Thursday, May 28, 2015

Silicon Photonics in the Year of Light
As we celebrate UNESCO's International Year of Light 2015, it is with some irony that earlier this month on May 12, 2015 IBM announced [1] it had fully fabricated and tested an integrated wavelength, multiplexed 100 Gb/sec silicon CMOS nano-photonic optical tranceiver chip whose design concept was first announced in December 2010 [2]. IBM's effort is echoed by consortiums of major players in the semiconductor industry, all in hot pursuit of silicon photonics [3] technology, and for good reason. The silicon photonics market is estimated to be worth $975 Million by the year 2020 [4]. Key players in this market include IBM, Intel Corporation and the CLR4 Innovators, PETRA (Japan's Photonics Electronics Technology Research Association), Hamamatsu Photonics, Finisar Corporation, Luxtera, Inc., ST Microelectronics, 3S Photonics, Oclaro Inc., Mellanox technologies, and Infinera Inc. The Optical Fiber Communication Conference and Exposition (OFC) [5] took place in Tokyo in March this year where many of these companies presented reports on their silicon photonic technologies.

IBM's recent announcement places emphasis on its compact device package and process proven fabrication techniques which will enable larger scale manufacturing of the devices for use in cloud computing and data centers in anticipation of “Big Data” network scaling. While IBM's most recent announcement is coincident with the International Year of Light, silicon photonics have been a glimmer in the eyes of semiconductor engineers for the past thirty years. In a recent SPIETV video presentation on silicon photonics, Yurii A. Vlasov of IBM [6] stated that during the past ten years over $1.5 B has been spent in the global quest for light based chip technology.  A portion of this investment resulted in an IBM silicon nano-photonic device design in which interconnected beams of wave guided light replace traditional electrical signals carried on copper circuit paths. Data carried by electrical signals on copper circuits propagate at only 66-70% of the speed of light as determined by dielectric properties near conductors and other physical factors. Device design engineers anticipated the day when this speed limitation would slow network server systems, as well as chip to chip intra-device data flow, and eventually singular on chip communications. To most smart phone owners, it's difficult to imagine a scenario in which electrical signals traveling at 66 to 70% of the speed of light becomes a limiting factor in computer chip design and performance, but in a world where global communications servers are linked by fiber optics, satellites and GPS navigation systems, cumulative device delay times have already created bottle necks in our global networks.

Interestingly, many original glass fiber optic cables transported light at about 70% of light's velocity in a vacuum.  New research and development in fiber optics and materials have resulted in the fabrication of optical fibers which can transport light at better than 99% of its vacuum velocity (186,282 miles/sec).  New silicon photonic devices are being optimized to provide maximum light velocity [7] while simultaneously implementing WDM (Wave Division Multiplexing) to maximize data speeds and throughput.  As such, silicon photonic devices might vary in performance as per the designs of their manufacturers.  

Why are silicon photonics important? I'll review some important observations I've made in my November 2013 blog article, “The Cloud of Nations” [8].

On a global scale we observe that in a vacuum, light travels approximately 186,282 miles per second and can circle the earth in 134 milliseconds (about one tenth of a second). If we consider the earth as a large computer, the continents might be compared to microprocessors interconnected by our global Internet, milliseconds apart. If we observe our immediate, personal photometric sphere of existence, we find that in one nanosecond light travels one foot. How precise must our world be?

Chip Level Clock Skew

In the metrics of semiconductors nanosecond measurements are woefully imprecise and we must calibrate metrology in picoseconds in order to measure the speed of data traversing millimeter sized computer chips. In the semiconductor industry we refer to differences in data arrival time on a computer chip as clock skew. On a real semiconductor device, electrical signals on copper conductors travel at approximately 66-70% of light speed. An acceptable clock skew range approximating 20 to 200 picoseconds (pico = 10^-12) usually provides acceptable device performance but this specification can vary across device types and circuit designs. For each tick of the chip's master system clock, billions of transistor gates must be switched on and off in precise unison. A microprocessor operating at a 2 gigahertz clock speed must have sufficient temporal uniformity across the device so that the arrival time of gate switching signals are all within an acceptable time window. If the switching signal's arrival time falls outside this window, the microprocessor and the program it's running will crash. Careful design considerations go into device fabrication technique and wafer processing to ensure product and performance quality. More information on computer chip clock skew can be found in the paper “Skew Variability in 3-D ICs with Multiple Clock Domains” [9] Hu Xu, Vasilis F. Pavlidis, and Giovanni De Micheli, LSI - EPFL, CH-1015, Switzerland Email: {hu.xu, vasileios.pavlidis, giovanni.demicheli}

As clock speeds increase and device geometries decrease, the <10nm design node will pose additional technical challenges in device timing and data throughput. The evolution of light speed photonic interconnects for on and off chip communication will minimize concerns with clock skew across chips and networks as device structures shrink to critical dimensions (CD) <10nm and become stacked in dense arrays. As the integration of CMOS silicon nano-photonics matures, we may eventually see microprocessors, memory chips and 3D devices structured on photonics instead of conventional copper conductors. In the interim, IBM anticipates that the more immediate advantages afforded by silicon nano-photonic inter-chip speeds could dramatically improve cloud server/network performance, and advance its Exascale super computer initiative which could potentially perform over a thousand times the speed of currently available systems. While the initial focus of silicon nano-photonics targets network centers and cloud servers, the wider proliferation of the technology can profoundly effect our evolving Internet.

