Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
Keeping up with our constantly connected lifestyle of instant messages and video streaming has its repercussions. Data centers have been gobbling up resources becoming a significant share of the energy used worldwide. To keep up with demand and curb the environmental impact, data centers must become more efficient. A large portion of the increase in data traffic and power consumption is attributed to services enabled by novel methods of computing, Machine Learning (ML) and Artificial Intelligence (AI). Another substantial fraction is due to the transport and routing of data through optical links, which relies on the switching and modulation of optical signals. Barium titanate (BTO), a ferroelectric material, displays a strong Pockels effect, very well suited to control the properties of light beam propagating through it. In this thesis, insights of creating a novel generation of photonic components to fulfill the requirement of performance and power consumption are presented.First, using specialized hardware for ML and AI can make the desired computation substantially more efficient. Programmable Photonic Integrated Circuits (PIC), are a viable solution to build photonic neuromorphic computing units and benefit from the advantages of the photonics domain. However, a non-volatile photonic phase shifter is an important missing piece to build such circuits. Here, BTO-based photonic devices are shown to enable such a memory element.By studying the ferroelectric nature of BTO thin films, multi-state non- volatility is presented. An investigation of imaging the ferroelectric domains of the BTO thin films by means of Second Harmonic Generation (SHG) was conducted. Spatially resolved SHG enables scanning of a 4-5 ÎŒm square area of the BTO thin films, and the ferroelectric domains defined by the orientation of their spontaneous polarization show different SHG responses. With this, a polarization vector is attributed to each pixel of the scanned image, thus an image of the ferroelectric domains is captured.The dynamics of ferroelectric domain switching in the BTO thin films is also studied by using the electro-optic effect. By applying electrical pulses, the ferroelectric configuration of the BTO thin film is modified, and in a photonic device different non-volatile states are reached. By changing the time duration of these pulses a switching model was established that is consistent with the crystalline structure of our sample. A full concept and demonstration of a ferroelectric non-volatile photonic phase shifter is then presented, displaying all the relevant metrics for operation in a PIC. The first ever non-volatile photonic phase shifter demonstrated, a symmetric 3-bit device with switching energies of only 4.6-26.7 pJ, insertion loss of only 0.07 dB, and no absorption change for different levels.Second, by increasing the efficiency of optical switches lower power consumption can be achieved in data centers. The networks within and outside data centers are connected via optical fibers by switches and routers. Increasing the efficiency of optical switches will undeniably boost the performance of the network. Utilizing a material such as BTO with a high Pockels coefficient, low optical loss, and high speed has huge potential for an effective optical switch. A 1 x 8 cascading Mach-Zehnder Interferometer architecture BTO- based optical switch is presented displaying promising results for a next generation optical switch.
, , , ,