Science AMA Series: I’m Jessie Rosenberg, a Research Staff Member in the Silicon Photonics team at the IBM TJ Watson Research Center. I do research on CMOS-compatible silicon electro-optic modulation technology for use in inter- and intra-chip low power and high bandwidth optical interconnects. AMA!


Hi Reddit,

I’m Jessie Rosenberg, a research scientist at IBM Watson Research Center. I started my physics career as a young child dreaming of black holes and understanding how the world worked, a dream that eventually drove me to skip high school altogether and start college at the age of 13. In graduate school I was drawn to the field of photonics, the study of light for technological applications, and the many possibilities it offered for improving technology used in everyday life.

When I finished my PhD in Applied Physics at the California Institute of Technology at the age of 23, I immediately started working at IBM on their silicon photonics platform, developing technology to intimately integrate optical and electrical devices onto the same silicon chip. By leveraging the same fabrication techniques that are used today for highly advanced computer systems, silicon nanophotonics brings the advantages of high-bandwidth optical communication down to the chip scale and into mass production, aiming to do the same thing for optics that the integrated circuit did for electronics.

I was honored to be selected as one of the Forbes 30 Under 30 list of innovators in science in 2011, have been an Associate Editor for The Optical Society journal Optics Express since 2014, and served as a Program Chair for the Science and Innovations track of the Conference on Lasers and Electro-Optics (CLEO) 2016. I’m looking forward to taking your questions!

I will be back to answer your questions at 1pm EDT (10am PDT, 5pm UTC).

Did you face many challenges when attempting to surpass high school and earn your PhD at such a young age?


I was lucky to have a lot of support from my family and everyone around me. For my first year of college, arguably the most difficult transition, I attended the Mary Baldwin PEG program, a residential program at Mary Baldwin College for students who want to start college early. Then I transferred to Bryn Mawr College in my sophomore year as a regular student.

For the most part I was surprised that, unless it came up in conversation, generally no one guessed my age. People see what they expect to see! I had a pretty normal college experience (though less partying than most, I think!), with plenty of friends and great experiences in classes. By the time I went to grad school when I was 17, the age difference didn't matter much anymore.

Since photons follow the laws of quantum physics, will we be able to make qubits with light and integrate them into a chip within the next decades ?


Many groups are already working on this! There are many different ways to make qubits – using optical photons as qubits is difficult, as the nonlinearities at optical frequencies are low and it is challenging to create deterministic single photon sources (though researchers are making progress on this!). At IBM we use superconducting qubits, which store information as microwave photons. The quantum computing group here has demonstrated a 5-qubit quantum computer integrated on a chip, which is presented as the IBM Quantum Experience for anyone to run experiments on. The Quantum Experience site is full of tutorials and videos, it’s a great resource to learn about quantum computing in general as well as our work at IBM.

These are stationary qubits, but we also need to be able to transmit qubits over long distances. For that, we need to convert between stationary and long-range qubits, so-called “flying” qubits. Optical photons work well as flying qubits, since they can propagate with low loss in optical fiber. Coupling between stationary and flying qubits is a significant challenge, since it has to be done with high fidelity and without disturbing the information in the qubit. This is a main research area for our group right now.

How close are we to optical processing units?


In general, all-optical processing is a difficult problem, because photons tend not to interact with each other very strongly. Of course, we have nonlinear optics, but that only takes effect at high power levels or in very specific conditions (high quality factor resonators or exotic materials). Compared to electrons, which interact very strongly, it’s much more difficult to make all-optical logic gates, for example. Many groups are working on this, and I’m sure we will get there!

But for now, the main application for optics in computing in the near-term is for interconnects, pulling data on and off the chip and sending it to other computers, memory, etc. In fact, the interconnect bandwidth is the main limitation on today’s computer systems, not the sheer amount of computational power. This is where silicon photonics is seeing its first applications.

Is this intended to replace the electronic connections between two silicon chips? Would we have motherboards with fiber optics on them?


As data rates get higher, the propagation distance over copper interconnects gets shorter. At some distance, there’s a crossover point where transitioning to optics and back will actually save you energy. As optical transceivers become more efficient and required bandwidths get higher, that crossover point becomes shorter. It started out with telecom distances, over many many kilometers, and now is down to the hundreds-of-meters level in datacenters. Over time I definitely think optics will reach into the circuit board!

There has been work, at IBM and other places, on an “optical backplane,” which is essentially what you described. Generally on PCBs the approach is to use an optical routing layer containing polymer waveguides. See, for example, these two papers from IBM on card-to-card and module-to-module data transmission over optical PCBs.

