Performance characteristics of active silicon microelectrode arrays.

Abstract

Extracellular neural recording is an important technique for advancing understanding of brain tissue organization, for enhanced diagnosis and treatment of CNS-related disorders, and for the design of neural prosthetic devices. This thesis describes continued development of active, micromachined silicon thin-film electrode arrays for chronic multiple single-unit CNS recording and investigates the recording capabilities of such structures. These electrode arrays utilize a heavily doped (p$\sp{++}),$ 15$\mu$m-thick silicon substrate which supports 32 electrodes. They have been integrated with 3 $\mu$m CMOS circuitry for electronic recording site positioning (via a 32:8 switch matrix), amplification of the neural signals (in-band ac gain of 300, 100 Hz-7 kHz; dc gain $<$1), reduction of channel output impedances ($<$500 $\Omega),$ and multiplexing of the recordings from selected sites onto a single output lead. A number of challenges in the application of these active devices have been overcome, including the development of an encapsulation scheme which protects the integrated circuits from the micromachining etchants and from the extracellular fluid. Stress-compensated, 0.75 $\mu$m-thick LPCVD dielectrics provide protection above and below polysilicon electrode interconnects, while 1 $\mu$m of planarizing low temperature (420$\sp\circ$C) oxide is used over aluminum interconnect in the circuit areas. Additional layers of PECVD nitride, metal and silicone can be added for enhanced chemical, electrical, and optical protection. The resulting process allows batch fabrication of the sensors with high yield and permits operation of both the transducers and circuitry in a challenging environment. A PC-based software interface for communication between the active probe and the user was developed which allowed the first successful neural recordings from any active electrode array. These recordings are comparable to recordings obtained via other methods in terms of signal-to-noise ratios (3:1 and better) and unit discriminability. This work also describes the investigation of a number of issues unique to active recording, such as optical noise sensitivity, input bias stability, baseline drift, referencing, and mixed analog-digital signal concerns associated with the sampling of low-level (100 $\mu$V) signals. Finally, the various circuit blocks required for active recording, their influence on recording ability, and design modifications for more successful active recording are addressed.