Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Lightator: An Optical Near-Sensor Accelerator with Compressive Acquisition Enabling Versatile Image Processing

Mehrdad Morsali, Brendan Reidy, Deniz Najafi, Sepehr Tabrizchi, Mohsen Imani, Mahdi Nikdast, Arman Roohi, Ramtin Zand, Shaahin Angizi

TL;DR

Lightator addresses the energy and latency challenges of cloud-dependent vision in IoT by delivering a photonic near-sensor accelerator with compressive acquisition that enables end-to-end DNN processing at the edge. The design uses a DMVA, MR-based All-in-One Convolver, and a Compressive Acquisitor to map weights to microring resonators and encode activations optically, while keeping activations handled electronically for flexibility. An end-to-end evaluation framework demonstrates 84.4 kilo FPS/W on average and substantial power reductions (up to ~$73\times$) against GPU baselines and prior photonic accelerators, with additional gains from mixed-precision configurations. The work underscores a practical, scalable path to energy-efficient, versatile edge vision with on-chip compression and kernel-size flexibility, reducing cloud dependency and data movement.

Abstract

This paper proposes a high-performance and energy-efficient optical near-sensor accelerator for vision applications, called Lightator. Harnessing the promising efficiency offered by photonic devices, Lightator features innovative compressive acquisition of input frames and fine-grained convolution operations for low-power and versatile image processing at the edge for the first time. This will substantially diminish the energy consumption and latency of conversion, transmission, and processing within the established cloud-centric architecture as well as recently designed edge accelerators. Our device-to-architecture simulation results show that with favorable accuracy, Lightator achieves 84.4 Kilo FPS/W and reduces power consumption by a factor of ~24x and 73x on average compared with existing photonic accelerators and GPU baseline.

Lightator: An Optical Near-Sensor Accelerator with Compressive Acquisition Enabling Versatile Image Processing

TL;DR

Lightator addresses the energy and latency challenges of cloud-dependent vision in IoT by delivering a photonic near-sensor accelerator with compressive acquisition that enables end-to-end DNN processing at the edge. The design uses a DMVA, MR-based All-in-One Convolver, and a Compressive Acquisitor to map weights to microring resonators and encode activations optically, while keeping activations handled electronically for flexibility. An end-to-end evaluation framework demonstrates 84.4 kilo FPS/W on average and substantial power reductions (up to ~) against GPU baselines and prior photonic accelerators, with additional gains from mixed-precision configurations. The work underscores a practical, scalable path to energy-efficient, versatile edge vision with on-chip compression and kernel-size flexibility, reducing cloud dependency and data movement.

Abstract

This paper proposes a high-performance and energy-efficient optical near-sensor accelerator for vision applications, called Lightator. Harnessing the promising efficiency offered by photonic devices, Lightator features innovative compressive acquisition of input frames and fine-grained convolution operations for low-power and versatile image processing at the edge for the first time. This will substantially diminish the energy consumption and latency of conversion, transmission, and processing within the established cloud-centric architecture as well as recently designed edge accelerators. Our device-to-architecture simulation results show that with favorable accuracy, Lightator achieves 84.4 Kilo FPS/W and reduces power consumption by a factor of ~24x and 73x on average compared with existing photonic accelerators and GPU baseline.
Paper Structure (6 sections, 1 equation, 10 figures, 1 table)

This paper contains 6 sections, 1 equation, 10 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Lightator Architecture
  4. Hardware Mapping
  5. Experiments
  6. Conclusion

Figures (10)

  • Figure 1: MR input and through ports’ spectra after imprinting a parameter (using tuning signal). By adjusting the MR's resonant wavelength ($\lambda_{res}$) using the phase shifter, part of the input signal drops into the ring (through the coupling region) towards the drop port while the remaining propagates towards the through port, hence imprinting any parameter in the transmitted signals. FMHW is the full width at half maximum of the resonance spectrum.
  • Figure 2: High-level operational flow of Lightator.
  • Figure 3: Lightator architecture consisting of a sensor array and the optical core.
  • Figure 4: Components of the DMVA: (a) CRC, (b) Selector, (c) VCSEL driver, (d) Sample waveforms of CRC input from the pixel and respective outputs .
  • Figure 5: Implementing a 3$\times$3 kernel in an arm.
  • ...and 5 more figures