About

Advanced semiconductor devices are the backbone of modern society. In CMOS-and-Beyond group, we accelerate innovation for devices and systems that are energy efficient, versatile and integrable with advanced node CMOS devices and benchmark their performance for unconventional computing like Neuromorphic and Quantum Computing. To learn more about our research activities, please visit other sections of the webpage, follow us on LinkedIn and reach out to any of the contact persons.

Our LinkedIn Page

Background

Brain — Human brain provides a benchmark for designing electronic devices with ultra-low power consumption that can do computation and learning based on small amount of data that today’s AI algorithms cannot do. Image Created by: Microsoft Copilot.

Artificial intelligence (AI) has already achieved the capabilities to match, or even exceed, human
intelligence in solving specific classification tasks. However, running complex AI models on classical von Neumann computing architectures, i.e., modern computer chips, require vast amounts of energy and cause long latency due to constant data transfer between physically separated memory and the processing units. Achieving brain-like efficiency in computing requires nanoscale, energy-efficient, non-volatile and fast analog memory devices capable of reliably emulating the ultra-dense, low-power, low-latency operation of biological synapses and neurons.

Goal

Dense crossbar array of ferroelectric memories fabricated in the CMOS and Beyond team together with the device ferroelectric properties. Results to be published soon (manuscript under preparation).

Our main goal is to strengthen semiconductor device innovation for unconventional computing. This involves research on advanced non-volatile memories using ferroelectric devices, logic components using oxide and 2D semiconductors and optoelectronic components for photonic neuromorphic computing. One of our key focus is understanding defect dynamics in high-k oxides, semiconductors and their interfaces to improve their performance reliability and lifetime.

Impact

snn — ECG data classification using a spiking neural network (SNN) that can accelerate arrythmia classification at a fraction of energy and cost in comparison to today’s systems. Work in collaboration with NCKU, Taiwan. Manuscript under preparation.

Aim of our research is to advance unconventional computing hardware technology by combining
knowledge of device physics, nanotechnology, experimental materials science and simulations. We use AI-based electronic materials development to develop hardware in a more resource efficient way. This approach has the potential to drive the development of more efficient, scalable, and adaptable devices, with applications in memory storage, neuromorphic computing, cryogenic memory for quantum computation, high performance computing, space technologies and energy-efficient logic circuits, ultimately paving the way for broader adoption in the semiconductor industry.

One significant impact of this technology is highly energy efficient smart edge devices, capable of real-time decision that can find applications in healthcare, autonomous systems, space, robotics and many other fields.