
Spiking Neural Network Hardware
Energy-efficient CMOS neurons, spike-based learning and AER interconnects.
Read more →Home / Neuromorphic Systems
Bio-inspired, event-driven chips for energy-efficient machine learning.
Our group has made progress towards realizing Bio-inspired Neuromorphic and Cognitive Systems-on-a-Chip (SoCs) that will introduce a new paradigm for massively-parallel and event-driven computing platforms to transcend the limitations of traditional computing architecture in nano-scale CMOS technologies. We are building chips that perform pattern recognition with complexity an order of magnitude higher than traditional computing and signal processing architectures, and at power levels orders of magnitude lower than in traditional von Neumann computers.
In 2015, U. S. Office of Science and Technology (OSTP) announced a grand challenge that addresses three priorities- the National Nanotechnology Initiative, the National Strategic Computing Initiative (NSCI), and the BRAIN initiative to Create a new type of computer that can proactively interpret and learn from data, solve unfamiliar problems using what it has learned, and operate with the energy efficiency of the human brain. In alignment with this goal, our research goal is to dramatically improve computing system performance by orders of magnitude in capability and energy-efficiency, by breaking the processor-memory throughput bottleneck in von Neumann computers.
To realize neural-inspired computing, we are investigating circuit architectures for adaptive Machine Learning algorithms developed by the computational Neuroscience community. This is achieved by a large organization of mixed-signal circuit motifs comprised of event driven and energy efficient silicon neurons and a dense fabric of compact synapses integrated on CMOS chips. Our present research focus in this area include the following synergistic efforts:

Energy-efficient CMOS neurons, spike-based learning and AER interconnects.
Read more →
CMOS-NVM crossbars and memristive synapses that compute where the data is stored.
Read more →While being successfully demonstrated, the proposed cognitive computing architectures will realize a massively parallel, highly scalable and energy-efficient event-driven intelligent signal processing (ISP) platform to tackle complex real-time machine learning tasks such as pattern recognition, robotic control, visual pattern and motion detection and brain-machine interfaces with power consumption comparable to biological brains.
The complete, up-to-date list of publications is on Google Scholar.