Area- and Energy- Efficient Modular Circuit Architecture for 1,024-Channel Parallel Neural Recording Microsystem.

Abstract

This research focuses to develop system architectures and associated electronic circuits for a next generation neuroscience research tool, a massive-parallel neural recording system capable of recording 1,024 channels simultaneously. Three interdependent prototypes have been developed to address major challenges in realization of the massive-parallel neural recording microsystems: minimization of energy and area consumption while preserving high quality in recordings. First, a modular 128-channel Δ-ΔΣ AFE using the spectrum shaping has been designed and fabricated to propose an area-and energy efficient solution for neural recording AFEs. The AFE achieved 4.84 fJ/C−s·mm2 figure of merit that is the smallest the area-energy product among the state-of-the-art multichannel neural recording systems. It also features power and area consumption of 3.05 µW and 0.05 mm2 per channel, respectively while exhibiting 63.3 dB signal-to-noise ratio with 3.02 µVrms input referred noise. Second, an on-chip mixed signal neural signal compressor was built to reduce the energy consumption in handling and transmission of the recorded data since this occupies a large portion of the total energy consumption as the number of parallel recording increases. The compressor reduces the data rates of two distinct groups of neural signals that are essential for neuroscience research: LFP and AP without loss of informative signals. As a result, the power consumptions for the data handling and transmissions of the LFP and AP were reduced to about 1/5.35 and 1/10.54 of the uncompressed cases, respectively. In the total data handling and transmission, the measured power consumption per channel is 11.98 µW that is about 1/9 of 107.5 µW without the compression. Third, a compact on-chip dc-to-dc converter with constant 1 MHz switching frequency has been developed to provide reliable power supplies and enhance energy delivery efficiency to the massive-parallel neural recording systems. The dc-to-dc converter has only predictable tones at the output and it exhibits > 80% power conversion efficiency at ultra-light loads, < 100 µW that is relevant power most of the multi-channel neural recording systems consume. The dc-to-dc converter occupies 0.375 mm2 of area which is less than 1/20 of the area the first prototype consumes (8.64 mm2).