mlsanityData to 13 Oct 2026

A Comparative Analysis of Microrings Based Incoherent Photonic GEMM Accelerators

arXiv 2402.03149 ↗First posted 5 Feb 2024cs.AR cs.ET cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Several microring resonator (MRR) based analog photonic architectures have been proposed to accelerate general matrix-matrix multiplications (GEMMs) in deep neural networks with exceptional throughput and energy efficiency. To implement GEMM functions, these MRR-based architectures, in general, manipulate optical signals in five different ways: (i) Splitting (copying) of multiple optical signals to achieve a certain fan-out, (ii) Aggregation (multiplexing) of multiple optical signals to achieve a certain fan-in, (iii) Modulation of optical signals to imprint input values onto analog signal amplitude, (iv) Weighting of modulated optical signals to achieve analog input-weight multiplication, (v) Summation of optical signals. The MRR-based GEMM accelerators undertake the first four ways of signal manipulation in an arbitrary order ignoring the possible impact of the order of these manipulations on their performance. In this paper, we conduct a detailed analysis of accelerator organizations with three different orders of these manipulations: (1) Modulation-Aggregation-Splitting-Weighting (MASW), (2) Aggregation-Splitting-Modulation-Weighting (ASMW), and (3) Splitting-Modulation-Weighting-Aggregation (SMWA). We show that these organizations affect the crosstalk noise and optical signal losses in different magnitudes, which renders these organizations with different levels of processing parallelism at the circuit level, and different magnitudes of throughput and energy-area efficiency at the system level. Our evaluation results for four CNN models show that SMWA organization achieves up to 4.4×, 5×, and 5.2× better throughput, energy efficiency, and area-energy efficiency, respectively, compared to ASMW and MASW organizations on average.

Key facts. No citing papers in the map yet (data through 13 Oct 2026). Builds on 16 of its 46 references that are in the map. First posted on arXiv 5 Feb 2024.

Builds on

16 of 46 references.

References · 16

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,950
  2. 2014Going Deeper with Convolutionscited by 4,239
  3. 2017ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devicescited by 1,199
  4. 2017In-Datacenter Performance Analysis of a Tensor Processing Unitcited by 617
  5. 2017Efficient Processing of Deep Neural Networks: A Tutorial and Surveycited by 433
  6. 2016Neuromorphic Silicon Photonic Networkscited by 67
  7. 2021CrossLight: A Cross-Layer Optimized Silicon Photonic Neural Network Acceleratorcited by 29
  8. 2021A Survey on Silicon Photonics for Deep Learningcited by 23
  9. 2019Digital Electronics and Analog Photonics for Convolutional Neural Networks (DEAP-CNNs)cited by 22
  10. 2021ROBIN: A Robust Optical Binary Neural Network Acceleratorcited by 15
  11. 2021An Electro-Photonic System for Accelerating Deep Neural Networkscited by 15
  12. 2022A Silicon Photonic Accelerator for Convolutional Neural Networks with Heterogeneous Quantizationcited by 9
4 more
  1. 2018PCNNA: A Photonic Convolutional Neural Network Acceleratorcited by 8
  2. 2021Learning on Hardware: A Tutorial on Neural Network Accelerators and Co-Processorscited by 6
  3. 2022Photonic Reconfigurable Accelerators for Efficient Inference of CNNs with Mixed-Sized Tensorscited by 5
  4. 2023SCONNA: A Stochastic Computing Based Optical Accelerator for Ultra-Fast, Energy-Efficient Inference of Integer-Quantized CNNscited by 4

Built on by

Not cited yet.