Towards Analog VLSIArrays forNonseparable 3D Spatiotemporal Filtering
Emmanuel Mic. Drakakis · 2006
We report thedevelopment ofa newCNN based theoretical frameworkand itson-goinganalogVLSI implementationfor continuous-time discrete-space spatiotemporal filtering. Ourproposed formulation, whichstems fromthetheory oflinear cellular neural networks, iscapable of realizing awiderangeoflinear spatiotemporal dynamics. Based onourproposed framework, 3D nonseparable spatiotemporal filters aresynthesized. Thefilters arepresented intheformof continuous-time/discrete-space mixeddomaintransfer functions withresponses qualitatively resembling visual cortical bandpass spatiotemporal receptive fields. AnalogVLSIbuilding blocks suitable fortheimplementation oftherequired spatiotemporal dynamics arepresented. Simulation results onAMS 0.35,um process parameters ofa 5x5array, withelectrical parameters tunedtorealize thespatiotemporal dynamics ofoneofour synthesized 3Dbandpass filters, arepresented.