IBM's silicon photonic chip will address concerns across networks as four 25 Gb/sec optical tranceivers operating on as many different wavelengths will combine to provide a throughput of 100Gb/sec. Installed in server systems, these optical chips could dramatically improve data throughput, reducing network latency and clock skew by eliminating delays inherent in copper conductors found in servers and data center network interfacing. Using sub 100nm design rules (typically 90nm), electrical and optical device structures are formed on the same substrate to maximize efficiency. IBM estimates the device's bandwidth can accommodate the transfer of a high definition movie (approximately 1.5 to 2 Gigabytes) in two seconds.

To achieve maximum bandwidth over distance, most of today's data centers employ Vertical Cavity Surface Emitting Lasers (VCSEL) distributed on multi-mode optical fiber. Remotely located cloud servers have placed new demands on network resources as even larger bandwidths over greater distances are required. To illustrate the importance of IBM's initial 2010 achievement, note that IBM's on-chip silicon photonic wavelength multiplexing emulates the initial success of dense wave multiplexing utilized in early generation fiber optic networks, the major difference being that IBM has achieved this capability at the chip level.  Previous dense wave fiber optic implementations utilized bulky hardware to enhance long haul fiber optic bandwidth. In 1998, AT&T researchers transmitted 100 simultaneous optical signals, each at a data rate of 10 gigabits per second over a distance of 400 km utilizing dense wavelength-division multiplexing (DWDM) technology [10], allowing multiple wavelengths to be combined into one optical signal. This technique increased the total data rate on one fiber to one terabit per second. Although IBM's current silicon photonic chip design utilizes 4 discreet channel wavelengths providing 100Gbit/sec, data throughput is scalable and could be configured to accommodate many demanding applications. While IBM's silicon nano-photonic initiative is impressive, significant global competition is not far behind.

Intel's CLR4 Silicon Photonic Initiative

The silicon photonic initiative is also being pursued by Intel [11] who has helped form a consortium of companies to agree upon product designs and industry standards. The CLR4 Innovators [12] are an alliance of 27 companies inclusive of Intel, HP, Dell, MACOM and SEMATECH (see the above CLR4 link for a complete list).  In April 2014, Intel outlined its silicon photonic initiative in a presentation [13] which addresses the common requirements of the CLR4. The 100 Gb/sec silicon photonics under development utilizes 4X25 Gb/sec tranceivers which can span 1000 meters to link server centers.

Unfortunately, Intel's initiative was set back in February this year as a key component in its silicon photonic module did not meet specifications and quality control standards. [14] The new modules will not ship till the end of 2015 which means they won't be installed till early 2016. Intel's customers (and possibly CLR4 Innovator members) intending to integrate its silicon photonic technology are in a holding pattern till next year.

PETRA's Silicon Photonic Demonstration

Japan's Photonics Electronics Technology Research Association (PETRA), recently demonstrated [15] 100 Gb/sec transmissions over 300 meters using its silicon photonics device technology. The device's 5mm x 5mm package containing 4x25 Gbp/sec transceivers, can be expanded to 12 or more channels to obtain throughputs greater than 300 Gbp/sec.  This scalability enables accommodation of future demands and upgrades of installed network infrastructure.  

Silicon Photonic Research at IMEC

IMEC is also active in silicon photonic research. In February 2015 this year at the International Solid State Circuits Conference, IMEC and its collaborators released results of recent developments in their laboratories [16] demonstrating a 4x20 Ghz/sec wavelength division multiplexing hybrid CMOS silicon photonics tranceiver.  The larger global effort to standardize silicon photonic performance specifications sets the stage for future implementation over many platforms.

