Several PhD Positions - Co-Design for In-hardware Learning

We are seeking highly motivated PhD candidates to join our research team and work at the intersection of brain-inspired computing, novel circuits and architectures, and machine learning. The Peter Grünberg Institute - Neuromorphic Compute Nodes (PGI-14) conducts research across several areas with the goal of closing the gap between the computational abilities of modern computers and those of the brain, especially in terms of energy efficiency. We develop novel computing capabilities by incorporating some of the remarkable features found in biological brains. We aim to lower the energy consumption of today’s hardware while expanding our understanding of the fundamentals of computation. Practical application areas include AI and Machine learning, intractable NP-hard problems, scientific computing, and signal processing., * 1) Associative memory based computing: This project will utilize concepts including Dense Associative Memories and Modern Hopfield Networks to both a) physically realize such systems in electronics, particularly using novel content-addressable memory (CAM) designs, including tape-out opportunities, and b) exploring computing applications and algorithms that can leverage associative memories, including machine learning and LLMs, genomics, and classical data science methods (tree models, decision trees, finite automata). We will test these applications experimentally in taped-out IC chips.

• 2) Kolmogorov-Arnold Networks (KAN) as a promising complement to conventional neural networks. KANs offer efficient representations for complex nonlinear models, with unique research opportunities in algorithm and hardware co-design, particularly for resource-constrained environments. Applications focus on edge intelligence, scientific computing, and emerging AI. The project tasks will include resource-aware training methods and hardware implementation, including the design, implementation, and verification of KAN accelerator architectures, as well as participation in tape-outs for custom integrated circuits.
• 3) In-hardware learning: More broadly than the above topics, our team seeks to build electronic hardware to support and implement local learning rules suitable for recurrent, nonlinear networks with internal state variables. The above projects provide suitable constrained architectures for this study, but the opportunity exists for more flexible explorations in both learning rule types and the physical realization. This project will involve algorithm-hardware co-design ultimately targeting physical implemetations in integrated circuits composed of both CMOS and emerging memory technologies.

An overarching theme of the projects is hardware-software co-design, where both circuits/ architecture and algorithms/applications are developed jointly in a team environment and with expert collaborators., * Research & Infrastructure: You will participate in a world-leading international research environment with state-of-the-art equipment

• Team & Environment: You will have the opportunity to work in the highly interdisciplinary and exciting field of neuromorphic and brain-inspired computing, collaborating with physicists, neuroscientists, materials scientists, and engineers
• Networking & Exchange: You will participate in (international) conferences and project meetings and actively build your scientific network
• Supervision & Support: We will accompany your doctoral studies with continuous, expert guidance from your academic supervisor
• Work-life balance: We offer flexible working hours to help you balance your professional and personal life. You also have the option of flexible working (in terms of location), which is generally possible after consultation and in line with upcoming tasks and (on-site) appointments

Depending on the specific projects above you seek to join, your profile should cover some of the skills listed below.

• Masters degree in electrical/electronic engineering, computer science, computer engineering, physics, or related field
• Good understanding of computer architecture and hardware accelerator design is beneficial
• For circuit implementations, strong electronics background with experience in design and simulation of analog, digital, or mixed-signal circuits, including SPICE and related tools is strongly desired
• For algorithms projects, a strong background in coding and advanced mathematics is strongly desired
• Programming experience in Python and familiarity with machine learning frameworks
• Knowledge of emerging memory devices and computing technologies is beneficial
• Excellent communication skills and the ability to work in a team are essential
• Very good command of written and spoken English with extensive vocabulary is required for the international working environment (at least B2 level according to the CEFR), ideally supported by a certificate confirming the language level.

• Fair remuneration: Depending on your qualifications and assigned responsibilities, you will be classified according to pay group 13 (75%) of the TVöD-Bund. In addition to the basic salary, there is an additional year-end bonus under the collective pay agreement amounting to 75% of a monthly salary, as well as capital-forming benefits. All information about the TVöD-Bund collective agreement can be found on the BMI website (pay scale table on page 74 of the PDF download)
• Perspective: The position is initially for a fixed term of 3 years but with the prospect of longer-term employment
• Support for international employees: Our International Advisory Service makes it easier for international employees to get started

In addition to exciting tasks and a collegial working environment, we offer you much more: https://go.fzj.de/benefits