PhD Photonic Integrated Switches for High Performance AI Compute Clusters

PhD Photonic Integrated Switches for High Performance AI Compute Clusters

Published Deadline Location
3 Jan 15 Feb Eindhoven

You cannot apply for this job anymore (deadline was 15 Feb 2024).

Browse the current job offers or choose an item in the top navigation above.

Job description

The smart optical networks lab (SONL) at the Electro-Optics Communication (ECO) group at Eindhoven University of Technology is recruiting for a PhD position to research novel nanosecond O-band photonic integrated switches for low latency and highly scalable AI compute clusters.

The electro-optical communications (ECO) group in the Faculty of Electrical Engineering at TU/e is a globally recognised, leading scientific and applied research group focused on exploiting light for communication and quantum systems. We apply our knowledge in collaboration with other scientists at TU/e and more recently within the newly formed Eindhoven Hendrik Casimir Institute (EHCI) to develop the required solution for many of the relevant challenges in communication and sensing systems. The group expertise spans from the fundamentals and physics of photonics, optics, the design and fabrication of photonic integrated circuits (PICs) systems to exploiting optical linear/non-linear signal processing to unlock fiber capacity and relevant higher layer protocols required to operate modern optical communication networks. Based in the purposely built FLUX building at the TU/e Campus, the ECO group has access to 300m2 of labs for conducting experimental research and is supported by a state-of-the-art 800m2 cleanroom. With 11 tenured scientists and as many as 70 PhDs, postDocs and senior researchers, the ECO group is a vibrant and exciting research group perfectly suited for talented and ambitious scientists. The group is active in spin outs and starts-ups (e.g. Astrape Networks, Microalign, PhotonX Networks) and carries out bilateral industrial research with major stakeholders in the communications industry.

The PhD project targets the design, prototyping, and demonstration of novel nanosecond O-band photonic WDM switches and the fast control to enable novel low latency highly scalable and flat interconnect AI compute clusters. Machine learning clusters and artificial intelligence (AI) training have become increasingly popular in recent years. The recent introduction of OpenAI's ChatGPT made large-scale models available to the public, which enables the integration of AI in everyday objects and tasks. As a result, the model scale is doubling yearly, causing a proportional increase in the overall model size. These developments require a highly efficient and scalable network to interconnect larger and larger amount of compute processors. Today's AI compute processors have O-band (1310 nm) multi-Terabit/s interfaces to share intermediate data for distributed training and processing, generating large traffic flows.  Low and deterministic latency will be required for specific application in data centre for AI training compute clusters. The interconnect network based on electrical packet switches will not be able to support these requirements due to power consumption of the electrical interconnects that will require complex and inefficient cooling solution, and the footprint of the electrical interconnections that will exceed the available space in packet switch chip assembly.

Transparent optical switches allow the network to be upgraded in terms of data rate, modulation format and wavelength grid without having to replace the network electronic switches. Furthermore, the lack of an O/E/O conversion reduces power consumption compared to the electronic counterparts and reduces the amount of costly optical transceivers. Moreover, photonic integration allows for many optical functions on a small footprint, combined with the potential of economic mass-manufacturing. Therefore, the implementation of fast and lossless photonic switches is crucial to enable the demonstration of optically switched network for AI training computing clusters with low power consumption and ultra-low and deterministic latency. Most of the photonic integrated switches operate in the C-band (1550 nm) for telecom, only a few O-band switches have been so far investigated.

The focus of the PhD activity is on demonstrating disruptive new network architectures for AI compute clusters utilizing optical switching to increase efficiency and reduce latency and power consumption.  Therefore, the PhD candidate will investigate and design novel O-band photonic integrated WDM switch architectures to implement and demonstrated a low latency optically switched networks for AI training compute clusters.

Responsibilities and tasks

The purpose of the project is the design and development of nanoseconds O-band photonic WDM switches and the fast control to enable novel low latency highly scalable and flat interconnect AI compute clusters. To achieve these objectives, the project is organized into the following main activities:
  1. Investigate disruptive new network architectures for AI compute clusters utilising optical switching and numerically assess the performance via software simulation tools.
  2. Define the system specifications which will be then translated into requirements for the design and implementation of the O-band photonic WDM switch architecture and the selection of the hybrid photonic technologies to fabricate the switches. 
  3. Design, simulate, fabricate and test the O-band photonic integrated switches exploiting diverse hybrid photonic technology platforms.
  4. Packaging and assessing the O-band photonic integrated switches in collaboration with the industrial partners.
  5. Develop the optoelectronic interfaces and controls to enable remote reconfiguration operation.
  6. Develop and evaluate fast and efficient scheduling algorithms for fast control and reconfiguration of the optical AI compute clusters.
  7. Realize a small-scale compute cluster lab testbed to demonstrate and evaluate the performance of the innovative low latency and high capacity optical switched AI computer network architecture empowered by the developed O-band photonic integrated WDM switches and controls.
The PhD candidates will contribute to the TU/e efforts in establishing collaboration with the other researchers in the project and contribute to the related project reporting, scientific publication and dissemination activities.

Specifications

Eindhoven University of Technology (TU/e)

Requirements

Candidates for this challenging projects must have:
  • Master of Science degree in Electrical Engineering or in Applied Physics.
  • Solid knowledge of optical fiber communications techniques and/or for photonic devices and circuits.
  • Knowledge of photonic devices and circuits / integration technologies and software simulation tools for optical fiber communications.
  • Knowledge of one or two programming languages e.g. OPNET, OMNET, and VHDL, Verilog HDL FPGA programming are appreciated.
  • He/she must be familiar with optical and electronic laboratory measurement equipment, optical fiber system and components, semiconductor devices, and electronic circuits.
  • Previous cleanroom experience will be considered an advantage but is not required.
  • Proven team-working capabilities and communication skills
  • English proficiency (both verbal and written)
  • Willingness to start as soon as possible

Conditions of employment

A meaningful job in a dynamic and ambitious university, in an interdisciplinary setting and within an international network. You will work on a beautiful, green campus within walking distance of the central train station. In addition, we offer you:
  • Full-time employment for four years, with an intermediate evaluation (go/no-go) after nine months. You will spend 10% of your employment on teaching tasks.
  • Salary and benefits (such as a pension scheme, paid pregnancy and maternity leave, partially paid parental leave) in accordance with the Collective Labour Agreement for Dutch Universities, scale P (min. €2,770 max. €3,539).
  • A year-end bonus of 8.3% and annual vacation pay of 8%.
  • High-quality training programs and other support to grow into a self-aware, autonomous scientific researcher. At TU/e we challenge you to take charge of your own learning process.
  • An excellent technical infrastructure, on-campus children's day care and sports facilities.
  • An allowance for commuting, working from home and internet costs.
  • A Staff Immigration Team and a tax compensation scheme (the 30% facility) for international candidates.

Specifications

  • PhD
  • Engineering
  • max. 38 hours per week
  • University graduate
  • V36.7168

Employer

Eindhoven University of Technology (TU/e)

Learn more about this employer

Location

De Rondom 70, 5612 AP, Eindhoven

View on Google Maps

Interesting for you