Silicon Photonics and The "Internet of Things"

Silicon photonics can assist us in resolving a potential net neutrality dilemma. Much has been debated regarding the FCC's recent adaption of a net neutrality policy which precludes Internet service providers from charging premiums for high speed commercial traffic on the web while intentionally throttling speeds on their networks. Conspicuous consumption of bandwidth by consumers is now the norm given HD and 4K video programming available on the web. If you travel frequently and patronize fast food establishments or coffee shops and use their complementary Wi-Fi systems, you become acutely aware of the limitations on available bandwidth. Many fast food restaurants preface their Wi-Fi login screens with disclaimers of suitability for a particular use, proclaiming “intended for text and email only” and “not intended for video streaming”. All too often, email and other essential services are slow when compared with Asian and European networks. The intentional throttling of Wi-Fi bandwidth often lowers video resolutions to “wax paper” quality. It's disheartening to see 1080P video rendered at resolutions below that of 1960s era analog broadcast quality. Will the FCC's net neutrality policy effectively address these concerns? Should the concept of net neutrality also accommodate the anticipated “Internet of Things” (IoT)? The IoT envisions the interconnection of “things” on the web to include household utility monitors, appliances, cars, watches, and any imaginable “thing” assigned an IP address. A few months ago I read a story describing hackers launching malware they had placed on someone's refrigerator sporting an on board computer. The fridge's door featured an HD flat screen providing an interactive family kiosk in the kitchen, but connected to the web it also kept malware in cold storage. Should ISP's be required to provide equal bandwidth and network routing for email, video conferencing, refrigerators and toasters? A humorous question, but if you're video conferencing from the fridge while snacking in the kitchen, net neutrality/equality for appliances becomes a pertinent IoT policy discussion. That said, the new IPv6 Internet protocol will provide an exponential increase in available URL's required for new email addresses, web sites and “things” attached to the web. Accordingly, the projected increase in Internet traffic can only be accommodated by enhancing routing systems and cloud services, ensuring network speeds are optimized and remain viable as global connectivity escalates.

In this regard we might agree that from both consumer and enterprise perspectives, silicon photonic solutions are long overdue and will enjoy rapid acceptance when available for shipment and implementation.

Closing Thoughts

- It will be interesting to track the evolution of silicon photonics as market demands are met with the delivery of Intel's technology to the CLR4 group of companies.

- IBM will seek to optimize its positioning in the enterprise markets, leveraging its silicon nano-photonics, cloud server infrastructure and possible ramp of MRAM technology for strategic product applications. IBM's advances could also provide “warp speed” for networks if additionally optimized with its fasp (TM) Aspera [17] file transfer platform.

- IMEC could leverage its own silicon photonic designs as its partnered R&D effort yields finished products ready for deployment.

- PETRA and its member companies have demonstrated competitive silicon photonic technology with the ability to scale for future throughput demands. 

- Possible additions to the silicon photonic market equation are prospects for more efficiently deployed Software Defined Networking (SDN) [18] which could reduce traditional hardware and infrastructure costs while enhancing network speed, efficiency and reliability.

- Lucent Bell Labs has introduced an ultra-dense network architecture designed for silicon photonics at the optical network unit. [19] A proof of principal experiment demonstrated an FDMA architecture providing up to eight 300-MBd 16QAM (Quadrature Amplitude Modulation) subbands providing a bidirectional data rate of 9.6 Gb/sec.

As silicon photonic infrastructure expands to enhance our global networks, we may eventually see the migration of this technology to microprocessor and memory devices for use in super computing and consumer products.  Congratulations to IBM, Intel and the global technology alliance members who are working to enlighten our future with photonics.

Please join me in supporting SPIE and the International Year of Light 2015 [20] (click on the icons below for additional site links).

Thomas D. Jay
Semiconductor Industry Consultant
Thomas D. Jay YouTube Channel

Visit my new Amateur Radio blog at:

Corporate, private entities or publications referenced or linked in this article are the respective owners of their logos, trademarks, service marks, media content and intellectual property.  Unless otherwise disclosed, Thomas D. Jay has no financial interest in companies referenced in blog articles or other published media communications. No representation is made to either buy or sell securities. Opinions expressed by Thomas D. Jay are his own. Thomas D. Jay does not employ or otherwise utilize/authorize third party agents to express his opinions, represent his interests or conduct business on his behalf except where formally contractually designated.  Thomas D. Jay does not agree to indemnify or hold harmless vendors, clients or third parties to related contractual agreements and reserves the right to applicable legal remedies in lieu of arbitration.  These terms are subject to change. Concerned parties should check this blog site for periodic updates.

Acknowledgements and Reference Links

IBM Press release, IBM web site May 12, 2015

IBM Press Release, IBM web site, December 1, 2010


WeSRCH web site, MarketsandMarkets press release uploaded May 27, 2015

OFC web site

Yurii A. Vlasov, SPIETV, YouTube

Nature Photonics


InfoScience web site, Hu Xu, Vasilis F. Pavlidis, and Giovanni De Micheli, LSI - EPFL, CH-1015, Switzerland Email: {hu.xu, vasileios.pavlidis, giovanni.demicheli}

Fiber Optics History web site

Intel web site

Intel web site

Intel web site

PC World web site, Agam Shah, IDG News Service
February 2, 2015

OFC website

IMEC web site

IBM Aspera web site


IEEE Xplore web site

International Year of Light 2015, Official Trailer, YouTube