Community-developed photonic inverse design built on Lumerical’s industry leading FDTD platform and Python API delivers unique solution
Vancouver, BC (1 March 2019)
Lumerical Inc., a leading developer of photonic design and simulation tools, announces the immediate availability of Pythonbased photonic inverse design (PID) built on Lumerical’s industry leading FDTD Solutions. This adjoint optimization-based PID implementation was developed by photonics Python community author Christopher Lalau-Keraly and interfaced to FDTD Solutions through Lumerical’s Python Automation API for use in integrated photonics.
Integrated photonics is already having a significant impact on telecommunications and data centers and it holds potential for a major impact in many other applications, such as health care, autonomous vehicles, and quantum computing. Putting powerful approaches like PID in the hands of photonics designers will help open the door to this new generation of devices and end products delivering these innovations to reality much sooner than with traditional methods.
Advancing the State of the Art
Reminiscent of other major shifts in designer productivity based on moving to design at a higher level of abstraction such as logic synthesis for electronics, PID fundamentally changes what is possible. Employing powerful optimization algorithms that leverage adjoint sensitivity analysis to explore design spaces that are impractical by other means, PID coupled with Lumerical’s industry leading FDTD Solutions enables designers to explore hundreds, thousands or more design parameters. This contrasts with traditional design approaches limited to a small number of established device designs, exploring much smaller parameter spaces, often with less than 10 parameters. This capability promises to enable the design of devices with entirely new functions, and more robust designs for existing devices leading to enhanced yield. In this way, PID is key to improving device quality and yield to help PICs become commercially viable, and able to scale.
“At Hewlett Packard Enterprise, our silicon photonics team has been exploring photonic inverse design. For us, it is very important that the eventual tool flow of future photonic designers in enterprise companies is more reliable because all of the pieces having commercial-grade support, combined with the flexibility of Python support. Lumerical’s PID delivers on this ideal situation,” said Thomas Van Vaerenbergh, Photonics Research Engineer at HPE. “PID allows photonic designers to squeeze out the last remaining dB’s and mdB’s loss reductions in state-ofthe-art fabrication platforms, a capability previously not available to photonic designers. PID’s access to the gradient information results in better results compared to black box methods with a similar application space such as CMAES.”
Lumerical Commitment to the Community
Christopher Lalau-Keraly’s open source PID offering ‘lumopt’, an extension of his work1 as a graduate student within Prof. Eli Yablonovich’s research group at UC Berkeley, is a parameterized geometry variant in which optimization is applied to a pre-defined parametric design shape. Lalau-Keraly said “I have always found Lumerical to be extraordinarily responsive to their customers. When they heard that I was developing this open source capability, they reached out to me, and were very supportive in providing their Python API, along with guidance and testing that hardened my open source PID implementation to be production-worthy.”
PID has already been adopted to benefit the wider community. University of British Columbia Prof. Lukas Chrostowski said “I have already inserted Lumerical’s PID into my edX silicon photonics course and OFC short course 432. It is exciting for students to compare their inverse design model created by an industrial grade tool with experimental data after we fabricate their design. It won’t be long before PID will be commonly taught in universities worldwide as companies will be looking to hire photonic designers with PID experience.”
The Transformative Power of Inverse Design
As an industry leader in photonics innovation, Lumerical strives to improve the way photonics systems are designed. “The initial response to our release of inverse design was overwhelming,” said Adam Reid, VP Engineering at Lumerical. “Numerous participants expressed relief that such a capability is now available in a commercial-grade simulation. The most common response was ‘When can we get it?’ One large commercial user told us that for their project, a first-of-its-kind multi-layer photonic component, meeting an aggressive efficiency spec and on a tight deadline would not have been able to succeed without lumopt and the Lumerical inverse design capability. Other customers have created designs with PID within hours that are comparable to the best designs that took months or years to design as published in prestigious scientific journals.”
Continuing on the parametric shape based inverse design capability announced today, Lumerical Labs, the advanced R&D group within Lumerical, is continuing to develop an advanced capability based on topology optimization (TO) in which designs are generated automatically from scratch without the need for specifying an initial shape. TO provides yet another level of optimal designs beyond that of the current parameterized variant.
With the release of PID, designers are now able to quickly generate layouts for tape-out using design tools they are already familiar with. Gilles Lamant, distinguished engineer at Cadence Design Systems said, “Through our collaboration with Lumerical on this project, we have pushed photonics innovation and demonstrated that a common mathematical model can be used in both the optimization environment and in the industry-standard Virtuoso Layout Suite using the Virtuoso CurvyCore technology. This is a key element in taking inverse design technology from academia to the industrial stage, enabling designers to access these advanced algorithms to optimize photonic device area, bandwidth, insertion loss and robustness.
Lumerical will demonstrate PID at the OFC Lab Automation Python Hackathon on Sunday, March 3rd at 8:00 pm in room 17B in the San Diego Conference Center and throughout OFC at Lumerical booth #5438.The Python based open source PID implementation now ships as part of Lumerical FDTD Solutions as a turnkey solution including examples in the application gallery. You can find the original source code on the GitHub site at https://github.com/chriskeraly/lumopt. The enabling Automation API is available immediately from Lumerical at https://www.lumerical.com/products/aapi.
Lumerical develops photonic simulation software – tools which enable product designers to understand light, and predict how it behaves within complex structures, circuits, and systems. Since being founded in 2003, Lumerical has grown to license its design tools in over 50 countries and its customers include 10 of the top 15 companies in the S&P 1200 Global IT index, and 46 of the top 50 research universities as rated by the Times Higher Education rankings. Lumerical’s substantial impact on the photonic design and simulation community means its tools are among the most widely cited in the scientific press, with references in more than 10,000 scientific publications and patents. Lumerical enables its customers to achieve more with light and establish a leading position in the development of transformative technologies employing photonics.
1. Adjoint shape optimization applied to electromagnetic design, Christopher M. Lalau-Keraly, Samarth Bhargava, Owen D. Miller, and Eli Yablonovitch. Optics Express, 9 September 2013.