What would you say is your biggest contribution/advancement thus far? How will the future of photonics science help propel our evolution, and possible interstellar travel?

Edit: wurd


Personally, I’m proud of the inline waferscale photonic test system I designed and constructed with the optical test team, that allows us to automatically measure all electrical and optical devices across a silicon wafer, even before the wafers are finished processing. The test system sits in the actual cleanroom of the semiconductor foundry, and tests every photonics wafer early on in the process, checking for certain wafer acceptance criteria. If any of the specs are out of line, that wafer can be scrapped before it goes through the long and expensive series of later processing steps and packaging.

As far as evolution – well, silicon photonics enables larger datacenters and supercomputers. Supercomputers can perform biological simulations such as protein folding, understanding protein folding helps develop better medical treatments, better medical treatments --> evolution???

For interstellar travel – larger datacenters can perform more analytics on bigger astronomical datasets (IBM is already working with the Netherlands Institute for Radio Astronomy to use Big Data to decode the Big Bang), which may help us find more Earth-like planets to colonize out among the stars. Once we solve that pesky problem of getting there…

Can you give a layman's explanation of what your title post means?? struggling to understanding the complicated sciency stuff that you're apparently really good at!

thanks for doing this AMA :)


Sorry for the jargon!

CMOS-compatible: Using the same silicon technology that’s used to make, for example, computer chips – this lets us mass-produce large and complex systems for a much lower cost than assembling all the components by hand.

Electro-optic modulation: Writing the data from an electrical signal onto an optical signal – having electrical data as input and optical data as output. That is my own main area of research as part of the silicon photonics team here at IBM, I’ve worked on resonant ring modulators that can enhance modulation efficiency by increasing the interaction between the electrical and optical signal (see this paper, for example).

Inter- and intra-chip: Both between multiple chips, and within the same chip. This refers to different length scales of communication – from very short reach optical links within one chip (not power-efficient with current technology, but a very interesting area of research!), out to longer links that can reach hundreds of meters to several kilometers within a datacenter, or even metro or telecom scale links that can be much longer.

Low power and high bandwidth: Energy efficient and carrying a lot of data. Datacenters, as the main application for silicon photonics right now, are very power hungry, and a significant portion of that power consumption comes from the communication between different parts of the datacenter. As bandwidth – the amount of data being transferred – increases due to increasing consumer demand, we need to keep the energy requirements for communicating that data in check. With silicon photonics, we can take advantage of a lot of tricks for increasing bandwidth – primarily wavelength division multiplexing, where we send multiple wavelengths of light carrying different data streams over the same fiber, transferring more data overall with a minimal increase in overhead.

Optical interconnects: A communication link between two electrical data sources, where the data is converted from electrical to optical at the input, and back to electrical at the output. In the case of our silicon photonic interconnects, there is a silicon chip, called a transceiver, on each side to perform the conversion, and optical fiber in between.

Thanks for doing this AMA! What key technical hurdles do you foresee needing to be overcome as you push to higher performance, particularly in terms of bandwidth? Any thoughts on the potential of other technologies such as Lipson's graphene on SiN modulator?


The next big jump will come from moving to higher-order modulation formats – encoding data with multiple intensity levels, polarizations, or additional wavelengths instead of just zeros and ones. We already have wavelength multiplexing, but we could add more wavelength channels there. We also have demonstrated a monolithic PAM-4 56 Gb/s modulator, which uses four intensity levels to code additional data at the same clock rate.

Past that, there are many new material systems that are very interesting, such as graphene, polymers, or nonlinear materials. The challenge there will be integrating them with high volume manufacturing and achieving high yield in order to develop a reliable commercial process. Integrating new materials with CMOS technology tends to be a slow process, however history shows that it can be done when it proves necessary!

Will electro-optic modulation ever be able to compete with thermal expansion controlled phase modulation in terms of insertion loss?

One application for integrated photonics that everyone is excited about is synthetic aperture LIDAR, and this apparently requires quite low optical loss to be practical, since the required power is so high. Can we expect to see an elecro-optic phase modulator at some point that's able to compete with heater-controlled phase shifters in terms of loss, but requires much less electrical control power to actuate?


Insertion loss is one issue with electro-optic modulation, the other is simply size. One option for electro-optic modulation would be to use forward-biased PN junctions, which can be short but have significant insertion loss from modulation due to injecting large numbers of carriers into the optical mode. Alternately, reverse-biased PN junctions have lower insertion loss, but must be quite long since the shift in refractive index is small, and that can have an effect on the pixel resolution. Ultimately to use electro-optic modulation in silicon, you need to move a certain amount of charge into and out of the optical mode, which will always involve some insertion loss. Incorporating materials with a linear electro-optic effect may be an option, but of course the materials integration is a challenge on a silicon platform.

