PHANES: ReRAM-based photonic accelerator for deep neural networks

Resistive random access memory (ReRAM) has demonstrated great promises of in-situ matrix-vector multiplications to accelerate deep neural networks. However, subject to the intrinsic properties of analog processing, most of the proposed ReRAM-based accelerators require excessive costly ADC/DAC to avoid distortion of electronic analog signals during inter-tile transmission. Moreover, due to bit-shifting before addition, prior works require longer cycles to serially calculate partial sum compared to multiplications, which dramatically restricts the throughput and is more likely to stall the pipeline between layers of deep neural networks. In this paper, we present a novel ReRAM-based photonic accelerator (PHANES) architecture, which calculates multiplications in ReRAM and parallel weighted accumulations during optical transmission. Such photonic paradigm also serves as high-fidelity analog-analog links to further reduce ADC/DAC. To circumvent the memory wall problem, we further propose a progressive bit-depth technique. Evaluations show that PHANES improves the energy efficiency by 6.09x and throughput density by 14.7x compared to state-of-the-art designs. Our photonic architecture also has great potentials for scalability towards very-large-scale accelerators.

Link

This is link to the paper published on DAC’22.

Towards Ultra-fast Photonic Hyperdimensional Computing Accelerator

Authors: Jiaqi Liu, Peiyu Chen, Wei Zhang, Jiang Xu.

In this paper, we propose the first Photonic Hyperdimensional Computing (HDC) accelerator to overcome the challenges of high-dimensional data processing. Our accelerator incorporates a Microring-based homogenous configurable Photon-Core that can be dynamically reconfigured to either the encoding stage or the classification stage, enhancing parallelism and throughput. We introduce a photonic record-based encoding scheme that leverages the wavelength-division multiplexing (WDM) and free spectral range (FSR) properties to improve the efficiency of the encoding stage. Additionally, we propose a customized high-speed photonic similarity check module that supports the calculation of Cosine similarity distance, enabling accurate classification results.

PhotonGraph: Photonic Accelerator for Graph Processing Algorithms

Authors: Jiaqi Liu, Xianbin Li, Xinyu Chen, Yinyi Liu, Peiyu Chen, Fan Jiang, Wei Zhang, Jiang Xu.

In this paper, we propose Theia, the first-ever phtonic accelerator for graph processing. The photonic cores in Theia which leverages Microring resonators to perform ultra-high-speed graph computations in the optical domain. Its architecture also incorporates the HBM-enabled optical network-on-chip (ONoC) to address the stringent memory access requirements of large-scale graph processing models. Additionally, a specialized data mapping scheme utilizing wavelength routing is introduced to enhance the efficient shuffling and routing of data structures in sparse and irregular graphs.

ACCL: Towards Accelerating Hyperdimensional Computing with FPGA

Authors: Jiaqi Liu, Xianbin Li, Xinyu Chen, Yinyi Liu, Peiyu Chen, Fan Jiang, Wei Zhang, Jiang Xu.

In this study, we present the first-ever Hyperdimensional Computing (HDC) accelerator based on Versal ACAP AIEngine designed to address the challenges associated with high-dimensional data processing. Our proposed accelerator features a homogeneous AIEngine array that can be dynamically reconfigured for either the encoding or classification stage, thereby enhancing parallelism and throughput. We also introduce a parallel record-based encoding computing scheme that capitalizes on the on-chip network to optimize the efficiency of the encoding stage. Moreover, we propose a customized high-speed very-long-instruction-width (VLIW)-based similarity check module capable of calculating Cosine similarity distance, which facilitates accurate classification results.

PhotonNTT: Energy-efficient Parallel Photonic Number Theoretic Transform Accelerator

Authors: Xianbin Li(co-first author), Jiaqi Liu(co-first author), Yuying Zhang, Yinyi Liu, Jiaxu Zhang, Chengeng Li, Shixi Chen, Yuxiang Fu, Fengshi Tian, Wei Zhang, Jiang Xu*; Design, Automation and Test in Europe Conference and Exhibition (DATE), Valencia, Spain

For the first time, we present a photonic NTT accelerator, PhotonNTT, with high energy efficiency and parallelism to address the above challenge. Our approach involves formulating the NTT into matrix-vector multiplication (MVM) operations and mapping the data flow into parallel photonic MVM units. A dedicated data mapping scheme is proposed to introduce free spectral range (FSR) and distributed RAM design into the system, which enables a high bit-wise parallelism level. The system’s reliability is validated through the Monte-Carlo bit error rate (BER) analysis. The experimental evaluation shows that the proposed architecture outperforms SOTA CIM-based NTT accelerators with an improvement of 50x in throughput and 63x improvement in energy efficiency.

Learning Styles Identification Model in a MOOC Learning Environment

Authors: Huang Jingya, Jiaqi Liu, International Conference on Computer Science and Education(ICCSE), Ningbo, China

Different online learners have different learning styles that are influenced by their prior knowledge and personalities, which necessitates the use of an online platform to identify these learning behaviors in order to enhance the course. Based on the Felder-Silverman model, we offer a novel learning style theory model suitable for MOOC education environments in this work. Then we extract high-dimensional features from the MOOCCube data set produced from China’s XuetangX platform. Furthermore, to identify online users’ learning styles, we apply a two-level hierarchical learning style classification model. First, a learning autonomy classification model is used to filter inactive learners by collecting the learner autonomy index from the data set. Then, to detect distinct learning styles, we construct a clustering-based behavior identification model using the Gaussian Mixture Model. Our hierarchical classification model demonstrates great capability and enables researchers to conduct analytical studies on the learning patterns of online learners.