Matthias Freiberger

A multiple-input strategy to efficient integrated photonic reservoir computing

Photonic reservoir computing has evolved into a viable contender for the next generation of analog computing platforms as industry looks beyond standard transistor-based computing architectures. Integrated photonic reservoir computing, particularly on the silicon-on-insulator platform, presents a CMOS-compatible, wide bandwidth, parallel platform for implementation of optical reservoirs. A number of demonstrations of the applicability of this platform for processing optical telecommunication signals have been made in the recent past. In this work, we take it a stage further by performing an architectural search for designs that yield the best performance while maintaining power efficiency. We present numerical simulations for an optical circuit model of a 16-node integrated photonic reservoir with the input signal injected in combinations of 2, 4, and 8 nodes, or into all 16 nodes. The reservoir is composed of a network of passive photonic integrated circuit components with the required nonlinearity introduced at the readout point with a photodetector. The resulting error performance on the temporal XOR task for these multiple input cases is compared with that of the typical case of input to a single node. We additionally introduce for the first time in our simulations a realistic model of a photodetector. Based on this, we carry out a full power-level exploration for each of the above input strategies. Multiple-input reservoirs achieve better performance and power efficiency than single-input reservoirs. For the same input power level, multiple-input reservoirs yield lower error rates. The best multiple-input reservoir designs can achieve the error rates of single-input ones with at least two orders of magnitude less total input power. These results can be generally attributed to the increase in richness of the reservoir dynamics and the fact that signals stay longer within the reservoir. If we account for all loss and noise contributions, the minimum input power for error-free performance for the optimal design is found to be in the ≈1 mW range.

Silicon photonics for neuromorphic information processing

We present our latest results on silicon photonics neuromorphic information processing based a.o. on techniques like reservoir computing. We will discuss aspects like scalability, novel architectures for enhanced power efficiency, as well as all-optical readout. Additionally, we will touch upon new machine learning techniques to operate these integrated readouts. Finally, we will show how these systems can be used for high-speed low-power information processing for applications like recognition of biological cells.

Toward neuro-inspired computing using a small network of micro-ring resonators on an integrated photonic chip

We present in this work numerical simulations of the performance of an on-chip photonic reservoir computer using nonlinear microring resonator as neurons. We present dynamical properties of the nonlinear node and the reservoir computer, and we analyse the performance of the reservoir on a typical nonlinear Boolean task : the delayed XOR task. We study the performance for various designs (number of nodes, and length of the synapses in the reservoir), and with respect to the properties of the optical injection of the data (optical detuning and power). From this work, we find that such a reservoir has state-of-the art level of performance on this particular task - that is a bit error rate of 2.5 10-4 - at 20 Gb/s, with very good power efficiency (total injected power lower than 1.0 mW).