Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Architecture-Level Modeling of Photonic Deep Neural Network Accelerators

Tanner Andrulis, Gohar Irfan Chaudhry, Vinith M. Suriyakumar, Joel S. Emer, Vivienne Sze

TL;DR

This work shows that similarities between Compute-in-Memory (CiM) and photonics let us use CiM system modeling tools to accurately model photonics systems, and demonstrates optimizations that reduce conversions and DRAM accesses to improve photonic system energy efficiency.

Abstract

Photonics is a promising technology to accelerate Deep Neural Networks as it can use optical interconnects to reduce data movement energy and it enables low-energy, high-throughput optical-analog computations. To realize these benefits in a full system (accelerator + DRAM), designers must ensure that the benefits of using the electrical, optical, analog, and digital domains exceed the costs of converting data between domains. Designers must also consider system-level energy costs such as data fetch from DRAM. Converting data and accessing DRAM can consume significant energy, so to evaluate and explore the photonic system space, there is a need for a tool that can model these full-system considerations. In this work, we show that similarities between Compute-in-Memory (CiM) and photonics let us use CiM system modeling tools to accurately model photonics systems. Bringing modeling tools to photonics enables evaluation of photonic research in a full-system context, rapid design space exploration, co-design, and comparison between systems. Using our open-source model, we show that cross-domain conversion and DRAM can consume a significant portion of photonic system energy. We then demonstrate optimizations that reduce conversions and DRAM accesses to improve photonic system energy efficiency by up to 3x.

Architecture-Level Modeling of Photonic Deep Neural Network Accelerators

TL;DR

This work shows that similarities between Compute-in-Memory (CiM) and photonics let us use CiM system modeling tools to accurately model photonics systems, and demonstrates optimizations that reduce conversions and DRAM accesses to improve photonic system energy efficiency.

Abstract

Photonics is a promising technology to accelerate Deep Neural Networks as it can use optical interconnects to reduce data movement energy and it enables low-energy, high-throughput optical-analog computations. To realize these benefits in a full system (accelerator + DRAM), designers must ensure that the benefits of using the electrical, optical, analog, and digital domains exceed the costs of converting data between domains. Designers must also consider system-level energy costs such as data fetch from DRAM. Converting data and accessing DRAM can consume significant energy, so to evaluate and explore the photonic system space, there is a need for a tool that can model these full-system considerations. In this work, we show that similarities between Compute-in-Memory (CiM) and photonics let us use CiM system modeling tools to accurately model photonics systems. Bringing modeling tools to photonics enables evaluation of photonic research in a full-system context, rapid design space exploration, co-design, and comparison between systems. Using our open-source model, we show that cross-domain conversion and DRAM can consume a significant portion of photonic system energy. We then demonstrate optimizations that reduce conversions and DRAM accesses to improve photonic system energy efficiency by up to 3x.
Paper Structure (7 sections, 5 figures)

This paper contains 7 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Photonics Modeling Tool
  3. Evaluation
  4. Accelerator Energy Breakdown
  5. Throughput
  6. Full-System (Accelerator+DRAM)
  7. Architecture Exploration

Figures (5)

  • Figure 1: Albireo architecture. As data traverse the $DE$, $AO$, and $AE$ domains, they leverage different movement and reuse opportunities but pay energy for data converters, notated $X/Y$ for conversion from domain $X$ to domain $Y$.
  • Figure 2: Energy breakdown validation.
  • Figure 3: Throughput for two DNN workloads. CiMLoop captures underutilization, which significantly degraded throughput for Albireo running AlexNet.
  • Figure 4: Memory exploration. Aggressively-scaled Albireo is dominated by DRAM energy. DRAM-energy-reducing operations such as batching and fusion are required to realize the benefits of aggressive scaling,
  • Figure 5: Increasing the amount of reuse in the analog and photonic domains can reduce data conversion energy, leading to a lower-energy system.