|
A key enabler of the IT revolution of the late 20th century was the development of electronic design automation (EDA) tools allowing engineers to manage the complexity of electronic circuits with transistor counts now reaching into the billions. Recently, we have been developing large-scale nonlinear photonic integrated logic circuits for next generation all-optical information processing. At this time a sufficiently powerful EDA-style software tool chain to design this type of complex circuits does not yet exist. Here we describe a hierarchical approach to automating the design and validation of photonic integrated circuits, which can scale to several orders of magnitude higher complexity than the state of the art.
Most photonic integrated circuits developed today consist of a small number of components, and only limited hierarchy.
For example, a simple photonic transceiver may contain on the order of 10 building-block components,
consisting of grating couplers for photonic I/O, modulators, and signal splitters/combiners. Because this is relatively
easy to lay out by hand (or simple script) existing photonic design tools have relatively little automation in
comparison to electronics tools. But demonstrating all-optical logic will require significantly more complex photonic
circuits containing up to 1,000 components, hence becoming infeasible to design manually.
Our design framework is based off Python-based software from Luceda Photonics which provides an environment to describe components, simulate their behavior, and export design files (GDS) to foundries for fabrication. At a fundamental level, a photonic component is described as a parametric cell (PCell) similarly to electronics design. PCells are described by geometric characteristics of their layout. A critical part of the design framework is the implementation of PCells as Python objects. PCell objects can then use inheritance to simplify design, and hierarchical designs can be made by creating composite PCells (modules) which consist of primitive building-block PCells (components). To automatically produce layouts, we built on a construct provided by Luceda called a PlaceAndAutoRoute cell: we create a module component by supplying a list of child cells, and a list of the desired connections between the cells (e.g. the out0 port of a microring is connected to a grating coupler). This functionality allowed us to write algorithms to automatically lay out the components: for instance, by laying out the first component and walking through the list of connections to check to see if the next component is already placed or not. The placement and orientation of the new component is determined by minimizing the length of a connecting waveguide. Our photonic circuits also utilize electrical signals to tune the photonic elements (setting propagation phases or microring resonant frequencies via thermo-optical tuning): the algorithm also routes the contacts for the metal heaters to contact pads at the edge of the circuit being designed where it can be contacted by electrical probes.
We are currently validating a test run fabricated over the summer, and will use detailed characterization results to prepare our final design cycle in which we aim to demonstrate complex operational logic circuits containing ~50-100 nonlinear resonators.
|
