Over the past decade, there has been a dramatic increase in the demand for data center storage and data transfer capacity; this has rapidly amplified the load placed on backplane-to-backplane data server interconnects within each center.
Image Credit: Iridian Spectral Technologies
Furthermore, the demand for data center interconnect bandwidth will continue to see a rise, driven by numerous factors.
Among these factors are improved technologies (e.g., flash memory and solid-state drive) that will see cloud storage becoming a more attractive prospect, as well as new requirements for dynamic allocation of server, storage and network resources.
Both commercial and consumer users demand continuous data availability, aided by distributing virtual computing and storage resources across a number of physical devices, and they expect accelerating speeds attached to that access.
Since a single request can activate several data exchanges between servers in one or more data centers, cloud providers have had to keep pace, maintaining warehouse-scale data center servers, advancing from 10 GbE to 100 GbE network adapters in common use; with 800 GbE (and even to 1.6 TbE) expected to become the standard within the just a few years.2
In a large data center that utilizes typical Clos topology (Fig. 1), hundreds of thousands of interconnects are necessary to sustain effective communication across both internal and external servers i.e., data centers in other locations.
Figure 1. Clos Topology. Image Credit: Iridian Spectral Technologies
Clos offers a direct interconnection between top-of-rack (TOR) switches and other servers, but every leaf switch in the leaf spine architecture connects to every switch across the entire network fabric. The spine switches are in the same category of connections to the leaf switches as the leaf switches are to the TOR switches.
Responding to evolving data center architectures and ever-increasing bandwidth demand, the traditional copper connections used to connect backplane to TOR switches and TOR to leaf switches have been replaced with optical interconnects to facilitate the higher data rate required.3
Consequently, data center designers and operators are persistently on the lookout for more capable optoelectronic transceiver components that are economical, compact and energy-efficient.
While conventional transceiver components used at these data rates in long-haul systems were once considered an option, they are inappropriate for data center use today since they are configured for longer-distance interconnects (leading to higher cost), while typically consuming more power and generally having a larger form factor.
Optical interconnect bandwidth enhancement utilizes single-mode fiber (SMF), multi-mode fiber (MMF), digital coherent optical transmission, dense wavelength division multiplexing (DWDM), multiple spatial modes and other techniques.
Each one of these techniques has been used to enhance optical interconnect data transfer rates to levels closing in on those of the long-haul fiber connections. This is due to the fact that all data passing through long-haul fiber has to be distributed around data centers at the same speed or faster.
The optical transceivers leveraged at each end of the fiber interconnect typically use optical filters to control the different wavelength channels used in WDM, CWDM, DWDM and other multiple-wavelength configurations.
Filters utilized in these shorter interconnect applications (≤ 1 km) use the same base filter coating technology as filters used in long-haul and metro applications. However, the optical design and filter size and thickness are modified in accordance with the needs of the compact data center products’ specific requirements.
In addition to compact, precise form factors, designers combining these parts require high transmittance and limited signal loss — as well as excellent reflectance, if operating in reflection — ensuring they do not have to unnecessarily increase amplification.
Relative energy consumption is also vital in this application, and using the best optical interconnect available optimizes energy efficiency while controlling any associated heat dissipation.
Custom filter solutions — including the conventional single-wavelength bandpass and edge pass units used in WDM systems, as well as DWDM edge filters — have shown to be a cost-effective and useful solution to address such needs. Etalons or other optical filters may also be used in the integrated laser assembly (ITLA) sources leveraged in these systems.
Limited Space ≠ Limited Options
A server backplane brings together many connectors, meaning each and every component containing a filter must be ultra-compact to minimize the volume it occupies.
In large server farms, hundreds or even thousands of servers take up large amounts of space with expensive maintenance costs per square foot since temperature-control of these vast spaces must be maintained while ensuring the physical security of the center.
Consequently, making significant space savings (e.g., trimming 30% or 40% of each component’s physical volume) across hundreds of thousands or even millions of components has a huge impact.
Distributed data centers situated in urban environments — small locations interconnected to form, in essence, one big data center — call for the same space-saving requirements due to the high cost per square foot of real estate in more populous locations.
The cost to manufacture such products at the volumes necessary has pushed a considerable amount of component and filter manufacturing to Asia.
However, Iridian Spectral Technologies can offer over 20 years of filter design experience, proprietary design tools and automation throughout part manufacture, cutting and testing to optimize efficiency and part quality while still conducting its manufacturing processes in North America.
This investment in automation allows Iridian to consistently manufacture etalons and filters at high volumes and maintain high quality standards while minimizing per-unit cost. Iridian routinely sells its product to low-cost manufacturers in Asia and elsewhere as a result of this combination of design expertise and manufacturing excellence.
