Starting from various aspects such as national strategy, technological development, and energy policies, China has launched a crucial national project - the "East West Computing" project in the context of "new infrastructure". For the first time, it has elevated computing power resources to the level of basic resources such as water, electricity, and gas, and coordinated the layout and construction of national integrated computing power network hubs.

At present, the scale of China's data centers has reached 5 million standard racks, with a computing power of 130EFLOPS (1.3 trillion floating-point operations per second). ICT research shows that the annual new investment scale of China's data centers is around 18-200 billion yuan. It can be foreseen that with the vision of "computing power infrastructure from the east to the west", the construction of servers and data centers in China will take on a new perspective. Power semiconductors provide the "blood" and "spinal cord" for power supply, and this field has given birth to many advanced players, such as Great Wall in mainland China, Eurolink, as well as companies such as Delta, Guangbao, and QunGuang in Taiwan. Opening the "Qunfang Spectrum" of the active and extremely important participants in this track, we found that there is also an important name - Shenzhen Shangyangtong Technology Co., Ltd. (hereinafter referred to as "Shangyangtong").

We interviewed Wang Bin, Vice President of the High Voltage Business Unit of Shangyangtong Company, on issues such as technology exploration, market layout, ecological construction, and one-stop solutions in the important segmented application scenarios of servers and data centers.

The uniqueness of servers and data centers and Shang Yangtong's "arsenal"

As mentioned earlier, driven by the top-level design of "East to West Computing", the domestic data center and server market is in a period of rapid growth, with terminal internet companies and large enterprises and institutions having a large demand for self built data centers. Considering that there will be a large demand for power devices in the data center and server power supply, it is bound to drive an explosive increase in the demand for power semiconductors in this application scenario; At the same time, the data center is a major energy consumer, requiring increasingly high efficiency of server power supply. It also has a deeper standard setting system for the performance of semiconductor devices. Secondly, there are clear regulations for reliability and long-term working life, and higher essential requirements are also put forward for semiconductor devices.

At present, among the main server power customers in the mainland, they have delivered in batches, and the terminal customers include star level Internet companies such as ByteDance and Baidu.

Wang Bin conducted a comprehensive analysis of the unique "landing" demand for power semiconductors in the server and data center markets. Undoubtedly, the performance matching of power semiconductors between large servers and data centers is definitely higher than that of consumer products such as mobile or desktop servers, This is mainly reflected in two aspects: Firstly, large data storage systems cannot withstand sudden power outages and require continuous power supply. Secondly, the CPU energy supply of servers and data centers often faces a situation of never running to being fully loaded in a very short period of time. The time difference from no load to full load is extremely short, in milliseconds for a long time, or even a few microseconds for a short time, which requires a very high transient power cycle

Generally speaking, the power semiconductor performance of large servers and data centers belongs to the industrial level, and its requirements for stability and reliability, including environmental tolerance, lie between the lower consumption level and the higher vehicle specification level. However, Wang Bin specifically pointed out that the performance requirements of power devices must be closely judged in conjunction with application scenarios, And cannot abstract conclusions: According to general understanding, the reliability of power semiconductors in data centers is lower than that of car grade applications. However, the backend of the power supply in data center servers is directly the CPU computing unit, and its reliability requirements are still much higher than other industrial grade applications. The short process from no-load to full load has a very high impact on the cold and hot cycles of power devices, and the performance demand in this area is even higher than that of car applications

By identifying the "pain points" of the application scenario, precise layout "prescriptions" can be formulated for the required technical knowledge in the application scenario. Compared to communication power supplies, automotive applications, and smart homes, Shangyangtong has gained opportunities to showcase its expertise in the research and production of high-end power semiconductor devices when entering the data center and server markets.

In this unique 'battlefield', we have always adopted leading super junction MOS technology, SGT MOS technology; Its products are more suitable for efficient, high power density, and high reliability power applications, almost perfectly meeting the reliability and energy efficiency requirements of data center server power supply.

Data centers often require a large amount of electricity, and have high requirements for power efficiency and reliability. Focusing on this difficulty, the design and development of high-voltage SJMOS have deeply considered the actual application needs of customers, using the most advanced technology and processes in China to provide customers with the latest generation of products. Moreover, Shangyang Tong's high-voltage SJMOS has lower conduction resistance and more optimized junction capacitance curve.

