Binaural hearing allows for accurate sound localization and provides important acoustic cues from the outside world. Mammals can process this large amount of parallel information using complex neurological wetware, namely a brain. Current CMOS technology does not permit fabrication of devices as complex as even the smallest mammalian brain, so the technology's current focus is to design circuits, both linear and non-linear, that emulate aspects of biological signal processing. Still, for the circuitry to be truly effective in low-power multiple sensor analog biomimetic systems, highly reproducible signal processing blocks are required on a single massively parallel integrated circuit.

I have developed reliable VLSI techniques, circuits, and software tools for use on a low-power acoustic processor, either as an analog system or as a mixed-mode System-on-a-Chip. These tools deal with disruptive non-idealities such as channel-to-channel variation, DC system drift, and DC offset problems that plague analog multichannel systems. First, I identify the sources of these challenges to our current-mode analog inner hair cell and auditory nerve (AIHCAN) integrated circuit. At the pre-fabrication level one can use my new layout tool, Cut-out my IC (COMIC), to rapidly create compensated transistors and place them. COMIC does all this while following established analog mismatch layout techniques. Still, this is not enough to achieve success with a non-linear system where steady-state information is critical, differential current tuning at various injection points is accomplished using the digital-to-analog converters calibrated output response (DACCOR), a micro-power mixed-mode active correction circuit. The DACCOR can consistently produce near-identical signal channels across each integrated circuit allowing a massively parallel system to be tuned for mismatch, post-fabrication.

Through an iterative approach of designing, building, and testing multiple integrated circuits, successful and consistent results have been shown with the AIHCAN integrated circuit, which was designed utilizing COMIC and DACCOR. Through AIHCAN I have demonstrated that each integrated circuit is capable of near-indistinguishable channel processing, thereby proving that low-power multiple sensor analog biomimetic systems can be fabricated with a very high yield.