Abstract-Microresonators have been utilized to construct optical interconnection networks. One of the drawbacks of these microresonators is that they suffer from intrinsic crosstalk noise and power loss, resulting in Signal-to-Noise Ratio (SNR) reduc-tion and system performance degradation at the network level. The novel contribution of this paper is to systematically study the worst-case crosstalk noise and SNR in a ring-based ONoC, the Corona. In the paper, Corona's data channel and broadcast bus are investigated, with formal general analytical models presented at the device and network levels. Leveraging our detailed analytical models, we present quantitative simulations of the worst-case power loss, crosstalk noise, and SNR in Corona. Moreover, we compare the worst-case results in Corona with those in mesh-based and folded-torus-based ONoCs, all of which consist of the same number of cores as Corona. The quantitative results demonstrate the damaging impact of crosstalk noise and power loss in Corona: the worst-case SNR is roughly 14.0 dB in the network, while the worst-case power loss is substantially high at -69.3 dB in the data channel.

While deep neural networks (DNNs) deliver state-ofthe-art accuracy on various applications from face recognition to language translation, it comes at the cost of high computational and space complexity, hindering their deployment on edge devices. To enable efficient processing of DNNs in inference, a novel approach, called Evolutionary Multi-Objective Model Compression (EMOMC), is proposed to optimise energy efficiency (or model size) and accuracy simultaneously. Specifically, the network pruning and quantisation space are explored and exploited by using architecture population evolution. Furthermore, by taking advantage of the orthogonality between pruning and quantisation, a two-stage pruning and quantisation cooptimisation strategy is developed, which considerably reduces the time cost of the architecture search. Lastly, different dataflow designs and parameter coding schemes are considered in the optimisation process since they have a significant impact on energy consumption and the model size. Owing to the cooperation of the evolution between different architectures in the population, a set of compact DNNs that offer trade-offs on different objectives (e.g., accuracy, energy efficiency, and model size) can be obtained in a single run. Unlike most existing approaches designed to reduce the size of weight parameters with no significant loss of accuracy, the proposed method aims to achieve a trade-off between desirable objectives, for meeting different requirements of various edge devices. Experimental results demonstrate that the proposed approach can obtain a diverse population of compact DNNs that are suitable for a broad range of different memory usage and energy consumption requirements. Under negligible accuracy loss, EMOMC improves the energy efficiency and model compression rate of VGG-16 on CIFAR-10 by a factor of more than 8.9× and 2.4×, respectively.

Network-on-chip (NoC) based multiprocessor system-on-chips (MPSoCs) have been proposed as promising architectures to meet modern applications' ever-increasing demands for computing capability under limited power budget. Understanding the behaviors of MPSoC applications is the key to design MPSoCs under tight power and performance constraints. In this case study, we systematically examine the computation and communication behaviors of four real applications on MPSoCs based on three popular NoC topologies. We formally model real multiprocessor applications as task communication graphs (TCG) to accurately capture their computation and communication requirements. We publicly release a multiprocessor benchmark suite called COSMIC online, which includes the TCG models. In this work, we analyze the spatial distributions of workloads and traf¿cs for each application, and evaluate their performance and energy ef¿ciency on various MPSoC architectures. Our study shows that fat tree based MPSoCs are good choices for applications requiring high network throughput.