A bio-inspired two-layer mixed-signal flexible programmable chip for early vision

Abstract

A bio-inspired model for an analog programmable array processor (APAP), based on studies on the vertebrate retina, has permitted the realization of complex programmable spatio-temporal dynamics in VLSI. This model mimics the way in which images are processed in the visual pathway, what renders a feasible alternative for the implementation of early vision tasks in standard technologies. A prototype chip has been designed and fabricated in 0.5 /spl mu/m CMOS. It renders a computing power per silicon area and power consumption that is amongst the highest reported for a single chip. The details of the bio-inspired network model, the analog building block design challenges and trade-offs and some functional tests results are presented in this paper.

On Analog Networks and Mixed-Domain Spatio-Temporal Frequency Response

“…3, enables the implementation of a general mixed-domain spatiotemporal rational transfer function in the form given below in (10), with a one to one correspondence of network connections and transfer function polynomial terms (10) with and the region of support [7] defined in 2-D space as (11) The quantities and are in general complex constants for , though we only consider real values in this presentation. The differential equation corresponding to (10) is given by (12) with parameters related to (10) (16) Comparing (10) and (16) reveals that the extra temporal derivative diffusion connections to linear CNNs enable a richer spatiotemporal characteristic to be realized.…”

“…From a filtering point of view, we have general support (recurring in both causal and anti-causal directions) discrete systems that are not necessarily recursible [9]. Despite nonrecursible structures not being suitable for direct implementation on delay based digital systems, analog VLSI arrays such as the CNNUM [10] are particularly suited for the implementation of general support dynamics. Preliminary results from our ongoing research on analog VLSI implementations of the network proposed here are reported elsewhere [11].…”

“…Looking at (10) again, at low values of close to zero, terms without dominate. The "spatial response slices" are approximated by (10) with all terms that have the factor removed. As we increase , the response will be a mix of terms with powers of as their factor.…”

“…Consider a general term in (10), as part of the numerator or denominator polynomial (18) where the coefficient is a real constant. The frequency response of (18) is given by (19) note that we only consider real coefficients in (10). Let (20) and (21) Both and incorporate the factor .…”

A comparison of CNN and LEGION networks

“…CNN has been applied to a number of local visual processing tasks, such as edge detection, corner detection, image translation from one position to another, etc., by taking advantage of single cell computation and local activity propagation due to the use of a feedback template [4] [8]. Though such tasks have not been explicitly addressed in the LEGION framework, it should be straightforward to apply LEGION to them.…”