Shih-Chii Liu Shih-Chii Liu Book: Analog VLSI: Circuits and Principles
by Liu, Kramer, Indiveri, Delbrück, and Douglas. Publisher: MIT Press. ISBN: 0-262-12255-3.
PROJECTS (suitable for Semesterarbeit or Diplomaarbeit)
The objectives of the neuromorphic group at INI are to explore computational principles of natural systems by developing implementations of insect and vertebrate sensory systems models; to develop sensori-motor control for robotics using aVLSI preprocessors, to develop technology for long-term adaptation and learning in spiking neuronal hardware, to understand computation in hybrid systems that compute with analog signals and spikes, and to develop proto-products. The senior members of the group include Tobi Delbrück and Giacomo Indiveri .
My research interests are in developing computational models of sensory and cortical processing including spike-based models suitable for implementation in hybrid silicon analog-digital systems.
INFRASTRUCTURE FOR A RECONFIGURABLE ASYNCHRONOUS EVENT-BASED MULTI-CHIP PROCESSOR
An example system consisting of a silicon AER retina and a cortex chip with integrate-and-fire neurons together with various types of synapses (including short-term depressing and facilitating synapses) have been used to describe the putative processing of visual cortical cells. This research was published in the 29th March 2001 issue of Nature. A preliminary report of the results from this cortical-like chip was presented at the ZNZ symposium in Zurich, 2001. We will use these networks for exploring brain-like architectures for computation.
SPIKING DENDRITIC NEURON MODEL
Together with YingXue Wang, we are looking at active dendritic circuits and studying the computational role of nonlinearities in a neuronal model. We implement these nonlinearities in AER VLSI multi-neuron circuits.
SPIKE-BASED SILICON COCHLEAS / MULTI-SENSORY FUSION
We are developing silicon cochleas with spike-based AER outputs in colloboration with Andre van Schaik at the University of Sydney in Australia. Example movies from the 32 channel binaural cochlea is a rolling spectrogram of spikes in response to a song, and the same output plotted in a 3D plot of channel, cochlea, and time. One of the goals of the sensory fusion project is to look at models where spike information from multiple modalities are fused or combined for different tasks. In particular, we fuse the cochlea spikes together with the spikes from the Dynamic Vision Sensor constructed by Lichtsteiner and Delbrück . This work is partially funded by an ONR NICOP grant.
SPIKE-BASED AUDITORY PROCESSING
Spike timing information from the cochlea can carry information useful for various auditory tasks. Applications of the silicon cochlea include auditory source localization and a particular sound recognition task, harmonic vs disharmonic sound recognition.
COMPUTATIONAL MODELING OF SENSORY FUSION
Together with Prashanth D'Souza and Richard Hahnloser, we are working on a computational model of visual and auditory spatial map alignment in the inferior colliculus of the barnowl using STDP learning. This work is partially funded by the Zurich Neuroscience Center.
We are developing neuromorphic motion chips which model the motion processing in the fly visual system. We have used the outputs of these chips in guiding the motion of a Koala robot. This platform can be used for developing insect-like sensori-motor models (e.g. optomotor response).
We are also collaborating with the Laboratory of Intelligent Systems at EPFL (directed by Dario Floreano) in developing new motion chips for use on a microflyer. The students and staff on this project include Jean-Christophe Zufferey, Antoine Beyeler, and Adam Klaptocz at LIS; and Rico Moeckel at INI. This work is sponsored by the Swiss National Foundation.
HOMEOSTASIS IN SPIKING NEURONS USING NONVOLATILE SILICON TECHNOLOGY
Together with Bradley A. Minch at Olin University, we are looking at ways of incorporating homeostatic, adaptation, and learning mechanisms in integrate-and-fire neurons using floating-gate technology.
CAVIAR - AN ASYNCHRONOUS VISION MULTI-CHIP EVENT-BASED SYSTEM
CAVIAR is an EU-funded 5th FET framework project which centers around the assembly of a spike-based asynchronous vision system. The partners in the projects are Bernabe Linares-Barranco at the National Microelectronics Center (IMSE) in Sevilla, Spain, Phillip Haefliger, Dr. Tor Sverre Lande at University of Oslo in Norway (UIO), and Anton Civit and Alejandro Linares-Barranco at University of Sevilla.
The components of CAVIAR-INI comprise a spike-based 64x64 silicon retina (INI), a convolution system (IMSE), a dimension-reduction chip consisting of a network of 32 x 32 spiking neurons with different synaptic dynamics (INI), and a network of 32 neurons with spike-based learning synapses (UIO). The spike outputs between modules are transmitted using the AER protocol and the mapping of events between modules is performed using a USB-AER board which was developed at USE. This board is based on a PCI-AER board developed at the NIH group in ROME.
EXAMPLE OF A NEUROMORPHIC RETINA PIXEL
Here is an example of a neuromorphic circuit with local gain control. The circuit is a modified version of Delbrück's receptor which models the adaptation process that results in local gain control. In Delbrück's receptor, the adaptation time constant is predefined at the design phase. The adaptation time constant of this receptor is controllable via a bias voltage.
Email: shihatini.phys.ethz.ch