Welcome to MIT 6.S918: Optical Computing in the Era of AI

Overview

We live in an age of big machine learning models, where modern deep neural networks comprise hundreds of billions of parameters. As these models continue to scale, the ever-growing requirements on energy efficiency and computation speed have sparked a new industry in designing specialized computing hardware optimized for neural networks.
In this constantly evolving landscape of technology, light-based computing, commonly referred to as "optical" or "photonic" computing, is a revolutionary paradigm shift promising higher computing frequency and less energy consumption than traditional digital computing. This course aims to introduce students to this exciting and rapidly growing field, focusing particularly on:

1. How can light be used for computing, and why should we build optical computing hardware?

2. What are the fundamental devices used for photonic computing?

3. How do these photonic devices integrate into modern computer systems for real-world computing workloads?

This course will integrate lectures, lab tours, demos, and a final team presentation on new research areas in photonic computing.

Course Information

Lecture hours: Wednesday / Friday, 2:30 – 4:00 pm
Location: 37-212

Office hours: Wednesday / Friday 4:00 – 5:00 pm
Location: 37-212

Background required

Both undergraduate and graduate students are welcome. We aim to make this course accessible to a broad range of backgrounds. However, students should have taken at least one electromagnetism course at the advanced undergraduate level and one course in computer architecture and systems (e.g., combinational and pipelined arithmetic-logic units (ALU), in-order pipelined microarchitecture, etc.). Some examples of MIT courses that would be adequate background (old course numbers in brackets) are:
Electromagnetism:
  • 6.2300 [6.013] Electromagnetic Waves and Applications
  • 6.S046/6.S976 Silicon Photonics
  • 6.6330/6.6331 [6.602/6.621] Fundamentals of Photonics
  • 6.6300 [6.630] Electromagnetic Waves

    • Computer Architecture:
  • 6.1910 [6.004] Computation Structures
  • 6.5900 [6.823] Computer System Architecture

    • Students should also be familiar with the basic principles underlying deep neural networks. Courses that cover this include 6.3900 [6.036] at the undergraduate level and 6.7900 [6.867] at the graduate level.
    Other courses may also be sufficient background. If you have questions about the prerequisites, please feel free to email the instructors.

    Course Staff

    instructors

    Advisors

    Guest Speakers

    Course Schedule (Tentative)

    Week 1

    • Jan. 10 (Wednesday): Lec 1: Introduction to Optical AI
    • Jan. 12 (Friday): Lec 2: Silicon photonic devices principles and hardware

    Week 2

    • Jan. 17 (Wednesday): Lec 3: Fundamentals of optical computing primitives
    • Jan. 19 (Friday): Lec 4: Benefits and constraints of photonics compared with electronics computing (guest lecture)

    Week 3

    • Jan. 24 (Wednesday): Lec 5: Building practical computer systems using optical computing primitives
    • Jan. 26 (Friday): Lec 6: Software stacks and operating systems for emerging optical computers

    Week 4

    • Jan. 29 (Monday): Lec 7: Recent advancements of digital hardware accelerators for AI (guest lecture)
    • Jan. 31 (Wednesday): Group project presentations