Silicon photonics products are now on the market. Why are they necessary and what does this mean for interconnect manufacturers?
By Neil Shurtz
In 1965, Fairchild Semiconductor’s Gordon Moore published his famous paper outlining a future where innovation would allow the number of transistors on an integrated circuit to double every couple of years. Remarkably, he was proven right for nearly 50 years. However, things eventually started to slow down and in 2015, Gordon Moore himself predicted his own law’s demise within the decade.
Part of the problem is that physically engineering gates at their current miniscule scale is becoming difficult. With conventional designs, it is hard to control the flow of electrons. Several clever innovations in this area present the possibility of allowing Moore’s Law to remain relevant into the future, even if innovation in transistors continues to languish. One of these innovations is silicon photonics.
Building a Balanced Computer
While researchers continue to work on ways to squeeze as many transistors onto microchips as possible, others are focusing on improving the speed and efficiency of everything around the switches at the heart of a microprocessor. No matter how fast and powerful processors become, the infrastructure around them — on-chip and inter-chip interconnects and cabling — will eventually become a bottleneck that limits their performance. With the ability to fit transistors onto chips slowing down, a technology has emerged that can augment current transistor densities for better performance.
In 2003, James D. Meindl of the Georgia Institute of Technology published a paper envisioning the future of Moore’s Law, relying not only on improvements in transistor density but also on the performance of interconnects. While he envisioned a world beyond silicon, the emerging technology of silicon photonics fuses traditional silicon electronics with optical technology. Optical interconnects and related optical technology offer far better performance than their copper counterparts, opening the possibility of building computers with connectivity performance that matches the performance of components like storage media and processors.
Unlocking the full potential of hardware through innovations in interconnects is helping companies develop the storage and compute products being deployed in today’s data center environments.
Fusing Optical and Silicon
The fusion of optical and silicon technologies allows chips with optical components to be built using traditional chip fabrication techniques, and opens the door for hybrid copper-optical components. All of which lends silicon photonics to more straightforward adoption at the intra-chip as well as the inter-chip and inter-machine levels.
Optical connections between chips and processors that integrate optical and silicon technology on the same chip provide significant performance improvements. These applications still face a number of roadblocks, including the high power consumption of the lasers that generate the light rays used by the system. However, optical connections between machines based on silicon photonics are now available and they are meeting some of data centers most pressing needs.
What You Can Buy Today
The current primary use of these transceivers is for switch-to-switch connections within the data center, but the possibilities for their use are quite broad, including inter-data center connections, as some models of the transceivers have a 10km useful range.
The need for such technology will grow as data center architectures continue to evolve in the direction of virtualized networks and “any-to-any” connectivity. These are some of the most easily accessible areas for making performance improvements, and address the needs of applications like AI and machine learning, which are beginning to tax traditional architectures and technologies with their unpredictable data access needs.
Based on the Prizm MT technology, the MXC fiber-optic connector was announced in 2013. This connector was the result of a joint development effort between Corning, Intel, and US Conec. For almost 20 years now, Intel has been involved in developing silicon photonics transceivers. This connector was developed to support this effort and its application as optical PCI-Express. The solution is now being pushed for Facebook’s Open Compute project, where they may be used for optical interconnects within a rack or connecting racks of servers. Because of this, every top connector manufacturer has developed MXC products.
What Connector Companies Need to Know
Quocirca data center analyst Clive Longbottom has said that while copper is not the future for data center connectivity, it has outperformed expectations and continues to hang on, even with the availability of other more mature optical technologies available as potential alternatives.
In addition, Longbottom predicts that the silicon photonics capabilities achievable through retrofitted means, like plugging a card in to the back of a machine, will stop at inter-machine connectivity. Other applications, like intra-chip photonics connectivity, will require a major, expensive, rip-and-replace that will not be financially feasible by many companies for years, whether the technology exists or not.
As with any new connectivity technology, standardization is also a concern. This is particularly true with a technology being pitched as eventually enabling any-to-any connectivity and hyperscale data centers. However, with the technology in such a state of infancy, firm standards have yet to emerge. IBM is using a different standard than Intel, for example.
Connector and cable manufacturers and distributors should anticipate the continuing rise of silicon photonics connectivity. Meanwhile, demand for conventional fiber optic and copper cabling will remain strong as existing systems protected by the costs of rip-and-replace continue to remain in service.
Chip level and inter-chip technologies remain stymied by technical challenges and the financial barriers of adoption. Yet, it seems inevitable that the age of copper will eventually yield to faster interconnects built on moving data with light. With escalating demands from advanced applications like AI and Deep Learning, and high-performance storage and compute hardware increasingly isolated by underperforming networks, the day when it makes financial sense to develop and install these technologies is coming.
To learn more about the systems and hardware that are transforming data centers and high-speed computing, see the report “Fiber Optic Connector and Cable Assembly Market 2017-2022,” available now from Bishop & Associates.
Neil Shurtz is a freelance writer based in Seattle. Shurtz’s areas of interest include sensor and telecom components, autonomous vehicles, and connected infrastructure.
Interested in a specific market? Click a market below for current articles and news.