The attoworld team is pleased to welcome Prof. Eli Yablonovitch to the Max Planck Institute of Quantum Optics. He is giving a colloquium talk on the 11th of July, where he will talk about his unconventional ideas for solving complex optimization problems. On this occasion, he will receive the “MPQ Distinguished Scholar” award, which aims at broadening and strengthening the relations between the Max Planck Institute of Quantum Optics and other research institutes worldwide by inviting eminent scientists to work at the Max Planck Institute of Quantum Optics for an extended period of time.
Prof. Yablonovitch's research interests include silicon photonics, telecommunications, optical antennas, new forms of photovoltaics, and the search for a low-voltage replacement for the transistor. Recently he has been investigating physics-based computing approaches to solving hard problems, such as the traveling salesman problem. Here Eli Yablonovitch talks to Thorsten Naeser about his research, some fascinating scientific ideas, and how the world might get its energy in the future.
You introduced the idea that strained semiconductor lasers could have superior performance due to the reduced valence-band (hole) effective mass. Could you elaborate on this idea?
Yes, semiconductor lasers are the dominant form of laser in the world, driving optical fibers that connect us to the internet. Every time we check our email or download a web page, we connect via a semiconductor laser. In semiconductor lasers, the stimulated emission is from electrons in the conduction band to holes in the valence band. Ideally, the effective mass should be as low as possible to make it easy to fill the bands, better to achieve population inversion. But in nature, the hole effective mass is heavier, making it harder to achieve population inversion.
There are actually two types of holes, heavy and light, which are degenerate at the top of the valence band. Any perturbation that breaks spherical symmetry could split the heavy and light holes, and then a light hole band would emerge at the top of the valence band, making it easier to achieve population inversion. It is most easy to achieve the spherical-symmetry-breaking by a slight lattice mismatch, for example by replacement of a pure layer of gallium atoms with a few percent indium atoms.
This insight greatly improves the output and reliability of semiconductor lasers, so that they form a reliable, permanent, communications fabric that is used by almost all of humanity, every day.
You are regarded as the father of the photonic band gap concept and coined the term “photonic crystal”. The geometric structure of the first experimentally realized photonic band gap is sometimes called “Yablonovite”. Could you tell us the story behind it?
In 1985, there appeared about a half-dozen proposals to make micro-cavity lasers that would suppress spontaneous emission that always competes with stimulated emission. Today we would call these Vertical Cavity Lasers, (VCSEL’s), which use dielectric layered periodic mirrors. But they don’t affect spontaneous emission very much, since spontaneous emission light can easily escape in the horizontal direction, even when the vertical direction is blocked. So I tried to solve this problem, and it was apparent that periodicity was needed in the horizontal directions as well, to block all directions of spontaneous emission.
So I tried to draw two-dimensional periodicity in the horizontal plane, and when I squinted at the paper, it appeared as a chessboard. Then I imagined the vertical periodicity, and it looked like three-dimensional chessboard, which falls under the face-centered-cubic category. But face-centered-cubic is the roundest Brillouin zone among the common crystal structures. If not, it could block light in one direction, but not in some other direction. Then I realized I had something because it was not otherwise obvious that the photonic crystal should be face-centered-cubic. A few years later it was discovered that the ideal photonic crystal should have a diamond structure, a sub-category of face-centered-cubic.
In spite of this progress, spontaneous emission is still not controlled, in technology applications like telecommunication lasers. Nonetheless, the 2-dimensional version of a photonic crystal is used as a coupler in Silicon Photonics that is used for internal telecommunications in data centers. The company that makes these silicon communication chips is Luxtera, and it was acquired by Cisco Systems. Once again, if you check your email, your message is most likely going through a 2-dimensional photonic crystal.
You are also exploring a low-voltage replacement for the transistor. How good are the chances that this will succeed?
