The researchers, who are based at the interdisciplinary Ginzton Lab in Stanford, have successfully employed computer based algorithms to optimise the device structure before making it in the lab in a systematic way which could be applied to similar devices in an array of fields.
A “multiplexer” is a device which reads in a series of inputs and then selects one to be transmitted through the device, thus acting as a ‘single-input, multiple-output’ switch. A demultiplexer is a device which performs the opposite function and is often used in conjunction with a multiplexer. Systems which consist of the two are often simply known as “multiplexers” and are very important for data communication.
Multiplexer systems allow costs to be cut down for systems since they allow several different information streams to be transmitted and received using a single channel, rather than creating a single channel for each stream.
Wavelength division multiplexing, or ‘WDM’, is an important area of silicon photonics and is already used in several devices. The idea behind WDM is to encode different pieces of information using different wavelengths of light which may then be transmitted together, for example down an optical-fibre, before being separated again at the other end. This method is useful because it means you can send a large amount of data whilst still using relatively low transmission rates, which are easier to handle by the devices receiving the data. The demultiplexer made in this paper is a “silicon wavelength demultiplexer” which splits an incoming mix of 1300nm and 1550nm light into two output waveguides, one for each wavelength.
Photonic device design is currently not automated, with much of it being done heuristically, and therefore is not ideal for mass production and commercialisation. The authors of this work hoped that if this process could be automated to the same extent as in electrical circuit design then this would be extremely beneficial.
The authors attempted to automate the design process by developing a set of algorithms which optimize the device design subject to a set of constraints. These consist of both physics constraints, the light must obey Maxwell’s equations, and device performance constraints, the signal must be ‘clean’ enough. They used their algorithm to optimize the device design on a computer before fabricating the optimized structure in the lab using a silicon layer on silicon dioxide. The structure was then tested and found to have very good device performance.