Circuits mixtes et microsystemes implantables dedies a l'enregistrement sans fil des biopotentiels neuronaux

Recordings of extracellular neural biopotentials from several neurons in the cortex have motivated a great deal of research towards the development of new prostheses for treating various neural disorders including motor impairments, epilepsy, and paralysis. Implementing a suitable interface to the cortex necessitates chronic use of a high resolution neural recording implant that can sample the simultaneous activity of hundreds of neurons. Neural interfacing circuits featuring high-channel counts, large data rates handling capabilities, small size, and ultra-low-power consumption are required. This work presents new dedicated mixed-signal circuits and low-power system architectures to accommodate high-density neural recording implants with a tremendous amount of channels. The main contributions presented in this work are reported in five papers published or submitted in journals of the IEEE and Springer.;Then, we propose a 16-channel mixed-signal integrated neural interfaces (INI) based on a new efficient parallelized architecture. This microsystem features a vertical integration approach to mitigate size and complexity. It is composed of two CMOS 0.18mum stacked chips mounted on the back of a stainless-steel microelectrode array. This multichip device features a data reduction mechanism employing an absolute value-detector and implements serial communications towards a host controller on a four-wire bus. The implemented data reduction strategy provides complete data integrity and achieves a maximum reduction factor of 48. Its specific architecture allows great flexibility and scalability to higher channel counts. Also, the proposed architecture is based on a new front-end including the low-noise amplifier mentioned above, a filter, sampling and digitization, which is very compact, flexible and scalable to very high channel counts.;We also present an automatic biopotential extractor for performing data reduction, whereas capturing complete neural waveforms, in high resolution INI. The implemented integrated circuit retrieves neuronal information with full integrity. It uses an analog signal processor based on the Teager energy operator (TEO), a 9th-order linear-phase delay filter and a dynamic comparator with latch. This detector achieves sub-microwatt power consumption using low-power OTA-C building blocks whose transistors are operated in weak inversion. It is optimized using a transconductance-efficiency design methodology.;A novel low-noise integrated neural amplifier achieving superior efficiency is first presented. This bioamplifier proposes the smallest reported size (0.05 mm2) and achieves one of the best noise efficiency factor (NEF). Its input referred noise is of 5.6 muVrms for power consumption below 10 muW. It employs a novel active DC rejection feedback loop implemented with a Miller integrator and a MOS-bipolar element which cancels the systematic offset induced at the electrode-tissue interface. The amplifier implemented in a CMOS 0.18-mum process provides a gain of 50 dB and exhibits a bandwidth ranging from 100 Hz to 9 kHz. In vivo recordings performed this amplifier and a stainless steel microelectrode demonstrates its excellent performance.