Under US National Science Foundation sponsorship we studied this problem for 10 years. Among the requirements, in addition to low voltage, an on/off ratio of 106 is required, as well as an on-state conductance of ~1milli-siemens per m of size. It is very difficult to achieve all three requirements in one device, the most difficult being the on/off ratio. A big energy barrier is needed to prevent current leakage, but if there are any defects, the current will leak through those defects. In effect, we need material perfection better than 1 part in 106 to prevent current leakage.
It appears this might be possible in carbon nano-ribbons, which are even smaller than carbon nano-tubes. But this will require 99.9999% perfect yield of the molecular scale nano-ribbons. This may require chemical purification methods to achieve the desired yield. This plan will demand years of materials research, to achieve this, and to fulfill the other requirements for a practical electronics technology.
You also do research in photovoltaics. What approach do you and your team take to this?
We believe that GaAs solar cells can be cheaper than silicon solar cells. GaAs is 10x more expensive than silicon, but it can be 100 times thinner, using less material, since the GaAs direct bandgap has a much higher optical absorption coefficient. To obtain the GaAs thin film, we use Epitaxial Liftoff, which peels ultrathin layers of epitaxial GaAs from a reusable GaAs substrates. But there was a surprise when making the new solar cells at Alta Devices Inc. Alta was achieving a higher open circuit voltage than had ever been achieved in GaAs solar cells. It revealed a new principle in photovoltaics: “A great solar cell also needs to be a great LED”.
Until 2011, the GaAs cells were grown on GaAs. But then the luminescent light would enter the GaAs substrate and become lost. In the epitaxial-liftoff cells, the substrate was removed and replaced by a thin film metallic mirror. That thin-film mirror provided mechanical support, but also redirected the downward luminescent photons back into the cell, to be re-absorbed, and create a higher density of electrons and holes. A higher density of excitation means a higher voltage. Using this idea Alta raised the efficiency record for 1 sun cells from 25% to 29.1%, which is a world record that will likely stand for a long-time.
But, business conditions were bad for Alta. The world has been awash in subsidized Chinese silicon panels, whose manufacture became a “jobs program” in China. It became impossible to pay for the development of a new material technology, when there was so much excess capacity in China. Alta had to close in 2019, but they are likely to hold the world record for decades to come.
How efficient do you think solar cells can become? Can they solve our future energy problems?
The demonstrated efficiency for single-junction flat-plate cells is 29.1%, and the costs are coming down. I don’t speak of a single technological solution in the near term. Solar panels are part of a blend of different energy technologies.
And in this context: What will our energy supply look like in, say, 100 years?
Solar panels will become the cheapest form of primary energy. They are mass manufactured under controlled conditions in factories, where the benefit of the “experience curve” will continue to bring costs down. I see the world transitioning from hunter/gatherer forms of energy, like coal and petroleum pulled from the ground, to a more agricultural model, where energy is converted from sunlight, as in farms.
But there is also great potential in conventional agriculture, where CRISPR gene-editing will raise the efficiency of leaves from ~4% today to ~30%, competing with the best solar panels, and especially storing the energy in carbo-hydrates.
In the very long-range future, the Physicist Freeman Dyson visualized a sphere of solar panels surrounding the sun, now called a “Dyson Sphere”.
And last but not least. What about your physics-based computing approaches to solving important questions like the traveling salesman problem? Could you please give us an idea of what you have in mind?
In physics, we usually write equations, but we could equally well write inequalities representing the same physics: The principle of least action, a system likes to go to the lowest energy level, the principle of least time, and several other such inequalities.
Each inequality in physics presents an opportunity for physics to directly perform optimization. The hardest problems in computer science are usually represented as optimization problems, among which is the traveling salesman problem, which has now been superseded by the “Amazon Prime” problem. If the physical machine is constantly being externally driven, as in electronic circuits, the relevant inequality is Onsager’s minimum heat generation principle, more universally called the principle of miniumun entropy generation.
There are now many published physical machines which solve the Ising problem; how to arrange magnets on a kitchen table, in order to minimize the energy? Most of these machines are successful, based on Onsager’s principle. Machine intelligence is also an optimization problem. In my talk at MPQ, I will try to project how physics based computing can find its role in the computing industry.