Throughout the various segmented markets of power semiconductors, silicon carbide has been relatively early applied in server power supplies. Thanks to the extremely small reverse recovery loss of silicon carbide (compared to ordinary FRD products), SiC SBD replaced silicon diodes in PFC earlier, significantly improving the efficiency of PFC. Generally, SiC diodes greatly reduce reverse recovery loss and can improve power efficiency by 1-2%. In this regard, the designed SiC SBD inherits the advantage of dynamic no reverse recovery, while reducing the high-temperature VF characteristics, which can further improve power efficiency; At the same time, it has better surge capacity to meet the safety requirements of customers during lightning transients.

In addition, power management in large data centers requires high redundancy or high capacity characteristics, and Oring MOS also has strong market competitiveness in solving this problem. Regarding this, Wang Bin explained: "Oring MOS is a very critical device at the output end of the server power supply - it is the last safety management device in the entire system power supply. The Shangyangtong 30V product has been applied to the server power supply and received high praise from customers. The product is very practical and durable. The designed product has high reliability and high impact resistance considerations, allowing customers to receive good protection in abnormal usage environments

In summary, a complete set of power device solutions such as high-voltage super junction MOS, medium and low voltage SGTMOS, SiC SBD, etc. can be provided in the server power supply, and the product design is aimed at high-end power supply customers.

There are many killer products in the overall planning of the one-stop solution layout, which can reflect product performance and also consider the level of craftsmanship in the subsequent packaging. At present, the products we use in servers and data centers still have the largest amount of packaging in terms of TO packaging, represented by the traditional plugin packaging of TO247 and TO220. We are also developing towards some surface mount packaging directions, and we also have DFN 8 × 8 and 5 × This form can reduce the parasitic inductance of our device products, which is beneficial for improving customer power density and efficiency. Surface mount packaging is suitable for higher end and higher power density power management devices. This form of packaging is relatively superior in size, easier to produce, and can be mechanically produced on a large scale. In contrast, TO packaging requires manual participation. Wang Bin elaborated on this in detail on Jiweiwang.

In the face of changes in the power supply system of future large data centers, Shangyangtong maintains strategic focus

Power efficiency and reliability may be the most important issues in the data center industry. Wang Bin pointed out, "From the power grid to traditional AC-UPS, then to server power and back-end on-board power supply, and ultimately to a power supply chain like CPU, the overall efficiency is estimated to be 50% -70% of the energy efficiency can truly reach the CPU end. In other words, about 30% -50% of the energy consumption is lost in the intermediate circuit, which is something the industry has been hoping that data centers can optimize and solve

At present, the most widely used uninterrupted power supply in domestic data centers is still traditional UPS, which mainly consists of three parts of circuits: rectifier AC-DC, inverter DC-AC, and static bypass. Faced with the problems of low efficiency, poor reliability, flexibility, and scalability of traditional UPS, many large domestic and foreign internet companies are continuously exploring distributed uninterrupted power supply systems. So, what impact will the future changes in the development layout of data center power supply have on the performance requirements of power semiconductor devices in servers and data centers?

Regarding this, Wang Bin's analysis: There may be some impact, but this change will not have a very fundamental difference in the demand for power management devices in data centers. Many large internet companies couple the AC end between uninterruptible power supplies and servers, such as Alibaba's highly integrated high-voltage distribution cabinets, transformers, HVDC, and other equipment, building a DCUPS system they call the 'Panama' system, simplifying complexity and optimizing system architecture and footprint Overall, there is no essential difference in the demand for power devices; On the other hand, many data center customers are also optimizing their heat dissipation methods, adopting liquid cooling, and so on. Although the architecture of the power supply system is changing, there will be no significant differences in the performance and usage of power semiconductors.

Conclusion: The close fit between R&D proportion and national top-level design

If computing power is becoming a "barometer" for measuring China's economic situation in the new development pattern, then the supply of electricity is the blood and bone behind computing power. Correspondingly, efficient, low-power, and intelligent power management solutions accurately indicate the health of blood and bone.

The energy management and data security of servers and data centers involve many top-level designs at the national level, such as "dual carbon" and big data security. Therefore, Shangyangtong's layout in this segmented market is cautious and stable, step by step, and has been successful for a long time. Wang Bin revealed, "Currently, our company's layout in the server and data center fields accounts for a high proportion of the company's research and development as well as revenue. We have put a lot of effort into such an industry closely related to data information, which is also a manifestation of actively responding to the country's' dual carbon 'call

With the trend of computing infrastructure becoming increasingly a public service, it can be foreseen that the research and development and layout of power semiconductors in data centers are infinitely vast and promising in the future.