Student project: Algorithm-hardware co-optimization for Transformer Neural Networks on the edge

This project aims to study the Transformer neural network from the hardware perspective and the applicability of co-optimizing the network with an event-driven multi-core accelerator for efficient edge processing.

What you will do

Transformer is an attention-based framework for deep learning that achieved remarkable performance on a wide range of tasks. Recently, it has become the backbone of state-of-the-art algorithms in natural language processing and computer vision, and the major technology advance behind large language models like ChatGPT. Bringing Transformer to the edge is challenging due to its high computational complexity. Its self-attention mechanism departs from the available AI accelerator architectures that mainly focused on the convolutional neural networks in recent years, creating a need to co-design novel hardware for the framework. Therefore, we seek to study the Transformer from the hardware perspective and the applicability of co-optimizing the network with a multi-core accelerator for efficient edge processing.

Tasks:

• Study the Transformer framework and targeted edge hardware.
• Analyze and locate the computation challenges of Transformer for efficient edge processing.
• Design space exploration for the algorithm-HW co-design of the target application.
• Study the hardware solutions for the computation challenges.
• Develop a set of scientific recommendations for algorithm-hardware co-optimization for Transformer.

The project can be divided into three phases: 1) feasibility studies on deploying Transformer on resource-constrained edge hardware, 2) strategy for algorithm-hardware co-optimization for Transformer, and 3) hardware implementation. The student is expected at least to demonstrate the results for phase one and two of the project.

What we do for you

Imec is one of the world's leading research institutes in micro and nano-electronics. The imec-NL lab at Holst Centre is a center of excellence in designing nano-electronics solutions for the Internet of Things and healthcare applications. In this internship project, you will be working on a cutting-edge research project under the supervision of expert researchers from diverse backgrounds. The outputs from the project may be published in high-impact journals/conferences (subject to the quality of the work). Imec-NL provides the required equipment, access to lab facilities, a workplace in the Holst Centre at High Tech Campus, and a monthly internship allowance during the internship.

Who you are

• M.Sc./Ph.D. students with a relevant background (non-European students are only eligible if they study in the Netherlands).
• Available for 9 months, preferably 12 months.
• Have good programming skills in Python.
• Have excellent knowledge of neural networks and computer architecture.
• Knowledge of programming in deep learning frameworks (e.g., TensorFlow, PyTorch) is a big plus.
• Are in good command of spoken and written English.
• Motivated student, good communicator, easy collaborator, and eager to work independently and expand knowledge in the field.

Interested

Does this position sound like an interesting next step in your career at imec? Don’t hesitate to submit your application by clicking on ‘APPLY NOW’.

If you wish to apply, then please submit your full resume and a cover letter.