Embedded Files
The Sensors Group develops neuromorphic sensors, network models, and algorithms that exploit the organizing principles of neural computation underlying biological sensors and brain processing. Much of our research is within the field of neuromorphic engineering, a field that is inspired by the way our brain computes.
Research areas
Research areas
dynamic vision sensor silicon retina event cameras (basic US Patent 77286269)
event-based silicon cochlea audio sensors and audio TinyML designs
learning to control and hardware inference of dynamical models for MPC
conversion methods from Analog Neural Networks to Spiking Neural Networks
embedded hardware for audio and visual edge devices
Our neuromorphic DVS, DAVIS, and DAS event sensors emulate the operating principles of the biological retina and cochlea. The retina pixels or cochlea channels output activity-driven events that encode features useful for downstream tasks. This asynchronous detection process leads to a compressed, sparse output with low latency and high temporal resolution. Furthermore the retina sensor "event cameras" provide high dynamic range of illumination.
Our neuromorphic digital AI deep neural network accelerators also exploit key notions of sparse brain computing to save energy and time. Our CNN accelerators NullHop and its descendants exploit activation sparsity and our multiple generations of RNN accelerators DeltaRNN, EdgeDRNN, and SPARTUS exploit temporal sparsity and weight sparsity.
We use our sensors and neural hardware to build micropower IoT audio keyword recognition devices and to do nonlinear optimal control of robots.
Page updated
Google Sites
Report abuse