Fault-tolerant
Interconnect Technologies for SoCs (2D,
3D, Si-Photonics, Hybrid)
Complex signal
processing systems-on-chip contain dozens
of components made of processor cores,
DSPs, memory, accelerators, and I/Os, all
integrated into a single die area of just
a few square millimeters. Such complex
systems need a novel on-chip interconnect
closer to a sophisticated network than
current bus-based solutions. This network
must provide high throughput and low
latency while keeping area and power
consumption low. We research
advanced interconnect technology for
embedded multicore SoCs targeting both FPGA
and ASIC platforms. In
particular, we investigate 3D-TVS
integration, fault tolerance methods,
photonic communication protocols,
low-power mapping techniques, and
low-latency adaptive routing.
Ultra Low-power
Neuromorphic Systems and AI-Accelerators
Neuromorphic
computing uses spiking neuron network
models to solve machine learning problems
in a more power/energy-efficient way when
compared to the conventional artificial
neural networks.We research
adaptive low-power spiking neuromorphic
systems and SoCs empowered with our
earlier developed fault-tolerant
three-dimensional on-chip interconnect
technology. In particular, we
investigate adaptive
configuration methods to enable the
reconfiguration of different network
parameters (spike weights, routing,
hidden layers, topology, etc.),
fault-tolerant and thermal-aware mapping
methods, and on-line learning
algorithms.
Our AI and
neuromorphic AI technology is used in
various real-world applications,
including anthropomorphic
robotics, embedded
medical, and distributed
energy
harvesting
Anthropomorphic Robotics
Restoring grasping
and movement for people with amputations
and neurological
impairment is imperative for
retrieving independence of amputees.
Prosthetic limbs, which are
becoming widespread therapeutic
solutions, can significantly
restore grasping and improve the quality
of life of people with amputations or
neurological disabilities. However, unlike
living agents that combine different
sensory inputs to perform a complex task
accurately, most prostheses use
uni-sensory input, offer limited degrees
of freedom and need long patient training.
Sensors enabling environmental perception
and efficient control algorithms are
needed since they strengthen the system's
reliability while reducing the cognitive
burden for the amputees. We study limb
prostheses and anthropomorphic robotics
based on non-invasive innovative neural
interfaces and advanced
low-power neuromorphic SoC control for
real-time communication and
processing. Currently,
we develop adaptive neuromorphic
multi-degree-of-freedom prosthesis limbs
with tactile feedback to restore grasping
and sensation for persons with amputation
or neurological impairments. We use
non-invasive technologies directly
interfacing the environment with the
residual arm or legs.
|