Silicon photonics is an emerging approach to the design of components for electronics that aims to replace electricity and copper wires with light and fibre optics. In recent years, many engineers and physicists worldwide have been specializing in this new field of research, with the hope of paving the way towards the development of faster, more energy-efficient and better-performing devices.
In order to work effectively, silicon photonic components should integrate electrical and optical components onto a single piece of silicon. While in principle, this could be achieved, it has so far proved to be very challenging.
Silicon photonics-based devices could have a broad range of applications, such as improving the efficiency of data centres. For instance, the technology could help tech companies to store the growing amount of information exchanged and shared online.
One of the leading companies who are currently investigating the potential of silicon photonics is IBM. A few years ago, researchers working at IBM’s T.J. Watson Research Lab, in Yorktown Heights, New York, created one of the first working prototypes of silicon photonics chips. These ground-breaking components can beam data between racks and servers at speeds of up to 100 gigabytes per second, which could greatly improve and aid the development of faster and more efficient computers.
Benjamin Lee, IBM Research staff member at the IBM T.J. Watson Research Lab in Yorktown Heights, New York
Image courtesy of IBM
Benjamin Lee is one of the researchers who worked on IBM’s silicon photonic chip prototype. For around a decade, Lee has become an expert in short-reach optical interconnects and integrated photonic switching systems, and he has been involved in numerous photonics-related research projects at IBM.
In this interview, Lee outlines some of the recent developments in IBM’s silicon photonics R&D, including a project presented at last year’s VLSI Symposium, a leading international conference on semiconductor technology and circuits that generally takes place in Kyoto, Japan.
Ingrid Fadelli: Firstly, could you briefly introduce yourself and tell us a bit about your academic and professional background, as well as your role at IBM Research?
Benjamin Lee: I am Dr Benjamin Lee, and I was principal investigator for the project that led to the work that Dr Jonathan Proesel is reporting at VLSI. I received my B.S. in Electrical Engineering from Oklahoma State University and my M.S. and Ph.D. in Electrical Engineering from Columbia University.
My graduate thesis work investigated silicon photonic devices and architectures for optical processing systems. I joined IBM as a postdoctoral researcher in 2009 and became a research staff member in 2010.
For the last ten-plus years, I’ve investigated optical components and architectures for accelerating today’s computer communication networks.
IF: Could you tell us how the recent advancements in silicon photonics presented at the VLSI conference came about and what previous research it was based on?
BL: Our project, ONRAMPS–Optical Networks using Rapid Amplified Multiwavelength Photonic Switches–introduces a second-generation chip-scale rapid photonic switching technology, therefore maturing it towards commercialisation.
ONRAMPS has several groundbreaking accomplishments. First, we demonstrated a photonic switch chip with monolithic integration of all the low-level control electronics. As a result, we don’t have to make thousands of electrical connections to the chip: instead, a few simple digital interfaces can be used to program and optimise all aspects of the chip.
Second, we built a strictly non-blocking eight-by-eight (8x8) switch using this technology. The switch is built from a large matrix of optical interferometers and controls the path of light through the switch by applying current into diodes and resistors within those interferometers.
Third, we developed volume-compatible assembly processes and used them to fully package the fibre-attached flip-chip bonded photonic switch module. This is key for bringing down the cost of future optical switches.
IF: How does the technology that you have developed compare to more conventional technologies for similar purposes?
BL: Optical circuit switching has been widely promoted for use within datacentre networks but has failed to provide revolutionary improvements that justify cost. Part of the problem is that most optical circuit switching technologies are expensive and slow, requiring many milliseconds to change the light’s path.
Our technology uses techniques from the microelectronic industry for fabrication and packaging to provide the potential to reach low cost in high volume. Furthermore, the technology can reconfigure the switch state in nanoseconds, which is the timescale of computer communications.
A close-up of a fourth-generation IBM multiport photonic switch chip (released in 2020), which contains eight input ports and eight output ports
Image credit: IBM via IEEE Xplore
IF: More generally, how do you feel that light could be used for different electronics applications? For instance, how could it impact the efficiency and performance of transistors, circuits, and computer processing at large?
BL: Today, light is used extensively in the world’s complex computing systems, such as in the cloud, in high-performance computing, and in the physical connections that comprise the internet. Light is primarily used to transmit information from one place to another, whereas electronics, such as transistors, are primarily used to process information.
As the processing occurs in electronics, when transmitting information over a very short distance, it usually doesn’t make sense to transfer the information into light and transfer it back into electrical signals again. However, as the distance grows between two endpoints, or as the amount of information being transmitted grows, a threshold is reached where using optics to transmit the signal can be more affordable and use less power. As computing systems scale in size, speed, and performance, the relative use of optical communications increases.
Optical circuit switching (OCS) technologies play a converse role. Today, as I transmit a signal optically, I must convert it into an electronic signal at multiple points along the path to compute the route from starting point to ending point. If the amount of transmitted information is very large and the processing required is very small (such as a simple switching function), then it wouldn’t make sense to transfer the signal back and forth from light to electricity. This is the idea behind optical circuit switching: let the signal remain light, and perform simple routing operations on the streams of optical bits as they pass by.
OCS is used widely in very long-distance networks, and it can also bring many benefits to tomorrow’s computer communication networks as well.
IF: Are there any challenges to the use of light in the development of electronics? If so, what are the major ones and how could they be overcome?
BL: There are many challenges to the use of light in the development of electronics, but the primary challenge is rearchitecting computer networks around a circuit-switching paradigm.
IF: On the other hand, what are the key benefits of using light, and what was the key motivation that inspired you to start experimenting with its use in electronics development?
BL: The benefits of OCS technology include its ability to reduce the bandwidth bottlenecks found in computer networks, improve latency, and decrease power consumption. It looks hopeful that silicon photonic switching will bring such benefits into datacentre networks with the speed and cost needed to have an impact. If the technology can be commercialised, all these benefits combine to provide more computation with less energy.
IF: How does your recent work ladder up to IBM's focus on AI and hybrid cloud technology?
BL: The ONRAMPS project (the work reported at VLSI this summer) targets future cloud networks, which require much more tightly coupled heterogeneous compute resources. In these networks, many types of computer chips (CPU, memory, GPUs, AI accelerators, FPGAs, and even quantum accelerators) will need to be tightly coupled, but in a manner that can be flexibly provisioned, efficient, and cost-effective to support diverse client workloads. The photonic switch can intercept these requirements.
IF: How do you see the field of silicon photonics developing in the future? For instance, how could it impact current technology development methods?
BL: Photonic breakthroughs have erupted over the past decade with many successful commercial solutions, but the promise of photonics has always been the ability to integrate many optical devices into a common platform at a low cost. With this work, we illustrate a new photonic function enabled by the large-scale integration of many optical switching devices. If the remaining challenges can be overcome, this technology can bring significant value to future computer networks.
As existing devices are approaching their maximum performance and efficiency, silicon photonics research truly has the potential to bring about huge change, facilitating the development of more advanced and faster technologies. Benjamin Lee and his colleagues at the IBM T.J. Watson Research Lab are continuing their valuable work in the field, with the ultimate aim of preparing photonic components for widespread commercial use.
To read more about Benjamin Lee’s work and keep updated on the new technology that he develops at IBM, visit his researcher profile page at IBM.com.