kantver - Fotolia


How will silicon photonics technology affect data center connectivity?

Silicon photonics technology has been around for over a decade, but its integration into the data center is still in its infancy. Will the connectivity benefits be worth the wait?

Silicon photonics technology has been on the horizon since Intel first announced its initiative at the dawn of the 21st century. The goal was to develop components that finally integrated optical and electronic signaling on the same silicon device, allowing high-bandwidth data communication over distances that copper conductors simply could not reach.

Let's take a closer look at the basic ideas behind silicon photonics and some of the benefits and challenges they can present for data center teams.

Q. What is silicon photonics technology and how does it affect the data center?

Silicon devices are the foundation of every processor, chipset, ASIC and memory component. At the same time, photonic devices -- components built to carry photons of light, rather than electrons, over great distances -- have been a mainstay of modern networks, providing high-bandwidth connectivity between buildings, across continents and around the world.

Silicon photonics uses existing complementary metal oxide semiconductor chip fabrication technologies to build transistors and optical components on the same silicon die at the same time, rather than separate electrons and photons on different devices. Silicon photonics fabricate the microscopic components needed to make direct conversions from electrical signals to light signals and back again, as well as the supporting optical components, to guide and separate optical wavelengths. Signals between silicon photonics chips are connected using optical fibers.

So how does silicon photonics technology affect the data center? The initial crop of components in the market seems similar to conventional optical network adapters, which use fiber cables to connect devices across distances of a mile or less. This can service high-speed network connections between buildings, though the components can also be used for high-bandwidth interconnections between data center servers. As time goes on, silicon photonics technology should allow for additional optical applications, bolstering the connectivity between servers and allowing for alternative compute models, such as server disaggregation.

Q. Are there any silicon photonics products currently available for data centers?

Practical silicon photonics products are still in their infancy, so it's challenging to predict how they will evolve.

Intel currently touts two silicon photonics products: the Intel Silicon Photonics 100G PSM4 QSFP28 Optical Transceiver and the Intel Silicon Photonics 100G CWDM4 QSFP28 Optical Transceiver. Both products support Ethernet switch, router and telecom connectivity for large enterprise and cloud-scale data centers with 100 gigabit Ethernet low-power optical links using single-mode optical fiber.

Both products also share features, such as a maximum of 3.5 watts of uncooled power dissipation and an electrical-side interface that supports the IEEE 802.3bm CAUI-4 standard, using four 25 gigabits per second (Gbps) data lanes. In addition, the products both employ a compact Quad Small Form-factor Pluggable form factor to allow many modules to be deployed in a relatively small area. The products are also controlled through an inter-integrated circuit bus that supplies a serial link between chips for management and diagnostics.

These early products represent just a sampling of the potential for silicon photonics. As the technology develops, the supported distances may shrink to facilitate faster connectivity options for systems and devices that are closer together.

Q. How does silicon photonics support data center disaggregation?

To understand the potential for silicon photonics, it's important to understand disaggregation in the data center.

Servers are traditionally monolithic entities, so all their components are assembled -- or aggregated -- into the same box. When a server cannot deliver more resources, a business needs to buy more servers in their entirety -- even though it may not need all of those additional resources. For example, if an application needs more CPU cores, an organization can deploy another server to service those added cores. But the memory and other resources in the server may sit idle.

The concept behind server disaggregation is to rethink the server into independent, functional subsystems. For example, processors, memory, storage and other key resources are divided into functional modules, and IT teams can add and rearrange those modules as desired within a common rack. Racks of disaggregated components would then fill the data center.

Moving forward, silicon photonics technology could be a crucial part of the connectivity between future disaggregated compute components, providing fast, reliable and low-cost optical/electronic translations that would work between modules sitting next to one another, scattered within a rack or mounted on racks across a data center.

Q. How will silicon photonics technology evolve into future data center products?

Practical silicon photonics products are still in their infancy, so it's challenging to predict how they will evolve. However, there are some possible directions for data center professionals to consider.

Initially, products like the Intel Silicon Photonics 100G PSM4 and CWDM4 QSFP28 Optical Transceivers will provide simpler, low-power and high-speed networking alternatives for large enterprise and cloud-scale data centers. This can encourage volume adoption of high-bandwidth network technologies within data centers and across metropolitan area network deployments.

Over the longer term, silicon photonics technology should continue to support even faster networking speeds, allowing the eventual move from 100 Gbps to 400 Gbps touted on network development roadmaps.

But the truly exciting future of silicon photonics technology is in chip-to-chip communication, allowing data to move between individual chips rather than between boxes or buildings. Nonvolatile memory chips have been demonstrated, offering the potential to store not just single bits per cell, but multiple bits per cell. The use of silicon photonics for memory poses the potential for vastly increased memory speed, as well as an enormous jump in memory density. This may allow silicon photonics to play a key role in the next major leaps in enterprise computing power.

Next Steps

Silicon photonics jump into DCI strategy with Juniper buy

How silicon photonics boosts data center connectivity

Efficiency gains key to disaggregated servers

Dig Deeper on Server hardware strategy