In recent years, Artificial Intelligence (AI) has made impressive progress and many applications are entering the daily lives of everyone. Expectations are that this is merely the beginning of a whole new era in which exciting new applications will arise in fields as for instance medicine, self-driving cars, drug design, robotics, arts, etc...
To accelerate this deployment, electronic engineers are investigating hardware implementations that can efficiently execute AI algorithms within low power constraints. One of the options to achieve this makes use of emerging resistive Random-Access Memory (RRAM) technology.
Most research focuses, however, on designing the RRAM properties such that they can meet the specifications needed for implementing existing AI algorithms. In this context, the use of RRAM as a multi-level data memory element for weight storage in deep learning algorithms is the most frequently encountered scenario. Some RRAM concepts show, however, intrinsic properties that can be exploited in very different ways to build learning systems.
The purpose of this PhD is to explore the opportunities offered by novel device technologies and to build a hardware implementation of artificial intelligence beyond conventional mainstream algorithm implementations.
This PhD contains three aspects:
- The student needs to study the device physics of novel technologies like for example resistive RAM, magnetic RAM, Ferroelectric-based device, etc... The student needs to identify what physical properties can be exploited for building AI systems. This can involve unconventional use of the devices for exploiting effects that are considered 'undesired' for normal use, like for example stochastic behavior of excessive variability. Several device options are or will be fabricated at the imec facility allowing for experimental investigation.
- The student will propose algorithmic concepts for exploiting bottom-up inherent learning properties of novel technology. This requires significant creativity and outside-of-the-box thinking. The student can use or adapt parts of existing machine learning algorithms for this purpose. We are particularly targeting a human-like concept-learning algorithm for building true artificial intelligence that goes well beyond mainstream machine learning.
- The student will at all stages of the research evaluated the practical feasibility of his or her ideas at the level of design and system level. The purpose is to arrive at a practically implementable concept on silicon.
Required background: master in science, master in engineering, additional degree or strong interest in cognitive sciences is a plus
Type of work: 40% modeling, 40%experimental, 10% literature
Supervisor: Rudy Lauwereins
Daily advisor: Robin Degraeve
The reference code for this position is 2020-057. Mention this reference code on your application form.
Chinese nationals who wish to apply for the CSC scholarship, should use the following code when applying for this topic: CSC2020-24.