Search for: All records

In this paper, we propose IMA-GNN as an In-Memory Accelerator for centralized and decentralized Graph Neural Network inference, explore its potential in both settings and provide a guideline for the community targeting flexible and efficient edge computation. Leveraging IMA-GNN, we first model the computation and communication latencies of edge devices. We then present practical case studies on GNN-based taxi demand and supply prediction and also adopt four large graph datasets to quantitatively compare and analyze centralized and decentralized settings. Our cross-layer simulation results demonstrate that on average, IMA-GNN in the centralized setting can obtain ~790x communication speed-up compared to the decentralized GNN setting. However, the decentralized setting performs computation ~1400x faster while reducing the power consumption per device. This further underlines the need for a hybrid semi-decentralized GNN approach. 

Retroreflective visible light communication and positioning (R-VLCP) is a novel ultra-low-power Internet-of-Things (IoT) technology leveraging indoor light infrastructures. Compared to traditional VLCP, R-VLCP offers several additional favorable features including self-alignment, low-size, weight, and power (SWaP), glaring-free, and sniff-proof. In analogy to RFID, R-VLCP employs a microwatt optical modulator (e.g., LCD shutter) to manipulate the intensity of the reflected light from a corner-cube retroreflector (CCR) to the photodiodes (PDs) mounted on a light source. In our previous works, we derived a closed-form expression for the retroreflection channel model, assuming that the PD is much smaller than the CCR in geometric analysis. In this paper, we generalize the channel model to arbitrary size of PD and CCR. The received optical power is fully characterized relative to the sizes of PD and CCR, and the 3D location of CCR. We also develop a custom and open-source ray tracing simulator – RetroRay, and use it to validate the channel model. Performance evaluation of area spectral efficiency and horizontal location error is carried out based on the channel model validated by RetroRay. The results reveal that increasing the size of PD and the density of CCRs improves communication and positioning performance with diminishing returns. 

Many real-world applications require real-time and robust positioning of Internet of Things (IoT) devices. In this context, visible light communication (VLC) is a promising approach due to its advantages in terms of high accuracy, low cost, ubiquitous infrastructure, and freedom from RF interference. Nevertheless, there is a growing need to improve positioning speed and accuracy. In this paper, we propose and prototype a VLC-based positioning solution using retroreflectors attached to the IoT device of interest. The proposed algorithm uses the retroreflected power received by multiple photodiodes to estimate the euclidean and directional coordinates of the underlying IoT device. In particular, the relative relationship between reflected light magnitude and reflected power is used as input to trainable machine learning regression models. Such models are trained to estimate the coordinates. The proposed algorithm excels in its simplicity and fast computation. It also reduces the need for sensory devices and active operation. Additionally, after regression, Kalman filtering is applied as a post-processing operation to further stabilize the obtained estimates. The proposed algorithm is shown to provide stable, accurate, and fast. This has been verified by extensive experiments performed on a prototype in real-world environments. Experiments confirm a high level of positioning accuracy and the added benefit of Kalman filtering stabilization. 

Convolutional Neural Networks (CNNs) are widely used due to their effectiveness in various AI applications such as object recognition, speech processing, etc., where the multiply-and-accumulate (MAC) operation contributes to ∼95% of the computation time. From the hardware implementation perspective, the performance of current CMOS-based MAC accelerators is limited mainly due to their von-Neumann architecture and corresponding limited memory bandwidth. In this way, silicon photonics has been recently explored as a promising solution for accelerator design to improve the speed and power-efficiency of the designs as opposed to electronic memristive crossbars. In this work, we briefly study recent silicon photonics accelerators and take initial steps to develop an open-source and adaptive crossbar architecture simulator for that. Keeping the original functionality of the MNSIM tool [1], we add a new photonic mode that utilizes the pre-existing algorithm to work with a photonic Phase Change Memory (pPCM) based crossbar structure. With inputs from the CNN's topology, the accelerator configuration, and experimentally-benchmarked data, the presented simulator can report the optimal crossbar size, the number of crossbars needed, and the estimation of total area, power, and latency.