‣ Nanoelectronic devices for neuroinspired/neuromorphic computing
‣Algorithm/device codesign for artificial neural nets
‣Transparent neural interfaces
‣Learning and memory
Illustration of our research by Peter Durand
Transparent Neural Interfaces
Calcium imaging has been used to monitor neural populations with single-cell resolution in the brain. Electrophysiological recordings provide high temporal, but limited spatial resolution, because of the geometrical inaccessibility of the brain. We develop transparent and flexible neural interfaces enabling recording of neural activity with very high resolution from populations and neurons.
•. M. Thunemann, Y. Lu, X. Liu et al, and A. Devor, D. Kuzum, Nature Communications, 9(1), 2035, 2018.
•. Y. Lu et al and, D. Kuzum. Advanced Functional Materials, 1800002, 2018.
• X. Liu et al and D. Kuzum, Frontiers in Neuroscience, 12, 32, 2018
•. Y. Lu, X. Liu, D. Kuzum, Current Opinion in Biomedical Engineering, 6, p. 138-147, 2018.
• E. Iseri and D. Kuzum, Journal of Neural Engineering,14.3, 031001, 2017.
Minimally Invasive Medical Implants
Biophysical Modeling of Learning and Memory
We develop graphene and shape memory based implants that can be translated into medical practice to establish a robust and resilient electrical interface with the human brain. We demonstrated that our technologies allow high resolution detection of cortical waves during epilepsy and can enable deep cortical stimulation from the surface.
• R. Zhao et al and D. Kuzum, IEEE IEDM, 2018.
• X. Liu, Y. Lu, and D. Kuzum, Scientific Reports, 8(1), 17089, 2018.
• Y. Lu et al, and D. Kuzum, Scientific Reports, doi:10.1038/srep33526, 2016.
Learning and memory are crucial for high-level cognitive behavior. The hippocampus plays important roles in memory formation and retrieval through sharp-wave-ripples. It has been hypothesized that certain neuron populations in the cortex exhibit coordinated activity with hippocampus during learning and memory retrieval. WE develop biophysical models to investigate mechanisms of hippocampal-cortical memory transfer.
• Xin Liu, D. Kuzum, Frontiers in Neuroscience, 13, 67, 2019.
• X. Liu et al and D. Kuzum, PLoS ONE 12(9): e0184542, 2017.
• H. Makino et al, D. Kuzum, and T. Komiyama, Neuron, 94(4), 880-890, 2017
Nanoelectronic Synaptic Devices for Neuroinspired Computing
As compared to biological systems, today’s programmable computers are 6 to 9 orders of magnitude less efficient. The superior features of the brain, lacking in today’s computational systems, are ultra-high density, low energy consumption, parallelism, robustness and plasticity. We develop programmable nanodevices for neuroinspired architectures to address inherent limitations of conventional computers.
• Y. Shi et al and D. Kuzum, Nature Communications, 9(1), 5312, 2018.
• S. Seo et al, D. Kuzum, H.-S. P. Wong, J.-H. Park L, Nature Communications, 9(1), 5106, 2018.
• S. Oh et al and D. Kuzum, IEEE Electron Device Letters, 39(11), 1768, 2018.
Algorithm-Device Codesign for Efficient Neural Networks
We investigate algorithm-device codesign approaches to develop efficient hardware for neural network models. Our approach harness nonlinear behavior of nanodevices to implement complex neural network operations with high efficiency while providing high accuracy and throughput.
• Y. Shi et al, and D. Kuzum, IEEE Transactions on Electron Devices, 2019
• S. Oh et al, and D. Kuzum, IEEE Electron Device Letters, 2019.
• Y. Shi et al, and D. Kuzum, Frontiers in Neuroscience,13, 405. 2019.