Cloud computing technology continues to evolve, and IP becomes the key "protagonist"

Cloud computing is seen as the next revolution in the technology world, which will bring fundamental changes to the way of working and business models.

Since the introduction of cloud computing technology, applications and computing resources have been continuously migrating from enterprise data centers to cloud environments. According to Gartner's forecast, by 2025, 80% of enterprises will shut down traditional data centers, and their physical infrastructure will be completely dependent on cloud providers (Figure 1).

Figure 1 – Transition to cloud infrastructure (Source: Gartner)

Three major trends and challenges driving the development of cloud computing

In this process, there are three main trends driving the development of the cloud computing market: the rapid growth of cloud data, the ubiquitous use of artificial intelligence to extract the meaning of data, and the expansion of cloud services to the edge of the network. Correspondingly, each development trend will pose challenges to SoC designers.

First, cloud data is growing exponentially. According to IDC, the amount of cloud data will triple from 2020 to 2025. (Figure 2)

Figure 2 - Global Internet data growth from 2010 to 2025 (Source: Data Age 2025)

The growth of data will bring new challenges to SoC developers, because they need to quickly and efficiently move large amounts of data between servers and within servers (i.e., between devices within the server). As a result, they will face challenges such as improving performance per watt, increasing computing power within existing capacity/heat limits, and providing high-speed, consistent chip interfaces for data access between devices.

Secondly, the amount of data collected, stored, and processed in cloud environments is increasing rapidly. Due to the large amount of data, traditional mechanisms for analyzing and extracting data meaning can no longer meet the needs. Therefore, companies are using artificial intelligence to analyze data.

To effectively extract value from the large amounts of data collected and stored in the cloud, multiple enhancements are needed to SoC and cloud infrastructure, including: using high-capability, high-performance AI accelerators for machine-based data analysis; using high-bandwidth memory/data interfaces to provide data to AI accelerators; and enabling AI accelerator expansion and data transmission through high-speed, low-latency accelerator interconnects.

Furthermore, there is a surge in the number of IoT devices connected to the cloud. The reason for the increase in these devices is that there are many latency-sensitive applications. In order to meet the minimum latency requirements of these applications, designers deploy relevant data and data processing capabilities closer to the point of use in distributed networks at the edge.

As cloud environments continue to expand to meet edge computing needs, SoC developers must meet new requirements for edge computing such as supporting increasing machine-to-machine communications and sub-millisecond application latency.

Specialized enhancements for each cloud data center market

The challenges mentioned above can be addressed by enhancing the six main functional areas of cloud computing (Figure 3), including computing servers, network infrastructure equipment, storage systems, visual computing solutions, telecommunications edge infrastructure equipment, and artificial intelligence/machine learning (AI/ML) accelerators.

Figure 3 - Six major functional areas of cloud computing

The first is the server. The growth of data volume has put forward the demand for faster server interfaces and more efficient memory. In addition, protecting data is also critical for cloud computing. To implement the necessary security algorithms in these high-speed interfaces, high-quality cryptographic IP for data encryption and decryption, security protocol accelerator IP for implementing high-speed security protocols, and trusted execution environments for providing root of trust and security key management are required. In order to avoid bottlenecks in various data paths, the IP used to implement these functions must be able to maintain line speed operation.

The second is network infrastructure equipment. Data growth requires faster network speeds. To this end, leading Ethernet switch manufacturers are already developing 800Gbps switches based on 112G SerDes. As data volume continues to grow, 1.6Tbps Ethernet may be launched in the next few years. Infrastructure switches that support 400Gbps Ethernet ports can use 56G x 8 or 112G x 4 SerDes electrical interfaces.

The third is storage. Storage systems manage the growing amount of data and use accelerators to process data. These developments include the use of computational storage, storage-class memory, cache coherent interfaces connected to persistent memory, and next-generation NVMe interfaces for higher data transfer speeds.

Fourth, visual computing solutions. As cloud applications continue to evolve, more visual content has emerged, and support for visual computing has become an additional feature of cloud infrastructure.

The fifth is telecommunications edge infrastructure equipment. The integration of cloud and edge will bring cloud services closer to end users, providing a richer, higher-performance, and lower-latency experience. To achieve fast response for control systems and other latency-sensitive applications, the best way may be to bring data collection, storage, and processing infrastructure closer to the point of use, that is, the edge of the network.

Finally, artificial intelligence for data analytics has become an important capability in cloud data centers. Because AI accelerators tend to process large amounts of data, the memory interface is often the bottleneck, making high-bandwidth memory valuable for these devices (Figure 4).

Figure 4 – Block diagram of a representative cloud-hosted AI accelerator

Additionally, AI accelerators for edge computing, especially aggregator and gateway applications, typically require the use of low-power DDR (LPDDR) memory. To support the scaling of AI solutions, accelerators must include a high-speed interface, such as 56Gbps or 112Gbps SerDes or HBI. High-speed interfaces between chips provide accelerator scaling and expansion capabilities to meet the needs of demanding AI applications.

IP Solutions for Cloud Infrastructure

All these technological changes and requirements will bring new challenges to SoC developers, which makes their demand for IP even stronger. Based on this, Synopsys provides a comprehensive portfolio of high-quality and silicon-proven IP products, enabling designers to develop SoCs that support current and future cloud computing applications. Synopsys' DesignWare® interface IP, processor IP, security IP, and foundation IP are optimized for high performance, low latency, and low power consumption, while supporting advanced process technologies from 16nm to 5nm FinFET.

The DesignWare® IP portfolio covers the vast majority of IP in many cloud SoC chips. In addition to IP solutions, DesignWare IP has been deployed in thousands of designs, helping designers shorten the time to production for global cloud computing applications.

Companies that have chosen DesignWare IP as key building blocks for SoCs in cloud environments include Habana Labs, an Intel company, which achieved first-time tapeout success for its Goya inference processor SoC using Synopsys' DesignWare controller and PHY IP for PCI Express 4.0. "After a thorough evaluation, we selected Synopsys' leading 16 GT/s DesignWare IP for PCI Express 4.0 because of its industry-leading reputation and advanced features required for the most demanding, data-intensive SoCs," said Eitan Medina, chief business officer at Habana Labs.

Starblaze Technology also used DesignWare IP to achieve the first successful tape-out of the STAR1000 SSD controller and put it into mass production. Using DesignWare ARC processors, foundation IP, security IP, and interface IP, Starblaze was able to reduce power consumption and I/O latency by 50%, reduce SoC area by 7%, and achieve the highest security through the best combination of power consumption, performance, and area.

AMD delivered its Ryzen and EPYC processors with Synopsys' DesignWare IP portfolio. "By choosing DesignWare IP with custom capabilities, we were able to not only meet our schedule but also keep our roadmap on track," said Rolands Ezers, senior director of I/O and Circuit Technology at AMD. DesignWare IP enabled AMD to beat out competitors with its EPYC server and Ryzen PRO desktop processors.

Looking ahead to the next generation of protocols, companies such as Achronix and Astera Labs have announced that they have adopted DesignWare PCIe 5.0 for high-performance cloud designs.

Conclusion: The key changes triggered by the cloud computing technology speech include the rapid growth of data volume, the expansion of cloud services to the edge of the network, and the widespread deployment of artificial intelligence to process massive data and gain insights from it. These changes will bring huge opportunities and challenges to SoC developers.

It has been proven that with high-quality, silicon-proven IP building blocks, designers can develop SoCs for high-end cloud computing solutions, strengthening applications including servers, networking, storage, visual computing, edge computing and artificial intelligence accelerators.