The design of digital circuits is currently dominated by hardware description languages such as Verilog and VHDL. This automation of circuit design has enabled the development of modern computer processors with billions of transistors.
Integrated photonic devices, on the other hand, are still designed by hand. A designer first picks a basic design based on analytic theory with a handful (2 – 6) of free parameters. The designer then tunes these parameters by running many optical simulations. This brute-force approach is inefficient and limits the designer to a small library of known devices.
Photonics, the science of guiding and manipulating light, is used in many applications such as optical interconnects, optical computing platforms for AI or quantum computing, augmented reality glasses, biosensors, medical imaging systems, and sensors in autonomous vehicles.
For all these applications, Stanford University professor Jelena Vučković said, many optical components must be integrated on a chip that can fit into the footprint of your glasses or mobile device. Unfortunately, the problems with high-density photonic integration are several. Traditional photonic components are large, sensitive to fabrication errors and environmental factors such as variations in temperature, and are designed by manual tuning with few parameters. So, Vučković and her team asked, “How can we design better photonics?”
Billions of U.S. dollars of basic and applied research funding have been invested during the last few years in ideas proposing inverse concepts. The photonics market could not make an exception to this global trend, and thus, several agenda-setting research groups have already started providing sophisticated tools, constrained optimization algorithms, and selective evolution techniques towards this direction.
Photonic Inverse Design
In the photonics inverse design process, scientists rely on sophisticated computational tools and modern computing platforms to discover optimal photonic solutions or device designs for a particular function. In this inverse process, the researcher first considers how he or she would like the photonic block to operate, then uses computer software to search the whole parameter space of possible solutions for the one that is optimal, within fabrication restraints.
What if, however, we were able to search the full space of fabricable devices? If we could do this successfully, we would be guaranteed to improve device performance and shrink device sizes. Unfortunately, the space of fabricable devices is absolutely enormous. For example, suppose we want to design a silicon photonics device with a 1×1 μm design region. If we divide it into 0.1×0.1 μm pixels, easily achievable with modern nanofabrication, and allow each pixel to either contain silicon (1) or not (0), then we have 2100, approximately 1030, possible devices.
Ideally, we’d want an optimization algorithm whose computational cost is independent of the number of free parameters in the design. This would allow us to design devices that exploit the full space of fabricable devices. Thankfully, it turns out that such optimization methods exist! Related methods have long been used in other fields such as aerospace design and machine learning, but they were only recently introduced to optical design.
Our research group has developed software package, Spins, that uses such methods to design arbitrary photonic devices. Spins is available as fully-featured optimization design suite available license through Stanford OTL. In addition, we also have an open source package, Spins-B. Our algorithm allows the user to ‘design by specification’, whereby the user simply specifies the design area and desired functionality of the device, and the algorithm finds a structure that meets these requirements.
One such device we designed using our algorithm is a TE / TM splitter . This device splits the two incident polarizations of light into separate output waveguides within a footprint of 2.8x 2.8 μm. In simulations, it has an insertion loss of only 0.9 dB and a crosstalk of less than 19 dB.
Applying inverse designed photonics to practical environments
Autonomous vehicles have a large lidar system on the roof housing mechanics that enable rotation of a beam to scan the environment. Vučković considers how this could be improved. “Can you make this system inside the footprint of a single chip, which would be just like another sensor in your car, and can it be inexpensive?” Through inverse design, her research group found optimal photonic structures to enable steering the beam with inexpensive lasers that are cheaper, and achieve 5 degrees of additional beam steering, than state-of-the-art systems.
Next up: scaling superconducting quantum processors onto a single diamond or silicon carbide chip. In this example, Vučković harkened back to the 2020 Dresselhaus Lecture delivered by Harvard Professor Evelyn Hu on leveraging defects at the nanoscale. By relying on impurities in these materials at low concentrations, naturally trapped atoms can be very useful for quantum applications. Vučković’s group is working on material developments and fabrication techniques that allow them to put these trapped atoms in desired positions with minimal defects.
“For many applications, letting computer software search for an optimal solution leads to better solutions than what you would design, or guess, based on your intuitions. And this process is material-agnostic, fully compatible with commercial foundries, and enables new functionalities,” said Vučković. “Even if you try to make something a little bit better than traditional solutions — smaller in a footprint or higher in efficiency — we can come up with multiple solutions that are equally good or better than what we knew before. We are relearning photonics and electromagnetics in this process.”
Photonic Inverse Design with Machine Learning
In the past two decades, the prevalence of information technology and the advances of hardware have been greatly accelerating machine learning and data science development. As such, machine learning has become the central research theme in computer vision, natural language processing, speech recognition, and much more.
Besides commercial and engineering applications, machine learning is assuming an ever-growing importance in scientific research. For example, it is becoming an indispensable tool for the design of molecule structures, planning of chemical syntheses, prediction of material functionalities, classification of celestial bodies, detection of high energy particles, and investigation of many-body systems.
Recently, the optical community has been progressively migrating the techniques of machine learning and data science into photonics research, with a number of successful applications including ultrafast optics, optical communication, and optical microscopy
With the astronomical capability of capturing essential features from vast amounts of high-dimensional data, machine learning models have become a promising tool to aid photonic design in various ways.