How do you generate the light in the first place, given that silicon has an indirect bandgap and III-V semiconductors are generally incompatible with silicon processes?


That’s a very important question! There’s excellent research being done in the area of integrating laser material directly with silicon – depositing germanium on silicon, or wafer-bonding III-V material onto silicon (either pre- or post-processing).

At IBM, we are taking a packaging approach with the laser, attaching a fully processed InP laser die using a self-aligned flip-chip process. This allows for better heat dissipation, and enables the laser to be tested for yield before it’s packaged onto the chip. Since the alignment between the laser and the silicon photonic waveguide is extremely sensitive, we incorporate a solder self-alignment process, where the laser chip is actually pulled into alignment against mechanical alignment features when the solder is annealed. We have shown sub-micron alignment accuracies using this technique, and anticipate high yield and manufacturability.

Do your circuits use coherent (lasers) or non-coherent (regular) light ? Which do you prefer?


Our silicon photonics platform requires coherent light, for many reasons. Most silicon photonic devices operate only with a single polarization, or behave differently for different polarizations, unless they are specifically designed to be dual polarization devices. Also, in the modulator, we use the electro-optic effect from moving carriers in and out of the waveguide to change the refractive index of the optical mode, which changes the phase of the light passing through the modulator. The modulator then includes some type of interferometer (a ring or Mach-Zehnder) to turn that phase modulation into amplitude modulation. If the light is incoherent, the phase also won't be coherent and the modulator will have no effect. We also rely on the coherence of the input light for our wavelength division multiplexer devices, which are formed from a series of cascaded Mach-Zehdner interferometers.

Hi Dr. Rosenberg! Thank you for doing this AMA!

I know that photonics are limited by the wavelength of light and smaller wavelengths carry a lot more energy, which would take more energy to produce and could also cause issues trying to control that energy. What is the current state of the art smallest photonic devices/buses/etc that we can fabricate in a lab, and what do you think is physically feasible for the future?


The considerations determining the wavelength of choice for various photonic applications are not so much the energy of the light, but rather the material systems we are working with and the wavelengths that they absorb at. We also have to contend with industry standards where they exist, such as for optical interconnects in datacenters, where the standard specifies the 1310nm wavelength band. Telecom operates at the 1550nm wavelength band, since that is where an efficient optical amplifier is available. Silicon has high absorption at wavelengths shorter than near-IR, so we wouldn't use visible light on a silicon platform, for example.

Hi Dr. Rosenberg, thank you for doing this AMA!

What challenges have you encountered as a female in a predominantly male industry? What steps does OSA take during its peer review process to avoid potential gender bias from reviewers?


I’ve been lucky throughout my career to have received a tremendous amount of support from amazing women that I’ve had the opportunity to work with. In my experience most women – and most men too! – have been happy to spend a few minutes to give advice, so that’s definitely a supportive network you can take advantage of. Even though there are still definitely some people who maintain a bias, I’ve found even more who are fighting hard to ensure that everyone has an equal chance.

OSA is in many ways a trend setter with CEO Elizabeth Rogan, 2018 President Ursula Gibson and Elizabeth Nolan, Chief Publishing Officer. Having female leaders in photonics sends a powerful message to all the young women in optics: you can get to the top.

As both a reviewer and associate editor (for Optics Express), I can say that I rarely spend time looking at author names except to look up related work and possible reviewers. As an author, if that’s something you’re concerned about, you could use first initials on your papers rather than your full name.

Are you actually talking chip to chip meaning emitter and receiver on each chip or is this a broadcast above type deal and all points receive (see) the same light/info?


At IBM we are focusing on silicon photonic optical transceivers, with a transmitter and receiver on each chip (see this press release).

Thanks for doing the chat today Dr. Rosenberg!

I'm attending CLEO for the first time next month. I know you have been involved with the technical committees, and I was wondering if you had any advice on what I should make sure to see or if you had any tips for getting the most out of the CLEO experience.


Congratulations on attending your first CLEO! I'd suggest attending as many tutorials as you can, we make sure to pick great speakers who can give an informative overview of their respective fields. It's an excellent way to broaden your knowledge base and come up with new ideas! The postdeadline sessions are also one of the highlights of the week, there are only a few concurrent sessions and they showcase the most exciting breaking research. I'd also suggest a walk through the exhibit floor, it can be very valuable to network with vendors and understand what's available on the market. Hope you enjoy your trip!

Additional Assets


This article and its reviews are distributed under the terms of the Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and redistribution in any medium, provided that the original author and source are credited.