Iridian Spectral Technologies’ design process is usually based on the client’s stated wavelength preferences: which wavelengths need to be multiplexed and how they will be spaced (e.g., do they require 5 nm, 10 nm or 20 nm spacing between each laser wavelength).
In addition, the client must specify the preferred tolerance on those sources, which helps establish how much bandwidth is necessary. From there, an examination of the conditions of use is conducted — specifically the angle of incidence and the beam’s cone angle.
Next up is the investigation of the form factor and part size based on where the customer needs the component to fit, taking into account both the space available for the component and the standard operating temperature (though given that data centers are usually are air-conditioned, temperature is typically a minor concern).
In design scenarios, such as the one detailed above, Iridian works in close collaboration with the client’s chief technical officer or technical team to provide a filter solution prior to moving on to initial production.
At this point, the client may assess the part and deliver any necessary feedback that can then be applied to a subsequent sample round or progress directly to high-volume production.
Even when conducting this level of meticulousness, Iridian Spectral can distribute filters in as little as four weeks (from receipt of customer specs to initial delivery) and etalons in about eight weeks.
Iridian aims to turn around a design and quote within two days and increase unit delivery into the tens of thousands per week (i.e., volume production) within eight to ten weeks. Consider the following examples:
- Design Changes
A recent high-volume customer was aiming to transition from a free-standing component — where free space is present on both sides of the filter product — to a much thinner part that featured an epoxy mount on the backside (i.e., epoxy applied directly to either send or receive optics).
This feature would allow the client’s module to be much smaller than a free-standing, edge-mounted orientation. The design unearthed a number of challenges related to the cone angle (i.e., how the light is coupled into and out of the filter) that were addressed through an iterative process with the client before increasing to high volume production.
- Volume Demands
A recent Iridian Spectral client required shipping of 10,000 parts a week: etalons measuring 7mm x 1.1mm.
- Precise Form Factors
A recent Iridian client required a part with lateral dimensions 0.5mm and smaller, for which an upgrade to equipment was necessary. The main challenge in that instance was handling after dicing; modifying the vacuum pickup to accommodate a smaller size.
Handling and packaging are the main challenges with miniaturized components, as is optical characterization (in some cases). When producing smaller form factors, optical characterization is typically performed at the wafer level, after which the ‘good areas’ are mapped out, facilitating the use of said good areas, packing them for clients.
Iridian can also offer its services as an additional or replacement source for clients whose existing product has previously been manufactured in high volumes — organizations seeking a supplementary partner to reproduce an identical part with all the same specs as the original, usually at a reduced cost or with higher package yield.
Iridian also ensures a good level of in-house language skills to deal with customers overseas. Moreover, the company has developed habits that enable efficient remote working scenarios, which were in place before the pandemic.
The progression of data centers for cloud storage and distributed computing has pushed the need for even better data transmission rates within data centers. Subsequently, this has necessitated innovative creativity in optical interconnects — both by design and the means of production to meet high-volume demand.
Iridian Spectral Technologies brings more than two decades of experience to the table, offering specialized optical filters and optoelectronics used in these centers.
Automated manufacturing, cutting and testing systems are commonly joined across all of Iridian’s products and product areas, guaranteeing maximum efficiency not only for this market area but across the entire business.
Using the same equipment with little or no reconfiguration, Iridian can load and produce a part for one client on one day and manufacture another client’s product the following day — all without jeopardizing quality or volume.
Furthermore, Iridian utilizes a number of various coating platforms, which enables the company to apply the most appropriate one to the part being produced: larger or smaller optics, higher- or lower-precision specifications.
Iridian is flexible and can adapt to the most efficient platform available for the application. Due to the fact Iridian works in numerous markets, the company possesses a broad range of tools to facilitate the most cost-effective solution, meeting the customer’s needs based on the product requirements.
- Zhou, X., Optical Fiber Technology (2017), http://dx.doi.org/10.1016/j.yofte.2017.10.002
- Ethernet Alliance, 2020 Ethernet Roadmap, https://ethernetalliance.org/technology/2020-roadmap/
- Qixiang Cheng, Meisam Bahadori, Madeleine Glick, Sebastien Rumley, and Keren Bergmen, Optica, Vol. 5, No. 11 / November 2018 / Optica, p 1354-1370 https://www.osapublishing.org/optica/fulltext.cfm?uri=optica-5-11-1354&id=399361
This information has been sourced, reviewed and adapted from materials provided by Iridian Spectral Technologies.
For more information on this source, please visit Iridian Spectral Technologies.