Multi Channel Signal Processing Integrated Neural Interfaces

THIS thesis presents two 0.35mum CMOS prototypes of multichannel integrated neural interfaces for distributed recording of neural activity in the brain. Each integrated neural interface contains 256 continuous-time recording channels, each comprised of a two-stage programmable gain amplifier and a bandpass filter with tunable cut-off frequencies. A parallel VLSI architecture with in-channel analog memory allows for truly simultaneous signal sampling. In-channel and peripheral switched capacitor circuits perform on-sensory-plane computationally expensive spatio-temporal signal processing in real time for applications such as data compression and pattern recognition. The circuits are optimized for low noise, low power and high density of integration as necessary for implantation. Golden and platinum 3-D electrode arrays are fabricated directly on the die surface for in vitro and in vivo experiments. An implantable microsystem comprised of an integrated neural interface prototype interfaced with a low-power high-throughput VLSI signal processor is validated in off-line epileptic seizure prediction.