High-speed connectivity is critical to a wide variety of end-markets that rely on fast, reliable and power-efficient transfer of data.
Alphawave Semi’s wired connectivity IP addresses six core infrastructure-oriented end-markets:
Hear from our President andCEO about how our designs address the rapidly growing need for high-speed connectivity in data centers.
In recent years, it has become essential for data center operators to build data centers all around the globe to address the growing volumes of data and the increased demand for enterprise-grade and consumer cloud services.
As hyperscale data centers grow in number and size, reliable high-speed wired connectivity becomes critical as a large number of servers, switches and storage devices need to be interconnected in a cost-efficient manner.
In addition, up to 76 per cent. of all data center internet traffic traverses within data centres due to interaction between compute resources, putting significant pressure on networking and interface bandwidth requirements.
Increasingly in recent years, hyperscale data center operators have been designing their own silicon for processing and connectivity functionality to take advantage of economies of scale, attain greater internal control over technology, and increase service differentiation.
To ensure their silicon matches or exceeds performance of merchant silicon, they have been relying on third-party silicon IP providers, like Alphawave Semi.
In a typical data center, data is transmitted between compute and storage devices through switches, routers and interface cards using copper or optical cables.
Wired connectivity IP from Alphawave Semi can be found in switch front- and backplanes, providing critical high-speed interfaces between switch ASICs and optical devices as optical links are transformed to electrical interfaces, and in high-speed interface cards that sit in servers, storage devices, and other appliances.
As AI models become more complex and multi-layered, they consume an increasing amount of compute, storage, and networking resources. The amount of compute used in largest AI training runs is estimated to have increased 300,000 times from 2012 through 2018, which caused a general increase in AI chip complexity and size.
As it becomes less practical and economical to build large monolithic die for AI, due to complexity and yield challenges, many AI suppliers choose to cluster a number of AI chips that have to share common resources such as memories and be able to talk to each other for greater performance. Therefore, interface connectivity can be a key bottleneck for AI chips and may prevent AI systems from reaching their full performance potential. Alphawave Semi’s silicon IP solutions solves this connectivity challenge.
As data processing requirements in data centers continue to grow, so does the proliferation of high-performance storage, particularly solid-state drives that are based on flash memory. Compared to hard-disk drives, flash-based solid-state drives are able to store more data per footprint and are much faster to access.
As requirements for fast data access and storage continue to grow, solid-state drive manufacturers are increasingly using the NVM Express (that is, NVMe) interface specification for external connectivity, which uses PCIe as the connectivity standard.
Historically, PCIe speeds have doubled every three years and are expected to increase to 128 Gbps. Such interface speeds require high-performing and reliable SerDes IP, like those provided by Alphawave Semi, in order to avoid performance bottlenecks related to external connectivity.
5G aims to increase cellular transmission speeds for wireless users up to 10 gigabit per second and expand bandwidth and density to accommodate a wide array of Internet of Things (IoT) devices.
An important part of the evolving wireless infrastructure architecture is the utilization of shorter frequencies, which necessitates a build-out of a greater number of antenna units interconnected through “front-haul” links with an array of distributed basebands. Because of this architectural separation, a significant strain is put on the wireline network, particularly on the front-haul and mid-haul portion that requires very low-latency reliable connectivity.
Connectivity interface has to keep up with evolution of high-performing specialized wireless integrated circuits such as transceiver and baseband units and support a variety of standards, including CPRI, JESD204, Ethernet and PCIe.
Advanced driver assistance systems, in-vehicle infotainment and autonomous driving systems are driving the need for interconnected cameras, sensors such as LiDARs and RADARs, displays and on-board processors.
It is expected that each autonomous car will be able to generate and consume up to 4 terabytes of data per hour of driving. High-speed wired connectivity is key to move this data from sensors to processing nodes and interconnect the automotive components with low latency. Due to stringent safety requirements, automotive connectivity has to be reliable and resilient to noise and interference under harsh environments.
In 2020, MIPI, a mobility-oriented global business alliance, has released the first automotive long-reach SerDes interface specification, which calls for data rates as high as 16 gigabit per second with a roadmap to 48 gigabit per second and beyond.