A group of three principal investigators will undertake a holistic study of prospects of nano-photonic interconnects in the context of integration with silicon technologies for improved computational speed, latency and lower power dissipation. An unique aspect of the project is that the PIs come from different backgrounds, namely photonic devices and computer architecture, and an integrated design methodology that will exploit feedback from devices to architectures will be central to this high-risk project with potential for large payoff. Specifically, this project will examine thermal sensitivities of nano-photonic devices and study its influence on computing architectures making use of optical interconnects. Conversely, the requirements imposed by architectural considerations on the device technology will be studied as well.
The potential broader impact of the project is largely technological. If successful, t has the potential to revolutionize the interconnect technology, which is one of the major bottlenecks in further progress of large scale computer hardware design today. Nevertheless, the PIs also plan to disseminate the results of their findings through workshops and class room efforts and to include women and underrepresented groups in their activities, which should enhance the broader impact of the project.
Integrated silicon nanophotonic devices are part of an emerging technology with the potential to drastically reshape the computing landscape, by challenging assumptions about microprocessor design based on traditional technologies. One of the primary applications for silicon nanophotonic devices is in designing interconnection networks on chip that use modulated laser light, in place of traditional electrical signaling, for communication among the various components in the system (e.g. CPU to CPU CPU to memory, etc.) Researchers in the separate fields of computer architecture and nanophotonic devices, however, have struggled to make a compelling case for the future adoption of these new techniques, in part because there has been a gap in understanding between computer architects and device researchers. This EAGER project set out to accomplish three objectives, which were met as follows: 1) Create a parametric library of nanophotonic devices that can be "plugged" into a circuit/architecture-level simulator, to be able to readily incorporate device-specific information into higher-level studies. Through an interdisciplinary team of graduate students, we studied and characterized many of the components that have been investigated in-house at Co-PI Lipson's group. We built a library of all the components and their relevant parameters. 2) Explore device and architectural solutions to optical power consumption, which ultimately affect temperature and stability of such designs. In one conception, we propose power-insensitive silicon microring resonators without the need for active feedback control. The passive control of the resonance is achieved by utilizing the compensation of two counteracting processes: the free carrier dispersion shift to shorter wavelengths (blueshift) and thermo-optic shift to longer wavelengths (redshift) of resonances. In another conception, we propose "athermal" modulators that consist of a ring resonator coupled to a Mach–Zehnder interferometer (MZI) with tailored thermal properties. The modified MZI, which compensates the temperature dependence of the ring resonator, is designed by considering the optical paths of the ring and the MZI. 3) Understand the semantic gap between architecture and nanophotonic device researchers. This project was extremely valuable to learn how architecture and device researchers can interact closely. Although we had success with architectural studies that loosely interacted with device researchers to obtain latency/bandwidth/power parameters, the level of mutual understanding that was accomplished through this closer interaction was significantly higher. There are significant semantic barriers across these disciplines, and the fact that both are fast-moving further complicates matters. We believe a library like the one we developed is an excellent means of communication.
