Product: FDTD Application: HPC on Amazon EC2
“Errors that reach fabrication and found on silicon can be costly and time consuming. The capability to run large scale physical simulations using the mature and approximation free FDTD method gives us a new level of confidence at tape-out.”, Yi Zhang, PIC Lead at Rockley
“Lumerical’s FDTD CPU-based HPC approach proved to be the fastest and most cost-effective. Turn-around of a large physical simulation fits in well with our design schedule and budget. Its compatibility with commercial cloud computing resources enables further scalability at critical times.”, Yi Zhang, PIC Lead at Rockley
Rockley Photonics is a world leader in the development of highly integrated photonic/electronic devices, based upon high-volume manufacturing in its proprietary large-waveguide platform. Rockley embraces the ever-growing trend towards increased complexity for photonics circuits and thus strives to develop new design and manufacturing methodologies. With increased complexity, many effects integral to the operation of the circuit are presently insufficiently considered during design and verification, such stray light interference or impacts of non-uniform temperature or stress fields in the PIC. Unfortunately, current circuit design methodologies based solely on system-level simulation with LVS are unable to capture such effects.
The impact of letting design errors slip through the cracks is significant. If a fault makes it to silicon, the cost of a re-spin will be expensive. More importantly, the increased time-to-market can have an immeasurable impact on market-share and brand reputation.
To date, Rockley has tackled the problem by working to improve DRC and LVS approaches. However, due to the complexity and urgency of upcoming designs, Rockley identified that more must be done to reduce risk. To verify that a design is “first-time-right”, they explored the use of large-scale physical simulations using FDTD.
Lumerical’s HPC Solution on Amazon EC2
Rockley considered several HPC solutions to accelerate large-scale simulations but in the end decided to go with a CPU-based cloud solution on Amazon AWS powered by Lumerical’s FDTD. This decision was driven by the accuracy and run-time performance of Lumerical’s FDTD coupled with its amenability to HPC computing, and the cost-effective flexibility of Amazon’s cloud solution. Flexible “cloud pack license” options offered by Lumerical for its FDTD engines makes this solution attractive as it enables cost-effective large-scale computing for critical periods of the design cycle.
Rockley conducted multiple 2D and 3D single time-domain simulations using Lumerical software on Amazon’s Elastic Compute Cloud (EC2) thus allowing for the extraction of high-resolution spectra. The scale of the simulations was significant for FDTD given the size of the design with a layout of approximately 6 mm2. A variety of hardware configurations was explored, enabling Rockley to choose the optimal configuration for their business and engineering needs.
Lumerical’s HPC solution using Amazon EC2 provided Rockley with a solution that is scalable, cost-effective, and flexible. Rockley was able to confirm the correct functionality of its large-scale photonics devices, thus reducing the risk of re-spin and time-to-market delays. Without Lumerical’s solution this level of verification would have not been possible as the run-time on conventional hardware would be unreasonably long. Further, the cost of procuring dedicated hardware would not be justifiable for a task needed only intermittently for a small portion of the design cycle.