In MALEPHICENT we want to realise reliable predictions of the performance of silicon photonic circuits in the presence of real-life fabrication variations and use this to optimize those circuit designs for maximum yield. Silicon photonics is becoming prevalent as photonic integrated circuit platform for interconnects and sensors. However, the technology is limited in circuit complexity by the design process, as there are no performant techniques to evaluate and compensate the effects of the extreme sensitivity to fabrication variations on the scale of large circuits, especially in combination with driver and control electronics.
In MALEPHICENT we address this by
(i) Creating a scalable modeling framework for electronic-photonic co-design and co-simulation, bridging the gap between the electronic and photonic domains. Leveraging machine learning techniques, we can then optimize photonic circuits at design-time, and actively compensate imperfections at run-time.
(ii) Establishing a robust and efficient design flow for PICs that allows scaling of circuits, while optimizing performance and fabrication yield at the same time. Today, photonic circuit design focuses on maximizing performance, and the yield and tolerance is only evaluated after fabrication. MALEPHICENT’s design framework aims to achieve short first-time-right design, which is fundamental for the rapid progress of modern PICs.