volume, high-hybrid integration of different rate product coexistence tables
Industry opportunities and solutions
The challenges posed by all of these new cloud-based data center business models are accompanied by opportunities. How does the optoelectronics industry address these challenges? The key is automation technology.
As shown in Figure 4, in the case of low predictive visibility in the market, optoelectronic device suppliers do need to adopt a high-volume, high-quality, low-cost manufacturing and data center timely mode to achieve rapid start-up production and shutdown capabilities. Flexible capacity. Automation technology can increase production quickly without worrying about the large amount of training required for high-volume manual production. Automation technology can also reduce reliance on labor, thereby reducing variable costs, especially during off-season demand. At the same time, device vendors can use the same automation platform and production process during the development and manufacturing phases, so automation technology minimizes the risk of new product introductions (NPIs). In addition, high-precision automation technology can solve complex process fabrications that cannot be achieved by hand and produce leading products.
Figure 4 shows the challenges of the data center business and the possible reactions in the field of optoelectronic manufacturing.
At present, the biggest challenge in the field of optoelectronic manufacturing is how to deal with the production of large quantities and highly mixed products. High product mixes often originate from NPIs or from conversions between different products. This conversion makes high-volume production particularly difficult because companies cannot increase production while keeping configurations changing. To solve this problem requires flexible, high-speed automation technology, how to make the automation equipment flexible and high speed has always been a problem. The next case study will show that by innovating in the automation industry, you can achieve high speed while maintaining flexibility.
Case Study – Flexible High Speed Patch Automation Platform
To clarify our understanding of automation solutions, the following is a case study of one of the key manufacturing processes (ie, patches). Figure 5 is a general step view of various components fabricated by optoelectronics. A variety of such components come from different types/generations to support a wide range of data center communication networks. There are several main process steps related to different finished products:
- For tube-box products (such as modules, high-power lasers), it is usually first necessary to complete the substrate/substrate chip (CoC/CoS) bonding, and then bond the CoC/CoS to the common substrate for the lens/mirror attachment. On, then package. The latest trend is the increasing number of chips or wafers (such as lasers, capacitors, and thermistors) that need to be bonded to a common substrate by eutectic or epoxy bonding.
- For PCB layer products (such as active optical cable (AOC) and on-board optical device (OBO)), the chip is directly attached to the PCB. A variety of chips, such as vertical cavity surface emitting laser (VCSEL) arrays, photodiode (PD) arrays, laser drivers, and transimpedance amplifiers (TIAs), need to be attached to the PCB. Usually the final lens attachment is done by active calibration, but now this process is increasingly done by passive wafer bonding by a high speed, high precision (3 μm) placement device.
- For TO-can products, the laser chip and other wafers are attached to the TO substrate or column after being inverted at 90°. To support the upcoming 5G wireless deployment, wavelength division multiplexing (WDM) lasers and electro-absorption modulated lasers (EMLs) are available in low-cost TO-can packages. These WDM/EML-TOs are much more complex than TO-can packages that only require 1–2 wafers when there are many chips that need to be bonded.
- For silicon photonics, there are more and more processes from CoC formation to interposer chip and chip to wafer (CoW) transition. There are still many processes that have not yet been determined, but the future will increasingly depend on wafer layer integration.
Figure 5 shows an example of how to automate the placement tools and processes for optimal efficiency.
Figure 6 is an example of how to implement a highly efficient automatic placement device and process. There are three columns in the chart – on the left is the “integrated” type of chip adhesive for R&D and low to medium volume. In the middle is the process consistent with the example in Figure 5. On the right is an example of a “special” type of flexible high-speed placement machine for HVM. Through the three key points we have listed below, you can understand how this approach can help the optoelectronics industry meet the optoelectronic manufacturing challenges discussed earlier:
- The common platform of the hardware and software layers can greatly reduce the risk of NPI by eliminating control or process changes that have been set up during the development phase. In the R&D/NPI phase, integrated patch devices can be used to develop multiple types of products. Automated machines can handle all types of patches and production processes in a single device. It is also fast enough to produce small/medium level mass production for the design and manufacturing process. Once demand arises and requires mass production, the production line can be organized into specific process combinations.
- High-volume, high-hybrid optoelectronic manufacturing requires flexible, high-capacity machines that can cover multiple wafers, processes, and products through one machine for one or more process groups. Optimize speed and throughput by introducing new overspeed eutectic modules and deploying multi-stage parallel processing (such as separating material processing from chip bonding steps) to instantly integrate replacement tools with zero time wastage.
HVM devices designed for next-generation products require fast, parallel innovation. For example, 3μm patch accuracy is improved from current mainstream optoelectronic processes using patch accuracy greater than 5μm. In addition to implementing new products, the high precision of the patch can increase the process benefits of current product manufacturing by reducing assembly errors as early as possible in the process.
Figure 6 shows an example of a platform approach for high volume and high hybrid optical manufacturing.
In summary, a high-speed automation platform that can produce 24/7 continuous production without sacrificing flexibility, precision and reliability is critical for high-volume, multi-variety optoelectronics manufacturing. A common platform between R&D and mass production, as well as timeless accuracy, is equally critical to rapid innovation and reduced NPI risk, as well as yield improvement and cost reduction in volume production.
