15 May 2018 in IEEE Journal of Selected Topics in Quantum Electronics
We present in our work numerical results on the performance of a $\mathbf {4 \times 4}$ swirl-topology photonic reservoir integrated on a silicon chip. Nonlinear microring resonators are used as nodes. We analyze the performance of such a reservoir on a classical nonlinear Boolean task (the delayed XOR task) for: various designs of the reservoir in terms of lengths of the waveguides between consecutive nodes, and various injection parameters (injected power and optical detuning). From this analysis, we find that this kind of reservoir can perform–for a large variety of parameters–the delayed XOR task at 20 Gb/s with bit error rates lower than $\mathbf {10^{-3}}$ and an averaged injection power lower than 2.5 mW.
25 July 2019 in IEEE Journal of Selected Topics in Quantum Electronics
As the performance increase of traditional Von-Neumann computing attenuates, new approaches to computing need to be found. A promising approach for low-power computing at high bitrates is integrated photonic reservoir computing. In the past though, the feasible reservoir size and computational power of integrated photonic reservoirs have been limited by hardware constraints. An alternative solution to building larger reservoirs is the combination of several small reservoirs to match or exceed the performance of a single bigger one. This work summarizes our efforts to increase the available computational power by combining multiple reservoirs into a single computing architecture. We investigate several possible combination techniques and evaluate their performance using the classic XOR and header recognition tasks as well as the well-known Santa Fe chaotic laser prediction task. Our findings suggest that a new paradigm of feeding a reservoir's output into the readout structure of the next one shows consistently good results for various tasks as well as for both electrical and optical readouts and coupling schemes.
31 October 2018 in IEEE Transactions on Neural Networks and Learning Systems
As Moore's law comes to an end, neuromorphic approaches to computing are on the rise. One of these, passive photonic reservoir computing, is a strong candidate for computing at high bitrates (>10 Gb/s) and with low energy consumption. Currently though, both benefits are limited by the necessity to perform training and readout operations in the electrical domain. Thus, efforts are currently underway in the photonic community to design an integrated optical readout, which allows to perform all operations in the optical domain. In addition to the technological challenge of designing such a readout, new algorithms have to be designed in order to train it. Foremost, suitable algorithms need to be able to deal with the fact that the actual on-chip reservoir states are not directly observable. In this paper, we investigate several options for such a training algorithm and propose a solution in which the complex states of the reservoir can be observed by appropriately setting the readout weights, while iterating over a predefined input sequence. We perform numerical simulations in order to compare our method with an ideal baseline requiring full observability as well as with an established black-box optimization approach (CMA-ES).
2 September 2020 in Scientific Reports
Physical reservoir computing approaches have gained increased attention in recent years due to their potential for low-energy high-performance computing. Despite recent successes, there are bounds to what one can achieve simply by making physical reservoirs larger. Therefore, we argue that a switch from single-reservoir computing to multi-reservoir and even deep physical reservoir computing is desirable. Given that error backpropagation cannot be used directly to train a large class of multi-reservoir systems, we propose an alternative framework that combines the power of backpropagation with the speed and simplicity of classic training algorithms. In this work we report our findings on a conducted experiment to evaluate the general feasibility of our approach. We train a network of 3 Echo State Networks to perform the well-known NARMA-10 task, where we use intermediate targets derived through backpropagation. Our results indicate that our proposed method is well-suited to train multi-reservoir systems in an efficient way.
20 May 2022 in Computer Vision and Image Understanding